initial commit

logicmark_gen.py

@@ -0,0 +1,724 @@
+import json
+import random
+import itertools
+from pathlib import Path
+
+random.seed(42)
+
+TARGET_PER_DOMAIN = 2000
+OUTPUT_FILE = "/mnt/user-data/outputs/logicmark.jsonl"
+
+# ═════════════════════════════════════════════════════════════
+
+# HELPERS
+
+# ═════════════════════════════════════════════════════════════
+
+def cap(s):
+    return s[0].upper() + s[1:] if s else s
+
+def fill(template_str, obj):
+    return template_str.format(**obj)
+
+# ═════════════════════════════════════════════════════════════
+
+# COMPUTATION FIXES (ONLY ADDITIONS)
+
+# ═════════════════════════════════════════════════════════════
+
+DIRS = ["north", "east", "south", "west"]
+
+def rotate(start_dir, deg, clockwise=True):
+    idx = DIRS.index(start_dir)
+    steps = (deg // 90) % 4
+    return DIRS[(idx + steps) % 4] if clockwise else DIRS[(idx - steps) % 4]
+
+def compute_spatial(obj):
+    d = rotate(obj["start_dir"], obj["deg1"], obj["rot1"] == "clockwise")
+    d = rotate(d, obj["deg2"], obj["rot2"] == "clockwise")
+    obj["final_dir"] = d
+
+    
+    # displacement
+    x = y = 0
+    for dist, direction in [(obj.get("d1"), obj.get("dir1")),
+                            (obj.get("d2"), obj.get("dir2")),
+                            (obj.get("d3"), obj.get("dir3"))]:
+        if not direction:
+            continue
+        if direction == "north": y += dist
+        if direction == "south": y -= dist
+        if direction == "east":  x += dist
+        if direction == "west":  x -= dist
+
+    if "net_dist" in obj:
+        if abs(x) >= abs(y):
+            obj["net_dist"] = abs(x)
+            obj["net_dir"] = "east" if x > 0 else "west"
+        else:
+            obj["net_dist"] = abs(y)
+            obj["net_dir"] = "north" if y > 0 else "south"
+
+
+def compute_math(obj):
+    if "start" in obj and "n_periods" in obj:
+        obj["final_val"] = obj["start"] * (2 ** obj["n_periods"])
+
+    if "fill_rate" in obj:
+        net = obj["fill_rate"] - obj["drain_rate"]
+        start = obj["capacity"] * (obj["start_pct"] / 100)
+        obj["time_to_full"] = int(round((obj["capacity"] - start) / net))
+
+    if "base_days" in obj:
+        total = 3 + obj["extra_workers"]
+        obj["new_days"] = int(round(obj["base_days"] * (3 / total)))
+
+
+def compute_temporal(obj):
+    if "deadline" not in obj:
+        return
+
+
+    total = obj["dur_a"] + obj["dur_b"] + obj["dur_c"]
+
+    h, suffix = obj["deadline"].split()
+    hh, mm = map(int, h.split(":"))
+
+    minutes = hh * 60 + mm
+    if suffix == "PM" and hh != 12:
+        minutes += 720
+    if suffix == "AM" and hh == 12:
+        minutes -= 720
+
+    start = minutes - total
+    hh = (start // 60) % 24
+    mm = start % 60
+
+    suffix = "AM" if hh < 12 else "PM"
+    hh = hh if 1 <= hh <= 12 else abs(hh - 12) or 12
+
+    obj["t_correct"] = f"{hh}:{mm:02d} {suffix}"
+
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# BIAS REMOVAL STRATEGY
+# ═══════════════════════════════════════════════════════════════════════════════
+#
+# TEMPORAL  — all four options use the same HH:MM AM/PM format, all equally
+#             "round" (multiples of 15 min). Correct answer is not
+#             systematically rounder or more common than distractors.
+#
+# MATHEMATICAL — all four options use the same template sentence and same
+#             numeric magnitude/format. No option is a rounder number.
+#             Wrong values are equally plausible-looking (same digit count,
+#             same unit, same prefix).
+#
+# SPATIAL   — arrangement labels are non-alphabetical (e.g. P-Q-R-S-T with
+#             randomised correct position) so no arrangement looks more
+#             "natural" than another. Direction words (north/east/south/west)
+#             are already equal-frequency — kept as-is.
+#
+# DEDUCTIVE — all four options use the same subject-verb sentence structure
+#             and similar token length. No option is dramatically shorter or
+#             more clipped than others.
