Datasets:
s1mb-generalization-contextual-score
English evaluation cases authored using GPT-6-Astra. This review contains answers; do not send it as model input.
contextual-score-001
Instruction
"How strongly does this record support the claim under the specified evidence rubric?"
State
"Claim: the valve was closed at noon. The stipulated definitive inspection record shows it fully open at noon; this observation establishes the claim is false."
Criteria
0(value: 0.0): "The record establishes the claim is false"1(value: 0.25): "No relevant evidence either way"2(value: 0.5): "Relevant but inconclusive indication"3(value: 0.75): "One direct reliable observation supports the claim"4(value: 1.0): "Two independent direct reliable observations support the claim"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
The definitive observation refutes closure.
contextual-score-002
Instruction
"How suitable is this investment-like game for the player's stated preferences? This is a fictional game, not financial advice."
State
{
"game": {
"duration_minutes": 30,
"maximum_loss_tokens": 10,
"prize": "book"
},
"player": {
"duration": 60,
"max_loss": 20,
"prize": "book"
}
}
Criteria
0(value: 0.0): "Violates the maximum acceptable loss"1(value: 0.25): "Loss limit respected, but neither preferred feature met"2(value: 0.5): "Loss limit respected, only preferred duration met"3(value: 0.75): "Loss limit respected, only preferred prize met"4(value: 1.0): "Loss limit respected, preferred duration and prize both met"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Only the prize preference matches.
contextual-score-003
Instruction
"How well does the proposal fit this caucus's stated priorities? Use the stated priority hierarchy, not your own politics."
State
{
"non_negotiable": "No new city debt",
"primary": "Public ownership",
"proposal": "Build publicly owned homes using new city debt.",
"secondary": "More homes"
}
Criteria
0(value: 0.0): "Violates a non-negotiable commitment"1(value: 0.25): "Meets no priority but violates no non-negotiable"2(value: 0.5): "Meets only the secondary priority"3(value: 0.75): "Meets only the primary priority"4(value: 1.0): "Meets both priorities without a prohibited feature"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
New debt violates the non-negotiable despite other benefits.
contextual-score-004
Instruction
"Score this study's applicability to the specified review question."
State
"The study concerns adults receiving tutoring, measures reading, and uses a randomized design. The review requires exactly those four characteristics."
Criteria
0(value: 0.0): "Matches none of population, intervention, outcome or design"1(value: 0.25): "Matches exactly one of the four dimensions"2(value: 0.5): "Matches exactly two of the four dimensions"3(value: 0.75): "Matches exactly three of the four dimensions"4(value: 1.0): "Matches all four dimensions"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
All four dimensions match.
contextual-score-005
Instruction
"Score this study's applicability to the specified review question."
State
{
"order": [
"population",
"intervention",
"outcome",
"design"
],
"review": [
"adults",
"tutoring",
"reading",
"observational"
],
"study": [
"adults",
"tutoring",
"reading",
"randomized"
]
}
Criteria
0(value: 0.0): "Matches none of population, intervention, outcome or design"1(value: 0.25): "Matches exactly one of the four dimensions"2(value: 0.5): "Matches exactly two of the four dimensions"3(value: 0.75): "Matches exactly three of the four dimensions"4(value: 1.0): "Matches all four dimensions"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Only design differs.
contextual-score-006
Instruction
"Rate the proposed meeting time for this participant."
State
[
"Meeting: Tuesday 10:00, 30 minutes, remote; participant is available.",
"Preferences: morning, at least 60 minutes, in person."
]
Criteria
0(value: 0.0): "Outside a mandatory availability window"1(value: 0.25): "Available but violates all three preferences"2(value: 0.5): "Available and meets one preference"3(value: 0.75): "Available and meets two preferences"4(value: 1.0): "Available and meets all three preferences"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Only the morning preference matches.
contextual-score-007
Instruction
"Score this study's applicability to the specified review question."
State
{
"review": {
"design": "observational",
"intervention": "exercise",
"outcome": "balance",
"population": "children"
},
"study": {
"design": "randomized",
"intervention": "tutoring",
"outcome": "reading",
"population": "adults"
}
}
Criteria
0(value: 0.0): "Matches none of population, intervention, outcome or design"1(value: 0.25): "Matches exactly one of the four dimensions"2(value: 0.5): "Matches exactly two of the four dimensions"3(value: 0.75): "Matches exactly three of the four dimensions"4(value: 1.0): "Matches all four dimensions"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
No dimension matches.
contextual-score-008
Instruction
"How well does this menu fit the diner's own requirements? Apply prohibitions before preferences."
State
"Menu: spicy lentil stew, hot and low cost. Diner permits all ingredients but prefers mild food, cold dishes and a formal multi-course service. The stew is a single casual dish."
Criteria
0(value: 0.0): "Violates an explicit prohibition"1(value: 0.25): "Permitted; none of three preferences met"2(value: 0.5): "Permitted; one preference met"3(value: 0.75): "Permitted; two preferences met"4(value: 1.0): "Permitted; all three preferences met"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
None of the three preferences matches.
contextual-score-009
Instruction
"Score this site's restoration priority using the local charter's category, not a personal judgment of cultural worth."
State
"The charter requires owner consent and prioritizes public access, active community use and imminent damage. The owner consents; all three factors are documented at this site."
Criteria
0(value: 0.0): "Ineligible under the charter"1(value: 0.25): "Eligible, but none of three priority factors applies"2(value: 0.5): "Eligible with one priority factor"3(value: 0.75): "Eligible with two priority factors"4(value: 1.0): "Eligible with all three priority factors"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
Eligibility and all three factors are established.
contextual-score-010
Instruction
"Rate this analytics plan against the client's explicit privacy agreement."
State
[
"Plan: anonymous counts; report visits and purchases only. Permitted.",
"Client requests visits, dwell time and device types."
]
Criteria
0(value: 0.0): "Violates a mandatory restriction"1(value: 0.25): "Complies, but provides none of three requested analyses"2(value: 0.5): "Complies and provides one requested analysis"3(value: 0.75): "Complies and provides two requested analyses"4(value: 1.0): "Complies and provides all three requested analyses"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Only visits match the requested analyses.
contextual-score-011
Instruction
"How useful is the quoted answer for the user's actual information need?"
State
[
"Answer: The museum opens at 10:00.",
"User needs the museum's admission price; verified price is 8 credits."
]
Criteria
0(value: 0.0): "Contradicts the requested fact"1(value: 0.25): "Unrelated to the requested subject"2(value: 0.5): "Same subject but no requested fact"3(value: 0.75): "Answers part of a multi-part request correctly"4(value: 1.0): "Answers the complete request correctly"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
The subject matches but the price is absent.
contextual-score-012
Instruction
"How well does the proposal fit this caucus's stated priorities? Use the stated priority hierarchy, not your own politics."
State
[
"Proposal: build publicly owned homes using new city debt.",
"No non-negotiable is breached. Primary priority: public ownership. Secondary: reduce city debt."
]
Criteria
0(value: 0.0): "Violates a non-negotiable commitment"1(value: 0.25): "Meets no priority but violates no non-negotiable"2(value: 0.5): "Meets only the secondary priority"3(value: 0.75): "Meets only the primary priority"4(value: 1.0): "Meets both priorities without a prohibited feature"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Public ownership is met; debt reduction is not.
contextual-score-013
Instruction
"How useful is the quoted answer for the user's actual information need?"