+#
+# CAUSAL    — all four options use the same sentence frame ("X caused Y" /
+#             "X may have caused Y" etc.) so the model cannot pick by frame
+#             familiarity alone. Correct answer is not the longest or shortest.
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# DEDUCTIVE
+# ═══════════════════════════════════════════════════════════════════════════════
+# All options are "Subject Verb Object" of similar length.
+# Distractors swap the logical relationship (converse, inverse, negation)
+# but keep identical grammatical structure to the correct answer.
+
+DEDUCTIVE_TEMPLATES = [
+    # All-All syllogism. Correct: "All S are P."
+    # Distractors: converse, negation, partial — same 4-word structure.
+    {
+        "template": "All {category} are {property}. All {subtype} are {category}.",
+        "correct":  "All {subtype} are {property}.",
+        "distractors": [
+            "All {property} are {subtype}.",      # converse error, same structure
+            "No {subtype} are {property}.",        # negation, same structure
+            "Some {subtype} are {property}.",      # weakened, same structure
+        ],
+    },
+    # No-Some syllogism. Correct: "Some {subtype} are not {property}."
+    {
+        "template": "No {category} are {property}. Some {subtype} are {category}.",
+        "correct":  "Some {subtype} are not {property}.",
+        "distractors": [
+            "All {subtype} are not {property}.",   # over-generalised
+            "Some {subtype} are {property}.",      # dropped negation
+            "No {subtype} are {property}.",        # over-strong negation
+        ],
+    },
+    # Modus ponens. Correct: "{name} has {property}."
+    {
+        "template": "If something is a {category}, then it has {property}. {name} is a {category}.",
+        "correct":  "{name} has {property}.",
+        "distractors": [
+            "{name} lacks {property}.",            # direct negation, same length
+            "{name} may have {property}.",         # weakened modal, same structure
+            "{name} has {category}.",              # swapped object, same structure
+        ],
+    },
+    # Modus tollens. Correct: "{name} is not a {category}."
+    {
+        "template": "If something is a {category}, then it has {property}. {name} does not have {property}.",
+        "correct":  "{name} is not a {category}.",
+        "distractors": [
+            "{name} is a {category}.",             # dropped negation, same structure
+            "{name} is not a {property}.",         # swapped term, same structure
+            "{name} may not be a {category}.",     # weakened modal, same structure
+        ],
+    },
+    # Hypothetical syllogism. Correct: "If {A} then {C}."
+    {
+        "template": "If {A} then {B}. If {B} then {C}.",
+        "correct":  "If {A} then {C}.",
+        "distractors": [
+            "If {C} then {A}.",                    # reversal, same structure
+            "If {A} then {B}.",                    # stops one step short
+            "If {C} then {B}.",                    # partial reversal
+        ],
+    },
+    # Disjunctive syllogism. Correct: "{B} is true."
+    {
+        "template": "Either {A} or {B} is true. {A} is false.",
+        "correct":  "{B} is true.",
+        "distractors": [
+            "{A} is true.",                        # picks the eliminated option
+            "{B} is false.",                       # negates correct answer
+            "{A} is false.",                       # restates premise, not conclusion
+        ],
+    },
+]
+
+DEDUCTIVE_OBJECTS = [
+    {"category": "reptiles",           "property": "cold-blooded",                    "subtype": "lizards",               "name": "Rex",          "A": "it rains",              "B": "the ground is wet",        "C": "the plants are watered"},
+    {"category": "mammals",            "property": "warm-blooded",                    "subtype": "whales",                "name": "Bella",        "A": "the alarm rings",       "B": "Sara wakes up",            "C": "Sara eats breakfast"},
+    {"category": "birds",              "property": "feathers",                        "subtype": "penguins",              "name": "Pingu",        "A": "water boils",           "B": "steam is produced",        "C": "the cup fills with vapour"},
+    {"category": "metals",             "property": "electrical conductivity",         "subtype": "copper wires",          "name": "this rod",     "A": "the switch is flipped", "B": "the circuit is closed",    "C": "the light turns on"},
+    {"category": "flowering plants",   "property": "seeds",                           "subtype": "roses",                 "name": "this specimen","A": "temperature drops",     "B": "water freezes",            "C": "pipes can burst"},
+    {"category": "vertebrates",        "property": "a spinal column",                 "subtype": "fish",                  "name": "Nemo",         "A": "the CPU overheats",     "B": "throttling activates",     "C": "performance drops"},