State
"Answer: 'The museum opens at 10:00.' User asks opening time. The verified current opening time is 09:00, and 10:00 is explicitly incorrect."
Criteria
0(value: 0.0): "Contradicts the requested fact"1(value: 0.25): "Unrelated to the requested subject"2(value: 0.5): "Same subject but no requested fact"3(value: 0.75): "Answers part of a multi-part request correctly"4(value: 1.0): "Answers the complete request correctly"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
The answer contradicts the verified requested time.
contextual-score-014
Instruction
"Rate the proposed meeting time for this participant."
State
"Meeting: Tuesday 10:00, 30 minutes, remote. The participant is available and prefers morning, no more than 30 minutes and remote attendance."
Criteria
0(value: 0.0): "Outside a mandatory availability window"1(value: 0.25): "Available but violates all three preferences"2(value: 0.5): "Available and meets one preference"3(value: 0.75): "Available and meets two preferences"4(value: 1.0): "Available and meets all three preferences"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
All three preferences and availability match.
contextual-score-015
Instruction
"Rate the proposed meeting time for this participant."
State
{
"availability": "Wednesday only",
"meeting": "Tuesday 10:00, 30 minutes, remote.",
"preferences": [
"morning",
"at most 30 minutes",
"remote"
]
}
Criteria
0(value: 0.0): "Outside a mandatory availability window"1(value: 0.25): "Available but violates all three preferences"2(value: 0.5): "Available and meets one preference"3(value: 0.75): "Available and meets two preferences"4(value: 1.0): "Available and meets all three preferences"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
Tuesday lies outside mandatory availability.
contextual-score-016
Instruction
"How adequate is this message as an apology for the event described?"
State
{
"event": "The sender broke the recipient's mug.",
"message": "You made me do it; this is your fault."
}
Criteria
0(value: 0.0): "Denies or blames the recipient for established harm"1(value: 0.25): "No relevant acknowledgment of the harm"2(value: 0.5): "Acknowledges harm without accepting responsibility"3(value: 0.75): "Acknowledges harm and accepts responsibility, but no repair"4(value: 1.0): "Acknowledges harm, accepts responsibility and offers relevant repair"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
The sender blames the recipient for established harm.
contextual-score-017
Instruction
"How much of the four-part checklist is logically established by the record? Treat unestablished items as unmet, not false."
State
{
"checklist": [
"approved",
"signed",
"funded",
"scheduled"
],
"record": "Not rejected; signature status unknown; funds requested; date proposed.",
"rule": "None of these phrases entails the corresponding checklist item."
}
Criteria
0(value: 0.0): "Zero items established"1(value: 0.25): "One item established"2(value: 0.5): "Two items established"3(value: 0.75): "Three items established"4(value: 1.0): "Four items established"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
No positive checklist condition is established.
contextual-score-018
Instruction
"If the primary server failed, how much of the four-feature service would remain available under this model? Score the fraction of features remaining."
State
[
"Four features: search, upload, download, billing.",
"Independent backup serves search and download. Upload and billing require the primary."
]
Criteria
0(value: 0.0): "No features remain"1(value: 0.25): "One of four features remains"2(value: 0.5): "Two of four features remain"3(value: 0.75): "Three of four features remain"4(value: 1.0): "All four features remain"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Two independent backup features remain.
contextual-score-019
Instruction
"Rate the proposed meeting time for this participant."
State
"Meeting: Tuesday 10:00, 30 minutes, remote. Participant is available then but prefers afternoon, at least 60 minutes and in-person meetings."
Criteria
0(value: 0.0): "Outside a mandatory availability window"1(value: 0.25): "Available but violates all three preferences"2(value: 0.5): "Available and meets one preference"3(value: 0.75): "Available and meets two preferences"4(value: 1.0): "Available and meets all three preferences"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
Availability holds but none of the preferences matches.
contextual-score-020
Instruction
"How suitable is this investment-like game for the player's stated preferences? This is a fictional game, not financial advice."
State
"Game: maximum loss 10 tokens, lasts 30 minutes, prize is a book. Player: accepts loss up to 10, prefers 60 minutes and a toy."
Criteria
0(value: 0.0): "Violates the maximum acceptable loss"1(value: 0.25): "Loss limit respected, but neither preferred feature met"2(value: 0.5): "Loss limit respected, only preferred duration met"3(value: 0.75): "Loss limit respected, only preferred prize met"4(value: 1.0): "Loss limit respected, preferred duration and prize both met"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
Loss is acceptable but neither preference matches.
contextual-score-021
Instruction
"Rate the translation's fidelity to the source as disambiguated by the context."
State
{
"source_meaning": "Do not send the file to Lee today.",
"translation": "Send the file to Lee today."
}
Criteria
0(value: 0.0): "Reverses the central instruction"1(value: 0.25): "Changes the central action to a different action"2(value: 0.5): "Preserves action but gets a required time or recipient wrong"3(value: 0.75): "Preserves all required meaning but adds an unsupported detail"4(value: 1.0): "Preserves all and only the required meaning"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
The prohibition is reversed into an instruction.
contextual-score-022
Instruction
"Score this exhibition plan against the community committee's stated conditions."
State
[
"Permitted plan: replica, community-written labels, no audio, paid entry.",
"Requests: community-written labels, audio testimony, free entry."
]
Criteria
0(value: 0.0): "Violates a non-negotiable condition"1(value: 0.25): "Permitted but meets none of three requested features"2(value: 0.5): "Permitted and meets one requested feature"3(value: 0.75): "Permitted and meets two requested features"4(value: 1.0): "Permitted and meets all three requested features"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Only community-written labels are present.
contextual-score-023
Instruction
"How well does the proposal fit this caucus's stated priorities? Use the stated priority hierarchy, not your own politics."
State
"Proposal: build publicly owned homes using new city debt. The caucus has no debt prohibition. Its primary priority is public ownership and its secondary priority is more homes."
Criteria
0(value: 0.0): "Violates a non-negotiable commitment"1(value: 0.25): "Meets no priority but violates no non-negotiable"2(value: 0.5): "Meets only the secondary priority"3(value: 0.75): "Meets only the primary priority"4(value: 1.0): "Meets both priorities without a prohibited feature"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
Both stated priorities are met without a prohibition.
contextual-score-024
Instruction
"How much of the four-part checklist is logically established by the record? Treat unestablished items as unmet, not false."
State
[
"Checklist: approved, signed, funded, scheduled.",
"Record: approved and signed; funding and scheduling unknown."
]
Criteria
0(value: 0.0): "Zero items established"1(value: 0.25): "One item established"2(value: 0.5): "Two items established"3(value: 0.75): "Three items established"4(value: 1.0): "Four items established"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Approval and signature are established.
contextual-score-025
Instruction
"Score the proposed assignment's fit to this worker's declared constraints."
State
[
"Assignment: four hours of audio editing using a licensed workstation.",
"Worker: five-hour cap, trained editor, no licensed workstation."
]
Criteria
0(value: 0.0): "Breaks a hard hours cap"1(value: 0.25): "Within hours cap but lacks both required skill and equipment"2(value: 0.5): "Within hours cap and has skill but lacks equipment"3(value: 0.75): "Within hours cap and has equipment but needs skill training"4(value: 1.0): "Within hours cap with both skill and equipment"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
The worker has skill but lacks equipment.
contextual-score-026
Instruction
"How much of the four-part checklist is logically established by the record? Treat unestablished items as unmet, not false."
State
"Checklist: approved, signed, funded, scheduled. The definitive record explicitly confirms approval, signature, full funding and an agreed scheduled date."
Criteria
0(value: 0.0): "Zero items established"1(value: 0.25): "One item established"2(value: 0.5): "Two items established"3(value: 0.75): "Three items established"4(value: 1.0): "Four items established"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
All four are explicitly established.
contextual-score-027
Instruction
"How urgent is the response under this fictional support contract?"
State
{
"contract": "Duplicate closed requests need no response.",
"ticket": "Exact duplicate of an already resolved and closed request."
}
Criteria
0(value: 0.0): "No response required"1(value: 0.25): "Respond within seven days"2(value: 0.5): "Respond within one day"3(value: 0.75): "Respond within one hour"4(value: 1.0): "Respond immediately"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
The duplicate rule sets no response requirement.
contextual-score-028
Instruction
"Rate this analytics plan against the client's explicit privacy agreement."
State
"Plan: store anonymous counts; report visits and purchases only. Agreement permits anonymous counts. Client requests dwell time, device types and referral sources, none provided."
Criteria
0(value: 0.0): "Violates a mandatory restriction"1(value: 0.25): "Complies, but provides none of three requested analyses"2(value: 0.5): "Complies and provides one requested analysis"3(value: 0.75): "Complies and provides two requested analyses"4(value: 1.0): "Complies and provides all three requested analyses"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
The plan complies but supplies none of the requested analyses.
contextual-score-029
Instruction
"Rate how suitable this ceremony is for this participant's explicitly stated practice."
State
"Ceremony: silent evening gathering outdoors, without invocations. It is permitted. The participant wishes for music, a morning time and an indoor venue."
Criteria
0(value: 0.0): "Conflicts with an absolute personal prohibition"1(value: 0.25): "Permitted, but meets none of the participant's three wishes"2(value: 0.5): "Permitted and meets one wish"3(value: 0.75): "Permitted and meets two wishes"4(value: 1.0): "Permitted and meets all three wishes"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
It meets none of the three wishes.
contextual-score-030
Instruction
"Rate the translation's fidelity to the source as disambiguated by the context."
State
[
"Translation: Send the file to Lee today.",
"Required source meaning: send the file to Lee tomorrow."
]
Criteria
0(value: 0.0): "Reverses the central instruction"1(value: 0.25): "Changes the central action to a different action"2(value: 0.5): "Preserves action but gets a required time or recipient wrong"3(value: 0.75): "Preserves all required meaning but adds an unsupported detail"4(value: 1.0): "Preserves all and only the required meaning"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
The action and recipient match but the required time does not.
contextual-score-031
Instruction
"Score this purchase against the buyer's budget and needs."
State
{
"buyer": {
"desired_warranty_years": 2,
"hard_budget": 60,
"required": [
"print",
"scan"
]
},
"item": {
"functions": [
"print",
"scan"
],
"price": 80,
"warranty_years": 2
}
}
Criteria
0(value: 0.0): "Exceeds the hard budget"1(value: 0.25): "Within budget but meets neither required function"2(value: 0.5): "Within budget and meets exactly one required function"3(value: 0.75): "Within budget and meets both functions but not the desired warranty"4(value: 1.0): "Within budget and meets both functions and desired warranty"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
The price exceeds the hard limit.
contextual-score-032
Instruction
"Score the proposed assignment's fit to this worker's declared constraints."
State
"Assignment: four hours of audio editing on a licensed workstation. Worker can spend five hours but has neither audio-editing skills nor the workstation."
Criteria
0(value: 0.0): "Breaks a hard hours cap"1(value: 0.25): "Within hours cap but lacks both required skill and equipment"2(value: 0.5): "Within hours cap and has skill but lacks equipment"3(value: 0.75): "Within hours cap and has equipment but needs skill training"4(value: 1.0): "Within hours cap with both skill and equipment"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
Neither skill nor equipment is available.
contextual-score-033
Instruction
"How adequate is this message as an apology for the event described?"
State
"Event: sender broke the mug. Message: 'I broke your mug and am sorry. I will pay for the equivalent replacement you choose.'"
Criteria
0(value: 0.0): "Denies or blames the recipient for established harm"1(value: 0.25): "No relevant acknowledgment of the harm"2(value: 0.5): "Acknowledges harm without accepting responsibility"3(value: 0.75): "Acknowledges harm and accepts responsibility, but no repair"4(value: 1.0): "Acknowledges harm, accepts responsibility and offers relevant repair"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
The message includes relevant restitution.
contextual-score-034
Instruction
"How useful is the quoted answer for the user's actual information need?"
State
{
"answer": "The museum opens at 10:00.",
"user_need": "What time does the museum open?",
"verified_facts": "Museum opens at 10:00."
}
Criteria
0(value: 0.0): "Contradicts the requested fact"1(value: 0.25): "Unrelated to the requested subject"2(value: 0.5): "Same subject but no requested fact"3(value: 0.75): "Answers part of a multi-part request correctly"4(value: 1.0): "Answers the complete request correctly"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
The answer supplies the complete requested fact.
contextual-score-035
Instruction
"Rate the translation's fidelity to the source as disambiguated by the context."
State
{
"source_meaning": "Send the file to Lee; no deadline is stated or implied.",
"translation": "Send the file to Lee today."
}
Criteria
0(value: 0.0): "Reverses the central instruction"1(value: 0.25): "Changes the central action to a different action"2(value: 0.5): "Preserves action but gets a required time or recipient wrong"3(value: 0.75): "Preserves all required meaning but adds an unsupported detail"4(value: 1.0): "Preserves all and only the required meaning"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Today is an unsupported added deadline.
contextual-score-036
Instruction
"How useful is the quoted answer for the user's actual information need?"
State
"Answer: 'The museum opens at 10:00.' User asks when the museum opens and closes. Verified hours are 10:00 to 18:00."
Criteria
0(value: 0.0): "Contradicts the requested fact"1(value: 0.25): "Unrelated to the requested subject"2(value: 0.5): "Same subject but no requested fact"3(value: 0.75): "Answers part of a multi-part request correctly"4(value: 1.0): "Answers the complete request correctly"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Opening is answered correctly but closing is omitted.
contextual-score-037
Instruction
"How well does this menu fit the diner's own requirements? Apply prohibitions before preferences."
State
[
"Menu: spicy lentil stew, hot, low cost, single casual dish; permitted.",
"Preferences: spicy food, cold dishes, formal multi-course service."
]
Criteria
0(value: 0.0): "Violates an explicit prohibition"1(value: 0.25): "Permitted; none of three preferences met"2(value: 0.5): "Permitted; one preference met"3(value: 0.75): "Permitted; two preferences met"4(value: 1.0): "Permitted; all three preferences met"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Only spiciness matches.
contextual-score-038
Instruction
"Rate how well this reply follows the recipient's communication brief."
State
"Reply: 'Hi Sam, Project Cedar is approved. Please reply today.' Naming is permitted. Brief requires a greeting, the project name and a request to reply today."
Criteria
0(value: 0.0): "Violates the brief's mandatory confidentiality rule"1(value: 0.25): "Confidentiality respected; none of three style requirements met"2(value: 0.5): "Confidentiality respected; one style requirement met"3(value: 0.75): "Confidentiality respected; two style requirements met"4(value: 1.0): "Confidentiality respected; all three style requirements met"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
All three style requirements are satisfied.
contextual-score-039
Instruction
"Rate this analytics plan against the client's explicit privacy agreement."