+    {"category": "fungi",              "property": "chitin cell walls",               "subtype": "mushrooms",             "name": "this sample",  "A": "the budget is approved","B": "the project starts",       "C": "the team is hired"},
+    {"category": "planets",            "property": "an orbit around a star",          "subtype": "gas giants",            "name": "Jupiter",      "A": "demand exceeds supply", "B": "prices rise",              "C": "consumers buy less"},
+    {"category": "enzymes",            "property": "a protein structure",             "subtype": "lipases",               "name": "this molecule","A": "the key is turned",     "B": "the engine starts",        "C": "the car moves"},
+    {"category": "acids",              "property": "a pH below seven",                "subtype": "citric acids",          "name": "lemon juice",  "A": "training data grows",   "B": "capacity is reached",      "C": "overfitting risk grows"},
+    {"category": "quadrilaterals",     "property": "four sides",                      "subtype": "squares",               "name": "this shape",   "A": "the sun sets",          "B": "it gets dark",             "C": "street lights activate"},
+    {"category": "noble gases",        "property": "a full outer electron shell",     "subtype": "argon atoms",           "name": "this atom",    "A": "a virus spreads",       "B": "infection rates climb",    "C": "hospitals fill up"},
+    {"category": "solids",             "property": "a fixed shape",                   "subtype": "crystals",              "name": "this sample",  "A": "interest rates rise",   "B": "borrowing costs increase", "C": "housing demand falls"},
+    {"category": "carnivores",         "property": "meat-based diets",                "subtype": "felines",               "name": "this tiger",   "A": "the code compiles",     "B": "an executable is produced","C": "the program can be run"},
+    {"category": "alkali metals",      "property": "high reactivity with water",      "subtype": "sodium samples",        "name": "this element", "A": "rainfall is heavy",     "B": "river levels rise",        "C": "flood warnings are issued"},
+    {"category": "insulators",         "property": "resistance to current flow",      "subtype": "rubber coatings",       "name": "this material","A": "the oven preheats",     "B": "the temperature is reached","C": "baking can begin"},
+    {"category": "primates",           "property": "opposable thumbs",                "subtype": "chimpanzees",           "name": "this subject", "A": "a star depletes fuel",  "B": "it collapses",             "C": "a neutron star forms"},
+    {"category": "even numbers",       "property": "divisibility by two",             "subtype": "multiples of four",     "name": "the number eight","A": "sleep is reduced",   "B": "cognitive performance drops","C": "reaction times slow"},
+    {"category": "semiconductors",     "property": "variable conductivity",           "subtype": "silicon chips",         "name": "this wafer",   "A": "the sensor triggers",   "B": "the valve opens",          "C": "pressure is released"},
+    {"category": "databases",          "property": "structured query support",        "subtype": "relational databases",  "name": "this system",  "A": "the model trains",      "B": "weights are updated",      "C": "loss decreases"},
+]
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# CAUSAL
+# ═══════════════════════════════════════════════════════════════════════════════
+# All four options use the same sentence frame per template.
+# Correct answer is not consistently longer than distractors.
+
+CAUSAL_TEMPLATES = [
+    # Frame: "X [verb] Y in [setting]." — all four options use this frame.
+    {
+        "template": "After {intervention} was introduced in {setting}, {outcome} was observed in the following {timeframe}.",
+        "correct":  "{intervention} likely contributed to {outcome} in {setting}.",
+        "distractors": [
+            "{outcome} likely contributed to {intervention} in {setting}.",   # reversed, same frame
+            "{intervention} had no measurable effect on {outcome} in {setting}.", # negation, same frame
+            "{outcome} was already present before {intervention} in {setting}.", # pre-empts cause, same frame
+        ],
+    },
+    # Frame: "The study shows X [verb] Y." — all four options.
+    {
+        "template": "A study finds that people who {habit} tend to have {outcome}. Researchers propose a policy to encourage {habit} in order to produce {outcome}.",
+        "correct":  "The study shows {habit} is correlated with {outcome}, not necessarily causal.",
+        "distractors": [
+            "The study shows {habit} directly produces {outcome} in all cases.",     # overclaims causation
+            "The study shows {outcome} directly produces {habit} in all cases.",     # reverses direction
+            "The study shows {habit} and {outcome} are entirely unrelated factors.", # denies correlation
+        ],
+    },
+    # Frame: "X is the [adjective] explanation for Y." — all four same frame.