State
{
"plan": "Permitted anonymous counts reporting visits and purchases only.",
"requested": [
"visits",
"purchases",
"referrals"
]
}
Criteria
0(value: 0.0): "Violates a mandatory restriction"1(value: 0.25): "Complies, but provides none of three requested analyses"2(value: 0.5): "Complies and provides one requested analysis"3(value: 0.75): "Complies and provides two requested analyses"4(value: 1.0): "Complies and provides all three requested analyses"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Two of the three requested analyses are provided.
contextual-score-040
Instruction
"Score the proposed assignment's fit to this worker's declared constraints."
State
{
"assignment": "Four hours of audio editing using a licensed workstation.",
"worker": {
"audio_skill": false,
"hours_cap": 4,
"licensed_workstation": true
}
}
Criteria
0(value: 0.0): "Breaks a hard hours cap"1(value: 0.25): "Within hours cap but lacks both required skill and equipment"2(value: 0.5): "Within hours cap and has skill but lacks equipment"3(value: 0.75): "Within hours cap and has equipment but needs skill training"4(value: 1.0): "Within hours cap with both skill and equipment"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Equipment is available but skill training is needed.
contextual-score-041
Instruction
"How well does the proposal fit this caucus's stated priorities? Use the stated priority hierarchy, not your own politics."
State
"Proposal: build publicly owned homes using new city debt. The caucus's priorities are lower debt and private ownership, but neither is a non-negotiable. The proposal meets neither priority."
Criteria
0(value: 0.0): "Violates a non-negotiable commitment"1(value: 0.25): "Meets no priority but violates no non-negotiable"2(value: 0.5): "Meets only the secondary priority"3(value: 0.75): "Meets only the primary priority"4(value: 1.0): "Meets both priorities without a prohibited feature"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
Neither priority is met, without an absolute prohibition.
contextual-score-042
Instruction
"How urgent is the response under this fictional support contract?"
State
"Contract: confirmed ongoing data loss requires immediate response. Ticket: logs confirm data is currently being lost."
Criteria
0(value: 0.0): "No response required"1(value: 0.25): "Respond within seven days"2(value: 0.5): "Respond within one day"3(value: 0.75): "Respond within one hour"4(value: 1.0): "Respond immediately"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
The immediate-response trigger is confirmed.
contextual-score-043
Instruction
"How useful is the quoted answer for the user's actual information need?"
State
{
"answer": "The museum opens at 10:00.",
"user_need": "How do I reset my router?",
"verified_facts": "Router reset requires its reset button."
}
Criteria
0(value: 0.0): "Contradicts the requested fact"1(value: 0.25): "Unrelated to the requested subject"2(value: 0.5): "Same subject but no requested fact"3(value: 0.75): "Answers part of a multi-part request correctly"4(value: 1.0): "Answers the complete request correctly"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
Museum hours are unrelated to router reset.
contextual-score-044
Instruction
"How urgent is the response under this fictional support contract?"
State
{
"contract": "A team-wide work blockage without data loss requires response within one hour.",
"ticket": "The whole team is blocked; no data is lost."
}
Criteria
0(value: 0.0): "No response required"1(value: 0.25): "Respond within seven days"2(value: 0.5): "Respond within one day"3(value: 0.75): "Respond within one hour"4(value: 1.0): "Respond immediately"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
The one-hour blockage category applies.
contextual-score-045
Instruction
"Score this site's restoration priority using the local charter's category, not a personal judgment of cultural worth."
State
{
"charter": "Consent required. Factors: public access, active community use, imminent damage.",
"site": "Consenting owner; public access and active community use; no imminent damage."
}
Criteria
0(value: 0.0): "Ineligible under the charter"1(value: 0.25): "Eligible, but none of three priority factors applies"2(value: 0.5): "Eligible with one priority factor"3(value: 0.75): "Eligible with two priority factors"4(value: 1.0): "Eligible with all three priority factors"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Public access and active use supply two factors.
contextual-score-046
Instruction
"Rate this analytics plan against the client's explicit privacy agreement."
State
{
"agreement": "No storage of any kind",
"plan": "Store anonymous counts; report visits and purchases, not referrals.",
"requested": [
"visits",
"purchases",
"referrals"
]
}
Criteria
0(value: 0.0): "Violates a mandatory restriction"1(value: 0.25): "Complies, but provides none of three requested analyses"2(value: 0.5): "Complies and provides one requested analysis"3(value: 0.75): "Complies and provides two requested analyses"4(value: 1.0): "Complies and provides all three requested analyses"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
Even anonymous storage violates the mandatory restriction.
contextual-score-047
Instruction
"How suitable is this investment-like game for the player's stated preferences? This is a fictional game, not financial advice."
State
"Game: at most 10 tokens lost, 30 minutes, book prize. Player accepts up to 10 tokens of loss and prefers a 30-minute game with a book prize."
Criteria
0(value: 0.0): "Violates the maximum acceptable loss"1(value: 0.25): "Loss limit respected, but neither preferred feature met"2(value: 0.5): "Loss limit respected, only preferred duration met"3(value: 0.75): "Loss limit respected, only preferred prize met"4(value: 1.0): "Loss limit respected, preferred duration and prize both met"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
The loss boundary and both preferences are satisfied.
contextual-score-048
Instruction
"Score this site's restoration priority using the local charter's category, not a personal judgment of cultural worth."
State
"Charter: owner consent required; priority factors are public access, active community use and imminent damage. Owner consents, but the private unused site is stable and closed to the public."
Criteria
0(value: 0.0): "Ineligible under the charter"1(value: 0.25): "Eligible, but none of three priority factors applies"2(value: 0.5): "Eligible with one priority factor"3(value: 0.75): "Eligible with two priority factors"4(value: 1.0): "Eligible with all three priority factors"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
The eligible site has none of the factors.
contextual-score-049
Instruction
"Score this exhibition plan against the community committee's stated conditions."
State
{
"plan": "Permitted replica display with community-written labels, audio testimony, paid entry.",
"requested": [
"community-written labels",
"audio testimony",
"free entry"
]
}
Criteria
0(value: 0.0): "Violates a non-negotiable condition"1(value: 0.25): "Permitted but meets none of three requested features"2(value: 0.5): "Permitted and meets one requested feature"3(value: 0.75): "Permitted and meets two requested features"4(value: 1.0): "Permitted and meets all three requested features"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Labels and audio are supplied, but entry is not free.
contextual-score-050
Instruction
"Rate how suitable this ceremony is for this participant's explicitly stated practice."
State
{
"ceremony": "Silent evening gathering outdoors, with no invocations.",
"prohibition": "No outdoor gatherings",
"wishes": [
"silence",
"evening",
"no invocations"
]
}
Criteria
0(value: 0.0): "Conflicts with an absolute personal prohibition"1(value: 0.25): "Permitted, but meets none of the participant's three wishes"2(value: 0.5): "Permitted and meets one wish"3(value: 0.75): "Permitted and meets two wishes"4(value: 1.0): "Permitted and meets all three wishes"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
The outdoor setting violates an absolute prohibition.
contextual-score-051
Instruction
"Rate this venue for the attendee using the rubric's priority order."