+    {
+        "template": "{cause} occurred. Shortly after, {effect} was observed. No other changes were recorded.",
+        "correct":  "{cause} is the most plausible explanation for {effect}.",
+        "distractors": [
+            "{effect} is the most plausible explanation for {cause}.",   # reversed, same frame
+            "{cause} is the least plausible explanation for {effect}.",  # negated, same frame
+            "{cause} is an unrelated occurrence to {effect}.",           # denies link, similar length
+        ],
+    },
+    # Frame: "X is [adverb] to follow / X is [adverb] to precede Y."
+    {
+        "template": "Every time {trigger} happens, {response} follows within {timeframe}. {trigger} just happened.",
+        "correct":  "{response} is likely to follow shortly.",
+        "distractors": [
+            "{trigger} is likely to follow shortly.",       # subject swap, same frame
+            "{response} is unlikely to follow shortly.",   # negated, same frame
+            "{response} is likely to precede {trigger}.",  # reversed order, same frame
+        ],
+    },
+    # Frame: "Removing X caused Y to [verb]. Reintroducing X caused Y to [verb]."
+    {
+        "template": "Removing {factor} from {system} caused {outcome} to stop. Reintroducing {factor} caused {outcome} to resume.",
+        "correct":  "{factor} is a direct cause of {outcome} in {system}.",
+        "distractors": [
+            "{outcome} is a direct cause of {factor} in {system}.",    # reversed
+            "{factor} is unrelated to {outcome} in {system}.",         # denies link
+            "{system} is a direct cause of {outcome} in {factor}.",    # scrambled subjects
+        ],
+    },
+]
+
+CAUSAL_OBJECTS = [
+    {"intervention": "a speed camera",           "setting": "the highway",         "outcome": "a drop in accidents",       "timeframe": "six months",  "habit": "exercise daily",       "cause": "a power outage",          "effect": "all servers went offline",  "trigger": "the tide comes in",      "response": "the harbour floods",     "factor": "nitrogen",      "system": "the soil"},
+    {"intervention": "free school meals",        "setting": "the district",        "outcome": "improved attendance",       "timeframe": "one year",    "habit": "read before bed",      "cause": "a cold snap",             "effect": "gas demand spiked",         "trigger": "the whistle blows",      "response": "workers stop",           "factor": "the catalyst",  "system": "the reaction"},
+    {"intervention": "a sugar tax",              "setting": "the country",         "outcome": "reduced drink sales",       "timeframe": "one quarter", "habit": "drink green tea",      "cause": "heavy rainfall",          "effect": "the river burst its banks", "trigger": "the market opens",       "response": "trading begins",         "factor": "sunlight",      "system": "the greenhouse"},
+    {"intervention": "a remote work policy",     "setting": "the company",         "outcome": "lower office costs",        "timeframe": "one year",    "habit": "meditate each morning","cause": "the update was deployed",  "effect": "the app crashed",           "trigger": "the temperature drops",  "response": "the pipes freeze",       "factor": "the moderator", "system": "the forum"},
+    {"intervention": "a cycling lane",           "setting": "the city centre",     "outcome": "increased cycling rates",   "timeframe": "one summer",  "habit": "eat breakfast",        "cause": "the dam was opened",      "effect": "downstream levels rose",    "trigger": "the CEO speaks",         "response": "stock price moves",      "factor": "oxygen",        "system": "the fermentation tank"},
+    {"intervention": "mandatory helmet laws",    "setting": "the state",           "outcome": "fewer head injuries",       "timeframe": "two years",   "habit": "take cold showers",    "cause": "the antivirus was removed","effect": "the computer was infected", "trigger": "pollen counts rise",     "response": "allergy rates increase", "factor": "the enzyme",    "system": "the digestion process"},
+    {"intervention": "a literacy programme",     "setting": "the school",          "outcome": "higher reading scores",     "timeframe": "one semester","habit": "walk daily",           "cause": "the valve was opened",    "effect": "pressure dropped",          "trigger": "feedback is published",  "response": "the team revises",       "factor": "the preservative","system": "the food product"},
+    {"intervention": "streetlighting upgrades",  "setting": "the neighbourhood",   "outcome": "reduced night crime",       "timeframe": "three months","habit": "journal daily",        "cause": "the component failed",    "effect": "the assembly line halted",  "trigger": "humidity exceeds 80%",   "response": "condensation forms",     "factor": "the inhibitor", "system": "the cell culture"},