State
{
"preferences": [
"quiet room",
"near transit",
"free entry"
],
"required": "hearing loop",
"venue": "Step-free entry, no hearing loop, quiet room, near transit, free entry."
}
Criteria
0(value: 0.0): "A required access feature is absent"1(value: 0.25): "Required access is present; none of three preferences met"2(value: 0.5): "Required access is present; one preference met"3(value: 0.75): "Required access is present; two preferences met"4(value: 1.0): "Required access is present; all three preferences met"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
The required hearing loop is absent.
contextual-score-052
Instruction
"How adequate is this message as an apology for the event described?"
State
{
"event": "Sender broke the mug.",
"message": "I broke your mug. I am sorry for the damage I caused."
}
Criteria
0(value: 0.0): "Denies or blames the recipient for established harm"1(value: 0.25): "No relevant acknowledgment of the harm"2(value: 0.5): "Acknowledges harm without accepting responsibility"3(value: 0.75): "Acknowledges harm and accepts responsibility, but no repair"4(value: 1.0): "Acknowledges harm, accepts responsibility and offers relevant repair"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Responsibility and harm are acknowledged but no repair is offered.
contextual-score-053
Instruction
"Score this exhibition plan against the community committee's stated conditions."
State
"The committee permits this replica display and requests community-written labels, audio testimony and free entry. The plan provides all three."
Criteria
0(value: 0.0): "Violates a non-negotiable condition"1(value: 0.25): "Permitted but meets none of three requested features"2(value: 0.5): "Permitted and meets one requested feature"3(value: 0.75): "Permitted and meets two requested features"4(value: 1.0): "Permitted and meets all three requested features"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
The plan fulfills every condition and request.
contextual-score-054
Instruction
"How adequate is this message as an apology for the event described?"
State
"Event: sender broke the recipient's mug. Message: 'Nice weather today.'"
Criteria
0(value: 0.0): "Denies or blames the recipient for established harm"1(value: 0.25): "No relevant acknowledgment of the harm"2(value: 0.5): "Acknowledges harm without accepting responsibility"3(value: 0.75): "Acknowledges harm and accepts responsibility, but no repair"4(value: 1.0): "Acknowledges harm, accepts responsibility and offers relevant repair"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
The message never acknowledges the incident.
contextual-score-055
Instruction
"How strongly does this record support the claim under the specified evidence rubric?"
State
{
"claim": "The valve was closed at noon.",
"record": "Two independent calibrated position logs both directly show closed at noon."
}
Criteria
0(value: 0.0): "The record establishes the claim is false"1(value: 0.25): "No relevant evidence either way"2(value: 0.5): "Relevant but inconclusive indication"3(value: 0.75): "One direct reliable observation supports the claim"4(value: 1.0): "Two independent direct reliable observations support the claim"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
Two independent direct observations support closure.
contextual-score-056
Instruction
"How well does this menu fit the diner's own requirements? Apply prohibitions before preferences."
State
"The diner permits the spicy lentil stew and wants spicy food, a hot dish and low cost. The menu meets exactly those three preferences."
Criteria
0(value: 0.0): "Violates an explicit prohibition"1(value: 0.25): "Permitted; none of three preferences met"2(value: 0.5): "Permitted; one preference met"3(value: 0.75): "Permitted; two preferences met"4(value: 1.0): "Permitted; all three preferences met"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
Every stated preference is met.
contextual-score-057
Instruction
"Score the proposed assignment's fit to this worker's declared constraints."
State
"Assignment: four hours of audio editing using a licensed workstation. Worker is a trained editor with that workstation and exactly four hours available."
Criteria
0(value: 0.0): "Breaks a hard hours cap"1(value: 0.25): "Within hours cap but lacks both required skill and equipment"2(value: 0.5): "Within hours cap and has skill but lacks equipment"3(value: 0.75): "Within hours cap and has equipment but needs skill training"4(value: 1.0): "Within hours cap with both skill and equipment"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
All required resources fit, including the inclusive cap.
contextual-score-058
Instruction
"Score this study's applicability to the specified review question."
State
"Study: adults, tutoring, reading outcome, randomized design. Review requires adults, exercise, balance outcome and observational design. Treat these categories as exact and disjoint."
Criteria
0(value: 0.0): "Matches none of population, intervention, outcome or design"1(value: 0.25): "Matches exactly one of the four dimensions"2(value: 0.5): "Matches exactly two of the four dimensions"3(value: 0.75): "Matches exactly three of the four dimensions"4(value: 1.0): "Matches all four dimensions"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
Only population matches.
contextual-score-059
Instruction
"Rate how suitable this ceremony is for this participant's explicitly stated practice."
State
{
"ceremony": "Silent evening gathering outdoors, no invocations.",
"permitted": true,
"wishes": [
"silence",
"evening",
"indoor venue"
]
}
Criteria
0(value: 0.0): "Conflicts with an absolute personal prohibition"1(value: 0.25): "Permitted, but meets none of the participant's three wishes"2(value: 0.5): "Permitted and meets one wish"3(value: 0.75): "Permitted and meets two wishes"4(value: 1.0): "Permitted and meets all three wishes"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Silence and evening timing meet two wishes.
contextual-score-060
Instruction
"Rate how suitable this ceremony is for this participant's explicitly stated practice."
State
"The participant permits the silent outdoor evening gathering and specifically wishes for silence, an evening time and no invocations."
Criteria
0(value: 0.0): "Conflicts with an absolute personal prohibition"1(value: 0.25): "Permitted, but meets none of the participant's three wishes"2(value: 0.5): "Permitted and meets one wish"3(value: 0.75): "Permitted and meets two wishes"4(value: 1.0): "Permitted and meets all three wishes"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
All three wishes are met.
contextual-score-061
Instruction
"Score this purchase against the buyer's budget and needs."
State
{
"buyer": {
"desired_warranty_years": 2,
"hard_budget": 100,
"required": [
"print",
"fax"
]
},
"item": {
"functions": [
"print",
"scan"
],
"price": 80,
"warranty_years": 2
}
}
Criteria
0(value: 0.0): "Exceeds the hard budget"1(value: 0.25): "Within budget but meets neither required function"2(value: 0.5): "Within budget and meets exactly one required function"3(value: 0.75): "Within budget and meets both functions but not the desired warranty"4(value: 1.0): "Within budget and meets both functions and desired warranty"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Only printing matches a required function.
contextual-score-062
Instruction
"How adequate is this message as an apology for the event described?"
State
[
"Event: sender broke the mug.",
"Message: I see that your mug is broken and you are upset, but I am not saying who caused it."
]
Criteria
0(value: 0.0): "Denies or blames the recipient for established harm"1(value: 0.25): "No relevant acknowledgment of the harm"2(value: 0.5): "Acknowledges harm without accepting responsibility"3(value: 0.75): "Acknowledges harm and accepts responsibility, but no repair"4(value: 1.0): "Acknowledges harm, accepts responsibility and offers relevant repair"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Harm is acknowledged without responsibility.
contextual-score-063
Instruction
"How urgent is the response under this fictional support contract?"
State
"Contract: cosmetic issues require response within seven days. Ticket: a cosmetic spacing issue with no loss of functionality."
Criteria
0(value: 0.0): "No response required"1(value: 0.25): "Respond within seven days"2(value: 0.5): "Respond within one day"3(value: 0.75): "Respond within one hour"4(value: 1.0): "Respond immediately"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
The cosmetic category has the seven-day requirement.
contextual-score-064
Instruction
"Score this exhibition plan against the community committee's stated conditions."
State
{
"non_negotiable": "Do not display the original object",
"plan": "Display the original object, with community-authored labels, audio testimony and free entry.",
"requested": [
"community labels",
"audio testimony",
"free entry"
]
}
Criteria
0(value: 0.0): "Violates a non-negotiable condition"1(value: 0.25): "Permitted but meets none of three requested features"2(value: 0.5): "Permitted and meets one requested feature"3(value: 0.75): "Permitted and meets two requested features"4(value: 1.0): "Permitted and meets all three requested features"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
Displaying the original violates the absolute condition.
contextual-score-065
Instruction
"Rate how suitable this ceremony is for this participant's explicitly stated practice."
State
[
"Ceremony: silent evening gathering outdoors, no invocations; permitted.",
"Wishes: silence, morning time, indoor venue."
]
Criteria
0(value: 0.0): "Conflicts with an absolute personal prohibition"1(value: 0.25): "Permitted, but meets none of the participant's three wishes"2(value: 0.5): "Permitted and meets one wish"3(value: 0.75): "Permitted and meets two wishes"4(value: 1.0): "Permitted and meets all three wishes"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Only the wish for silence is met.
contextual-score-066
Instruction
"If the primary server failed, how much of the four-feature service would remain available under this model? Score the fraction of features remaining."
State
"The service has search, upload, download and billing. A fully independent hot replica immediately serves all four if the primary fails, with no shared dependencies."
Criteria
0(value: 0.0): "No features remain"1(value: 0.25): "One of four features remains"2(value: 0.5): "Two of four features remain"3(value: 0.75): "Three of four features remain"4(value: 1.0): "All four features remain"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
The intervention leaves all four features available.
contextual-score-067
Instruction
"Rate how well this reply follows the recipient's communication brief."
State
{
"brief": "Never name Project Cedar; use a greeting, name the project, and request a reply today. Confidentiality overrides conflicting style requests.",
"reply": "Hi Sam, Project Cedar is approved. Please reply today."
}
Criteria
0(value: 0.0): "Violates the brief's mandatory confidentiality rule"1(value: 0.25): "Confidentiality respected; none of three style requirements met"2(value: 0.5): "Confidentiality respected; one style requirement met"3(value: 0.75): "Confidentiality respected; two style requirements met"4(value: 1.0): "Confidentiality respected; all three style requirements met"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
The explicit confidentiality priority overrides the conflicting style request.
contextual-score-068
Instruction
"Score this site's restoration priority using the local charter's category, not a personal judgment of cultural worth."
State
{
"charter": "Only owner-consented sites are eligible; priority factors are public access, active community use, imminent damage.",
"site": {
"active_use": true,
"imminent_damage": true,
"owner_consented": false,
"public_access": true
}
}
Criteria
0(value: 0.0): "Ineligible under the charter"1(value: 0.25): "Eligible, but none of three priority factors applies"2(value: 0.5): "Eligible with one priority factor"3(value: 0.75): "Eligible with two priority factors"4(value: 1.0): "Eligible with all three priority factors"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
Lack of owner consent makes the site ineligible.
contextual-score-069
Instruction
"Score the proposed assignment's fit to this worker's declared constraints."
State
{
"assignment": "Four hours of audio editing using a licensed workstation.",
"worker": {
"audio_skill": true,
"hours_cap": 3,
"licensed_workstation": true
}
}
Criteria
0(value: 0.0): "Breaks a hard hours cap"1(value: 0.25): "Within hours cap but lacks both required skill and equipment"2(value: 0.5): "Within hours cap and has skill but lacks equipment"3(value: 0.75): "Within hours cap and has equipment but needs skill training"4(value: 1.0): "Within hours cap with both skill and equipment"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
Four hours exceeds the cap of three.
contextual-score-070
Instruction
"Score this exhibition plan against the community committee's stated conditions."
State
"Plan: display an approved replica with curator-written labels, no audio, paid entry. Replica display is permitted. The committee requests community-written labels, audio testimony and free entry."
Criteria
0(value: 0.0): "Violates a non-negotiable condition"1(value: 0.25): "Permitted but meets none of three requested features"2(value: 0.5): "Permitted and meets one requested feature"3(value: 0.75): "Permitted and meets two requested features"4(value: 1.0): "Permitted and meets all three requested features"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
None of the three requested features is supplied.
contextual-score-071
Instruction
"Score this site's restoration priority using the local charter's category, not a personal judgment of cultural worth."
State
[
"Charter: consent required; factors are public access, active community use, imminent damage.",
"Owner consents; site is publicly accessible, unused and stable."
]
Criteria
0(value: 0.0): "Ineligible under the charter"1(value: 0.25): "Eligible, but none of three priority factors applies"2(value: 0.5): "Eligible with one priority factor"3(value: 0.75): "Eligible with two priority factors"4(value: 1.0): "Eligible with all three priority factors"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Only public access applies.
contextual-score-072
Instruction
"If the primary server failed, how much of the four-feature service would remain available under this model? Score the fraction of features remaining."
State
{
"features": [
"search",
"upload",
"download",
"billing"
],
"model": "All features require the primary; no backup."
}
Criteria
0(value: 0.0): "No features remain"1(value: 0.25): "One of four features remains"2(value: 0.5): "Two of four features remain"3(value: 0.75): "Three of four features remain"4(value: 1.0): "All four features remain"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
Failure removes every feature.
contextual-score-073
Instruction
"If the primary server failed, how much of the four-feature service would remain available under this model? Score the fraction of features remaining."
State
"Four features: search, upload, download, billing. After primary failure, an independent read-only cache serves search only; all other features stop."
Criteria
0(value: 0.0): "No features remain"1(value: 0.25): "One of four features remains"2(value: 0.5): "Two of four features remain"3(value: 0.75): "Three of four features remain"4(value: 1.0): "All four features remain"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
Only one of four survives.
contextual-score-074
Instruction
"Rate how well this reply follows the recipient's communication brief."
State
[
"Reply: Hi Sam, Project Cedar is approved. Please reply today. Project naming is permitted.",
"Brief: include a greeting, omit project names, avoid requests for action."
]
Criteria
0(value: 0.0): "Violates the brief's mandatory confidentiality rule"1(value: 0.25): "Confidentiality respected; none of three style requirements met"2(value: 0.5): "Confidentiality respected; one style requirement met"3(value: 0.75): "Confidentiality respected; two style requirements met"4(value: 1.0): "Confidentiality respected; all three style requirements met"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Only the greeting requirement is satisfied.
contextual-score-075
Instruction
"If the primary server failed, how much of the four-feature service would remain available under this model? Score the fraction of features remaining."
State
{
"assumption": "Backup remains operational after primary failure.",
"features": {
"billing": "primary only",
"download": "independent backup",
"search": "independent backup",
"upload": "independent backup"
}
}
Criteria
0(value: 0.0): "No features remain"1(value: 0.25): "One of four features remains"2(value: 0.5): "Two of four features remain"3(value: 0.75): "Three of four features remain"4(value: 1.0): "All four features remain"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Only billing is lost.
contextual-score-076
Instruction
"How strongly does this record support the claim under the specified evidence rubric?"
State
[
"Claim: the valve was closed at noon.",
"Record: flow was low, which may result from closure or from an empty tank; no direct observation."