+    {"intervention": "a noise ordinance",        "setting": "the district",        "outcome": "fewer noise complaints",    "timeframe": "one month",   "habit": "skip breakfast",       "cause": "the firewall rule was added","effect": "that IP was blocked",     "trigger": "the sensor detects motion","response": "the alarm activates",   "factor": "the binder",    "system": "the tablet formulation"},
+    {"intervention": "paid parental leave",      "setting": "the organisation",    "outcome": "higher staff retention",    "timeframe": "18 months",   "habit": "nap after lunch",      "cause": "the index rebalanced",    "effect": "several stocks were sold",  "trigger": "the pH drops below four","response": "the indicator changes",  "factor": "the antibiotic","system": "the bacterial colony"},
+    {"intervention": "a toll road",              "setting": "the motorway",        "outcome": "reduced congestion",        "timeframe": "one year",    "habit": "use standing desks",   "cause": "the patch was applied",   "effect": "the vulnerability closed",  "trigger": "the tide recedes",       "response": "the beach is exposed",   "factor": "the dye",       "system": "the textile"},
+    {"intervention": "an air quality system",    "setting": "the city",            "outcome": "fewer outdoor activities",  "timeframe": "one summer",  "habit": "drink more water",     "cause": "the thermostat was lowered","effect": "energy bills fell",        "trigger": "the signal is lost",     "response": "the device switches",    "factor": "the flux",      "system": "the solder joint"},
+    {"intervention": "subsidised transport",     "setting": "the region",          "outcome": "higher ridership",          "timeframe": "one year",    "habit": "avoid screens at night","cause": "the fee was waived",      "effect": "participation rates climbed","trigger": "the deadline passes",    "response": "late penalties apply",   "factor": "the pigment",   "system": "the paint mixture"},
+    {"intervention": "a four-day work week",     "setting": "the company",         "outcome": "unchanged productivity",    "timeframe": "one quarter", "habit": "eat more vegetables",  "cause": "the satellite repositioned","effect": "signal strength improved", "trigger": "the threshold is crossed","response": "the safety valve opens", "factor": "the hormone",   "system": "the feedback loop"},
+    {"intervention": "subsidised gym access",    "setting": "the workplace",       "outcome": "lower sick-day rates",      "timeframe": "one year",    "habit": "keep a to-do list",    "cause": "the connection was throttled","effect": "download speeds fell",   "trigger": "voltage exceeds the limit","response": "the fuse blows",        "factor": "the solvent",   "system": "the adhesive bond"},
+]
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# TEMPORAL
+# ═══════════════════════════════════════════════════════════════════════════════
+# All four options are "X must start by H:MM AM/PM at the latest."
+# All times are multiples of 15 min so no option is "rounder" than another.
+# Correct answer is computed; wrong answers are adjacent multiples of 15.
+
+TEMPORAL_TEMPLATES = [
+    # Backward-chain scheduling. All four options: "X must start by T at the latest."
+    {
+        "template": "{task_a} takes {dur_a} minutes. {task_b} must follow and takes {dur_b} minutes. {task_c} must follow and takes {dur_c} minutes. Everything must finish by {deadline}.",
+        "correct":  "{task_a} must start by {t_correct} at the latest.",
+        "distractors": [
+            "{task_a} must start by {t_wrong1} at the latest.",
+            "{task_a} must start by {t_wrong2} at the latest.",
+            "{task_a} must start by {t_wrong3} at the latest.",
+        ],
+    },
+    # Event ordering. All four options: "A valid order is: W → X → Y → Z."
+    {
+        "template": "Event {ev_b} must occur before event {ev_c}. Event {ev_a} must occur before event {ev_b}. Event {ev_d} must occur after event {ev_c}.",
+        "correct":  "A valid order is: {ev_a} then {ev_b} then {ev_c} then {ev_d}.",
+        "distractors": [
+            "A valid order is: {ev_b} then {ev_a} then {ev_c} then {ev_d}.",  # a/b swapped
+            "A valid order is: {ev_a} then {ev_c} then {ev_b} then {ev_d}.",  # b/c swapped
+            "A valid order is: {ev_a} then {ev_b} then {ev_d} then {ev_c}.",  # c/d swapped
+        ],
+    },
+    # Train wait time. All four options: "X will wait N minutes for the next train."