]
Criteria
0(value: 0.0): "The record establishes the claim is false"1(value: 0.25): "No relevant evidence either way"2(value: 0.5): "Relevant but inconclusive indication"3(value: 0.75): "One direct reliable observation supports the claim"4(value: 1.0): "Two independent direct reliable observations support the claim"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Low flow is relevant but has an alternative explanation.
contextual-score-077
Instruction
"Rate how well this reply follows the recipient's communication brief."
State
"Reply: 'Hi Sam, Project Cedar is approved. Please reply today.' Naming the project is permitted. Brief: no greeting, no project name, no requested action."
Criteria
0(value: 0.0): "Violates the brief's mandatory confidentiality rule"1(value: 0.25): "Confidentiality respected; none of three style requirements met"2(value: 0.5): "Confidentiality respected; one style requirement met"3(value: 0.75): "Confidentiality respected; two style requirements met"4(value: 1.0): "Confidentiality respected; all three style requirements met"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
The reply violates all three style requirements, but no confidentiality rule.
contextual-score-078
Instruction
"How strongly does this record support the claim under the specified evidence rubric?"
State
{
"claim": "The valve was closed at noon.",
"record": "Only a photograph of an unrelated office is available."
}
Criteria
0(value: 0.0): "The record establishes the claim is false"1(value: 0.25): "No relevant evidence either way"2(value: 0.5): "Relevant but inconclusive indication"3(value: 0.75): "One direct reliable observation supports the claim"4(value: 1.0): "Two independent direct reliable observations support the claim"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
The record contains no relevant evidence.
contextual-score-079
Instruction
"Score this purchase against the buyer's budget and needs."
State
"Item: 80 credits, prints and scans, two-year warranty. Buyer: budget 100, requires print and scan, wants a five-year warranty."
Criteria
0(value: 0.0): "Exceeds the hard budget"1(value: 0.25): "Within budget but meets neither required function"2(value: 0.5): "Within budget and meets exactly one required function"3(value: 0.75): "Within budget and meets both functions but not the desired warranty"4(value: 1.0): "Within budget and meets both functions and desired warranty"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Both functions fit but the warranty is too short.
contextual-score-080
Instruction
"Rate this venue for the attendee using the rubric's priority order."
State
"The venue has step-free entry, a quiet room, nearby transit and free entry. The attendee requires step-free entry and prefers exactly the other three features."
Criteria
0(value: 0.0): "A required access feature is absent"1(value: 0.25): "Required access is present; none of three preferences met"2(value: 0.5): "Required access is present; one preference met"3(value: 0.75): "Required access is present; two preferences met"4(value: 1.0): "Required access is present; all three preferences met"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
Required access and all preferences match.
contextual-score-081
Instruction
"How suitable is this investment-like game for the player's stated preferences? This is a fictional game, not financial advice."
State
{
"game": {
"duration_minutes": 30,
"maximum_loss_tokens": 10,
"prize": "book"
},
"player": {
"duration": 30,
"max_loss": 5,
"prize": "book"
}
}
Criteria
0(value: 0.0): "Violates the maximum acceptable loss"1(value: 0.25): "Loss limit respected, but neither preferred feature met"2(value: 0.5): "Loss limit respected, only preferred duration met"3(value: 0.75): "Loss limit respected, only preferred prize met"4(value: 1.0): "Loss limit respected, preferred duration and prize both met"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
The possible loss exceeds the hard tolerance.
contextual-score-082
Instruction
"How well does this menu fit the diner's own requirements? Apply prohibitions before preferences."
State
{
"menu": "Permitted spicy hot lentil stew, low cost, casual single dish.",
"preferences": [
"spicy",
"hot",
"formal multi-course"
]
}
Criteria
0(value: 0.0): "Violates an explicit prohibition"1(value: 0.25): "Permitted; none of three preferences met"2(value: 0.5): "Permitted; one preference met"3(value: 0.75): "Permitted; two preferences met"4(value: 1.0): "Permitted; all three preferences met"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Spiciness and temperature match, service does not.
contextual-score-083
Instruction
"Score this study's applicability to the specified review question."
State
[
"Study: adults / tutoring / reading / randomized.",
"Review: adults / tutoring / balance / observational. Categories are exact and disjoint."
]
Criteria
0(value: 0.0): "Matches none of population, intervention, outcome or design"1(value: 0.25): "Matches exactly one of the four dimensions"2(value: 0.5): "Matches exactly two of the four dimensions"3(value: 0.75): "Matches exactly three of the four dimensions"4(value: 1.0): "Matches all four dimensions"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Population and intervention match.
contextual-score-084
Instruction
"Rate the proposed meeting time for this participant."
State
{
"available": true,
"meeting": "Tuesday 10:00, 30 minutes, remote",
"preferences": [
"morning",
"at most 30 minutes",
"in person"
]
}
Criteria
0(value: 0.0): "Outside a mandatory availability window"1(value: 0.25): "Available but violates all three preferences"2(value: 0.5): "Available and meets one preference"3(value: 0.75): "Available and meets two preferences"4(value: 1.0): "Available and meets all three preferences"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Time and duration match, mode does not.
contextual-score-085
Instruction
"How much of the four-part checklist is logically established by the record? Treat unestablished items as unmet, not false."
State
{
"checklist": [
"approved",
"signed",
"funded",
"scheduled"
],
"record": "Approved, signed, fully funded; no date has been agreed."
}
Criteria
0(value: 0.0): "Zero items established"1(value: 0.25): "One item established"2(value: 0.5): "Two items established"3(value: 0.75): "Three items established"4(value: 1.0): "Four items established"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Three conditions hold; scheduling does not.
contextual-score-086
Instruction
"Rate this venue for the attendee using the rubric's priority order."
State
[
"Venue: step-free entry, quiet room, near transit, free entry; no other facilities.",
"Required: step-free entry. Preferences: quiet room, cafe, parking."
]
Criteria
0(value: 0.0): "A required access feature is absent"1(value: 0.25): "Required access is present; none of three preferences met"2(value: 0.5): "Required access is present; one preference met"3(value: 0.75): "Required access is present; two preferences met"4(value: 1.0): "Required access is present; all three preferences met"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Only the quiet room preference is satisfied.
contextual-score-087
Instruction
"Rate this venue for the attendee using the rubric's priority order."
State
"Venue: step-free entry, quiet room, near transit, free entry; no other facilities. Required: step-free entry. Preferences: parking, cafe, outdoor seating."
Criteria
0(value: 0.0): "A required access feature is absent"1(value: 0.25): "Required access is present; none of three preferences met"2(value: 0.5): "Required access is present; one preference met"3(value: 0.75): "Required access is present; two preferences met"4(value: 1.0): "Required access is present; all three preferences met"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
Required access exists but none of the preferences does.
contextual-score-088
Instruction
"How suitable is this investment-like game for the player's stated preferences? This is a fictional game, not financial advice."
State
[
"Game: lose at most 10 tokens, 30 minutes, book prize.",
"Player: loss limit 20, prefers 30 minutes and a toy."
]
Criteria
0(value: 0.0): "Violates the maximum acceptable loss"1(value: 0.25): "Loss limit respected, but neither preferred feature met"2(value: 0.5): "Loss limit respected, only preferred duration met"3(value: 0.75): "Loss limit respected, only preferred prize met"4(value: 1.0): "Loss limit respected, preferred duration and prize both met"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Only duration matches.
contextual-score-089
Instruction
"Score this purchase against the buyer's budget and needs."