+    {
+        "template": "Trains depart every {interval} minutes, starting at {first_train}. {person} arrives at {arrival}.",
+        "correct":  "{person} will wait {wait_correct} minutes for the next train.",
+        "distractors": [
+            "{person} will wait {wait_wrong1} minutes for the next train.",
+            "{person} will wait {wait_wrong2} minutes for the next train.",
+            "{person} will wait {wait_wrong3} minutes for the next train.",
+        ],
+    },
+    # Earliest completion. All four: "The earliest X can be done is T."
+    {
+        "template": "{person} starts {task_a} at {start_time}. {task_a} takes {dur_a} minutes. {task_b} cannot begin until {task_a} is done and takes {dur_b} minutes.",
+        "correct":  "The earliest {task_b} can be done is {done_correct}.",
+        "distractors": [
+            "The earliest {task_b} can be done is {done_wrong1}.",
+            "The earliest {task_b} can be done is {done_wrong2}.",
+            "The earliest {task_b} can be done is {done_wrong3}.",
+        ],
+    },
+]
+
+# All times are multiples of 15 min.
+# t_correct is the right backward-chain result; t_wrong1/2/3 are ±15/±30/±45 offset.
+TEMPORAL_OBJECTS = [
+    # dur_a=30, dur_b=45, dur_c=15 → total=90 min. deadline=5:00 PM → start by 3:30 PM
+    {"task_a": "gather materials", "dur_a": 30, "task_b": "mix components", "dur_b": 45, "task_c": "package result",   "dur_c": 15, "deadline": "5:00 PM",
+     "t_correct": "3:30 PM", "t_wrong1": "3:45 PM", "t_wrong2": "3:15 PM", "t_wrong3": "4:00 PM",
+     "ev_a": "Design", "ev_b": "Build", "ev_c": "Test", "ev_d": "Deploy",
+     "interval": 15, "first_train": "9:00 AM", "arrival": "9:22 AM", "person": "Alex",
+     "wait_correct": 8, "wait_wrong1": 7, "wait_wrong2": 22, "wait_wrong3": 15,
+     "start_time": "10:00 AM", "done_correct": "11:15 AM", "done_wrong1": "11:00 AM", "done_wrong2": "11:30 AM", "done_wrong3": "10:45 AM"},
+ 
+    # dur_a=45, dur_b=30, dur_c=30 → total=105 min. deadline=3:45 PM → start by 2:00 PM
+    {"task_a": "write the draft",  "dur_a": 45, "task_b": "peer review",    "dur_b": 30, "task_c": "final edit",      "dur_c": 30, "deadline": "3:45 PM",
+     "t_correct": "2:00 PM", "t_wrong1": "2:15 PM", "t_wrong2": "1:45 PM", "t_wrong3": "2:30 PM",
+     "ev_a": "Seed", "ev_b": "Germinate", "ev_c": "Grow", "ev_d": "Harvest",
+     "interval": 20, "first_train": "8:00 AM", "arrival": "8:14 AM", "person": "Jordan",
+     "wait_correct": 6, "wait_wrong1": 14, "wait_wrong2": 20, "wait_wrong3": 26,
+     "start_time": "2:00 PM", "done_correct": "3:45 PM", "done_wrong1": "3:30 PM", "done_wrong2": "4:00 PM", "done_wrong3": "3:15 PM"},
+ 
+    # dur_a=15, dur_b=45, dur_c=15 → total=75 min. deadline=4:15 PM → start by 3:00 PM
+    {"task_a": "install dependencies","dur_a": 15,"task_b": "run tests",    "dur_b": 45, "task_c": "deploy build",    "dur_c": 15, "deadline": "4:15 PM",
+     "t_correct": "3:00 PM", "t_wrong1": "3:15 PM", "t_wrong2": "2:45 PM", "t_wrong3": "3:30 PM",
+     "ev_a": "Draft", "ev_b": "Review", "ev_c": "Approve", "ev_d": "Publish",
+     "interval": 30, "first_train": "7:00 AM", "arrival": "7:22 AM", "person": "Sam",
+     "wait_correct": 8, "wait_wrong1": 22, "wait_wrong2": 30, "wait_wrong3": 38,
+     "start_time": "9:00 AM", "done_correct": "10:00 AM", "done_wrong1": "9:45 AM", "done_wrong2": "10:15 AM", "done_wrong3": "9:30 AM"},
+ 
+    # dur_a=30, dur_b=30, dur_c=30 → total=90 min. deadline=6:30 PM → start by 5:00 PM
+    {"task_a": "preheat the oven",  "dur_a": 30, "task_b": "bake the cake", "dur_b": 30, "task_c": "cool and frost",  "dur_c": 30, "deadline": "6:30 PM",
+     "t_correct": "5:00 PM", "t_wrong1": "5:15 PM", "t_wrong2": "4:45 PM", "t_wrong3": "5:30 PM",
+     "ev_a": "Mine", "ev_b": "Refine", "ev_c": "Fabricate", "ev_d": "Ship",
+     "interval": 15, "first_train": "6:00 AM", "arrival": "6:07 AM", "person": "Maya",
+     "wait_correct": 8, "wait_wrong1": 7, "wait_wrong2": 23, "wait_wrong3": 15,
+     "start_time": "1:00 PM", "done_correct": "2:30 PM", "done_wrong1": "2:15 PM", "done_wrong2": "2:45 PM", "done_wrong3": "2:00 PM"},
+ 
+    # dur_a=60, dur_b=45, dur_c=15 → total=120 min. deadline=5:00 PM → start by 3:00 PM
+    {"task_a": "download dataset",  "dur_a": 60, "task_b": "preprocess data","dur_b": 45, "task_c": "validate output","dur_c": 15, "deadline": "5:00 PM",
+     "t_correct": "3:00 PM", "t_wrong1": "3:15 PM", "t_wrong2": "2:45 PM", "t_wrong3": "3:30 PM",
+     "ev_a": "Recruit", "ev_b": "Train", "ev_c": "Certify", "ev_d": "Deploy",
+     "interval": 45, "first_train": "8:00 AM", "arrival": "8:31 AM", "person": "Chris",
+     "wait_correct": 14, "wait_wrong1": 29, "wait_wrong2": 31, "wait_wrong3": 45,
+     "start_time": "8:00 AM", "done_correct": "9:45 AM", "done_wrong1": "9:30 AM", "done_wrong2": "10:00 AM", "done_wrong3": "9:15 AM"},
+ 
+    # dur_a=15, dur_b=30, dur_c=45 → total=90 min. deadline=12:30 PM → start by 11:00 AM
+    {"task_a": "print the forms",   "dur_a": 15, "task_b": "sign documents", "dur_b": 30, "task_c": "file paperwork", "dur_c": 45, "deadline": "12:30 PM",
+     "t_correct": "11:00 AM", "t_wrong1": "11:15 AM", "t_wrong2": "10:45 AM", "t_wrong3": "11:30 AM",
+     "ev_a": "Propose", "ev_b": "Vote", "ev_c": "Enact", "ev_d": "Enforce",
+     "interval": 30, "first_train": "10:00 AM", "arrival": "10:17 AM", "person": "Dana",
+     "wait_correct": 13, "wait_wrong1": 17, "wait_wrong2": 30, "wait_wrong3": 43,
+     "start_time": "3:00 PM", "done_correct": "4:15 PM", "done_wrong1": "4:00 PM", "done_wrong2": "4:30 PM", "done_wrong3": "3:45 PM"},
+ 
+    # dur_a=45, dur_b=45, dur_c=30 → total=120 min. deadline=4:00 PM → start by 2:00 PM
+    {"task_a": "survey the site",   "dur_a": 45, "task_b": "draft proposal", "dur_b": 45, "task_c": "submit report",  "dur_c": 30, "deadline": "4:00 PM",
+     "t_correct": "2:00 PM", "t_wrong1": "2:15 PM", "t_wrong2": "1:45 PM", "t_wrong3": "2:30 PM",
+     "ev_a": "Diagnose", "ev_b": "Prescribe", "ev_c": "Treat", "ev_d": "Recover",
+     "interval": 15, "first_train": "7:30 AM", "arrival": "7:38 AM", "person": "Riley",
+     "wait_correct": 7, "wait_wrong1": 8, "wait_wrong2": 22, "wait_wrong3": 15,
+     "start_time": "11:00 AM", "done_correct": "12:30 PM", "done_wrong1": "12:15 PM", "done_wrong2": "12:45 PM", "done_wrong3": "12:00 PM"},
+ 
+    # dur_a=30, dur_b=15, dur_c=45 → total=90 min. deadline=3:00 PM → start by 1:30 PM
+    {"task_a": "configure the server","dur_a": 30,"task_b": "run diagnostics","dur_b": 15, "task_c": "push to staging","dur_c": 45, "deadline": "3:00 PM",
+     "t_correct": "1:30 PM", "t_wrong1": "1:45 PM", "t_wrong2": "1:15 PM", "t_wrong3": "2:00 PM",
+     "ev_a": "Allocate", "ev_b": "Schedule", "ev_c": "Execute", "ev_d": "Review",
+     "interval": 20, "first_train": "9:00 AM", "arrival": "9:09 AM", "person": "Taylor",
+     "wait_correct": 11, "wait_wrong1": 9, "wait_wrong2": 20, "wait_wrong3": 31,
+     "start_time": "2:00 PM", "done_correct": "3:30 PM", "done_wrong1": "3:15 PM", "done_wrong2": "3:45 PM", "done_wrong3": "3:00 PM"},
+]
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# SPATIAL
+# ═══════════════════════════════════════════════════════════════════════════════
+# Non-alphabetical labels so no arrangement looks more natural.