State
[
"Item: 80 credits, prints and scans only, two-year warranty.",
"Buyer: budget 100; needs faxing and laminating; wants two-year warranty."
]
Criteria
0(value: 0.0): "Exceeds the hard budget"1(value: 0.25): "Within budget but meets neither required function"2(value: 0.5): "Within budget and meets exactly one required function"3(value: 0.75): "Within budget and meets both functions but not the desired warranty"4(value: 1.0): "Within budget and meets both functions and desired warranty"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
Neither required function is present.
contextual-score-090
Instruction
"Rate the translation's fidelity to the source as disambiguated by the context."
State
"Translation: 'Send the file to Lee today.' Source, unambiguously in this context: delete the file today, without sending it."
Criteria
0(value: 0.0): "Reverses the central instruction"1(value: 0.25): "Changes the central action to a different action"2(value: 0.5): "Preserves action but gets a required time or recipient wrong"3(value: 0.75): "Preserves all required meaning but adds an unsupported detail"4(value: 1.0): "Preserves all and only the required meaning"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
Sending substitutes a different central action for deleting.
contextual-score-091
Instruction
"Rate the translation's fidelity to the source as disambiguated by the context."
State
"Translation: 'Send the file to Lee today.' The source requires exactly that action, recipient and day, with no other meaning."
Criteria
0(value: 0.0): "Reverses the central instruction"1(value: 0.25): "Changes the central action to a different action"2(value: 0.5): "Preserves action but gets a required time or recipient wrong"3(value: 0.75): "Preserves all required meaning but adds an unsupported detail"4(value: 1.0): "Preserves all and only the required meaning"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
The translation exactly preserves the stipulated meaning.
contextual-score-092
Instruction
"How strongly does this record support the claim under the specified evidence rubric?"
State
"Claim: the valve was closed at noon. One calibrated position log directly records closed at noon; there is no other evidence."
Criteria
0(value: 0.0): "The record establishes the claim is false"1(value: 0.25): "No relevant evidence either way"2(value: 0.5): "Relevant but inconclusive indication"3(value: 0.75): "One direct reliable observation supports the claim"4(value: 1.0): "Two independent direct reliable observations support the claim"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
There is one direct reliable observation.
contextual-score-093
Instruction
"Rate this analytics plan against the client's explicit privacy agreement."
State
"Plan: anonymous counts report visits, purchases and referrals. The agreement permits anonymous counts, and those are exactly the three requested analyses."
Criteria
0(value: 0.0): "Violates a mandatory restriction"1(value: 0.25): "Complies, but provides none of three requested analyses"2(value: 0.5): "Complies and provides one requested analysis"3(value: 0.75): "Complies and provides two requested analyses"4(value: 1.0): "Complies and provides all three requested analyses"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
All requested analyses are provided compliantly.
contextual-score-094
Instruction
"Rate this venue for the attendee using the rubric's priority order."
State
{
"preferences": [
"quiet room",
"near transit",
"cafe"
],
"required": "step-free entry",
"venue_features": [
"step-free entry",
"quiet room",
"near transit",
"free entry"
]
}
Criteria
0(value: 0.0): "A required access feature is absent"1(value: 0.25): "Required access is present; none of three preferences met"2(value: 0.5): "Required access is present; one preference met"3(value: 0.75): "Required access is present; two preferences met"4(value: 1.0): "Required access is present; all three preferences met"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Two preferences are present, but no cafe.
contextual-score-095
Instruction
"Score this purchase against the buyer's budget and needs."
State
"Item: 80 credits, prints and scans, two-year warranty. Buyer: budget 100, needs printing and scanning, wants a two-year warranty."
Criteria
0(value: 0.0): "Exceeds the hard budget"1(value: 0.25): "Within budget but meets neither required function"2(value: 0.5): "Within budget and meets exactly one required function"3(value: 0.75): "Within budget and meets both functions but not the desired warranty"4(value: 1.0): "Within budget and meets both functions and desired warranty"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 1.0}
Budget, functions and warranty all fit.
contextual-score-096
Instruction
"Rate how well this reply follows the recipient's communication brief."
State
{
"confidentiality": "Project naming permitted",
"reply": "Hi Sam, Project Cedar is approved. Please reply today.",
"style_requirements": [
"greeting",
"project name",
"no action request"
]
}
Criteria
0(value: 0.0): "Violates the brief's mandatory confidentiality rule"1(value: 0.25): "Confidentiality respected; none of three style requirements met"2(value: 0.5): "Confidentiality respected; one style requirement met"3(value: 0.75): "Confidentiality respected; two style requirements met"4(value: 1.0): "Confidentiality respected; all three style requirements met"
Target: {'0': 0.0, '1': 0.0, '2': 0.0, '3': 1.0, '4': 0.0}
Greeting and project naming match, but the action request does not.
contextual-score-097
Instruction
"How much of the four-part checklist is logically established by the record? Treat unestablished items as unmet, not false."
State
"Checklist: approved, signed, funded, scheduled. Record: explicitly approved; not known to be unsigned; funding undecided; scheduling undecided. Lack of known non-signature does not establish signing."
Criteria
0(value: 0.0): "Zero items established"1(value: 0.25): "One item established"2(value: 0.5): "Two items established"3(value: 0.75): "Three items established"4(value: 1.0): "Four items established"
Target: {'0': 0.0, '1': 1.0, '2': 0.0, '3': 0.0, '4': 0.0}
Only approval is established.
contextual-score-098
Instruction
"How well does the proposal fit this caucus's stated priorities? Use the stated priority hierarchy, not your own politics."
State
{
"non_negotiable": "None",
"primary": "Reduce city debt",
"proposal": "Build publicly owned homes using new city debt.",
"secondary": "More homes"
}
Criteria
0(value: 0.0): "Violates a non-negotiable commitment"1(value: 0.25): "Meets no priority but violates no non-negotiable"2(value: 0.5): "Meets only the secondary priority"3(value: 0.75): "Meets only the primary priority"4(value: 1.0): "Meets both priorities without a prohibited feature"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
Only the secondary housing priority is met.
contextual-score-099
Instruction
"How well does this menu fit the diner's own requirements? Apply prohibitions before preferences."
State
{
"diner": {
"preferences": [
"spicy",
"hot",
"low cost"
],
"prohibition": "No lentils"
},
"menu": "Spicy lentil stew, served hot, low cost."
}
Criteria
0(value: 0.0): "Violates an explicit prohibition"1(value: 0.25): "Permitted; none of three preferences met"2(value: 0.5): "Permitted; one preference met"3(value: 0.75): "Permitted; two preferences met"4(value: 1.0): "Permitted; all three preferences met"
Target: {'0': 1.0, '1': 0.0, '2': 0.0, '3': 0.0, '4': 0.0}
Lentils are prohibited regardless of preferences.
contextual-score-100
Instruction
"How urgent is the response under this fictional support contract?"
State
[
"Contract: a non-blocking functional defect gets a response within one day.",
"Ticket: a functional defect with a documented workaround and no blocked users."
]
Criteria
0(value: 0.0): "No response required"1(value: 0.25): "Respond within seven days"2(value: 0.5): "Respond within one day"3(value: 0.75): "Respond within one hour"4(value: 1.0): "Respond immediately"
Target: {'0': 0.0, '1': 0.0, '2': 1.0, '3': 0.0, '4': 0.0}
The defect fits the one-day category.