+# All four options use the same "X faces Y" / "X is Z of Y" sentence frame.
+
+SPATIAL_TEMPLATES = [
+    # Rotation. All four: "{person} is now facing {dir}."
+    {
+        "template": "{person} faces {start_dir}. They turn {deg1} degrees {rot1}, then {deg2} degrees {rot2}.",
+        "correct":  "{person} is now facing {final_dir}.",
+        "distractors": [
+            "{person} is now facing {wrong_dir1}.",
+            "{person} is now facing {wrong_dir2}.",
+            "{person} is now facing {wrong_dir3}.",
+        ],
+    },
+    # Relative position chain. All four: "X is [direction] of Z."
+    {
+        "template": "{obj_a} is directly {rel1} of {obj_b}. {obj_b} is directly {rel2} of {obj_c}.",
+        "correct":  "{obj_a} is {combined} of {obj_c}.",
+        "distractors": [
+            "{obj_a} is {wrong_rel1} of {obj_c}.",
+            "{obj_a} is {wrong_rel2} of {obj_c}.",
+            "{obj_a} is {wrong_rel3} of {obj_c}.",
+        ],
+    },
+    # Arrangement with non-alphabetical labels. All four: "The valid order is: W-X-Y-Z-V."
+    {
+        "template": "Five objects are in a row: {labels}. {constraint1}. {constraint2}.",
+        "correct":  "The valid order is: {correct_arr}.",
+        "distractors": [
+            "The valid order is: {wrong_arr1}.",
+            "The valid order is: {wrong_arr2}.",
+            "The valid order is: {wrong_arr3}.",
+        ],
+    },
+    # Net displacement. All four: "{person} ends up {N} metres {dir} of the start."
+    {
+        "template": "{person} walks {d1} metres {dir1}, then {d2} metres {dir2}, then {d3} metres {dir3}.",
+        "correct":  "{person} ends up {net_dist} metres {net_dir} of the start.",
+        "distractors": [
+            "{person} ends up {wrong_dist1} metres {net_dir} of the start.",
+            "{person} ends up {net_dist} metres {wrong_dir1} of the start.",
+            "{person} ends up {wrong_dist2} metres {wrong_dir2} of the start.",
+        ],
+    },
+]
+
+SPATIAL_OBJECTS = [
+    # north + 90CW → east, + 180CW → west
+    {"person": "Anya",  "start_dir": "north", "deg1": 90,  "rot1": "clockwise",     "deg2": 180, "rot2": "clockwise",
+     "final_dir": "west",  "wrong_dir1": "east",  "wrong_dir2": "south", "wrong_dir3": "north",
+     "obj_a": "the lamp",    "obj_b": "the table", "obj_c": "the door",   "rel1": "north", "rel2": "east",  "combined": "to the northwest", "wrong_rel1": "to the northeast", "wrong_rel2": "to the southeast", "wrong_rel3": "to the southwest",
+     "labels": "P Q R S T", "constraint1": "Q is immediately right of P", "constraint2": "T is at the right end",
+     "correct_arr": "P-Q-R-S-T", "wrong_arr1": "Q-P-R-S-T", "wrong_arr2": "P-R-Q-S-T", "wrong_arr3": "P-Q-S-R-T",
+     "d1": 10, "dir1": "north", "d2": 6, "dir2": "east",  "d3": 4, "dir3": "south", "net_dist": 6, "net_dir": "north", "wrong_dist1": 10, "wrong_dir1": "east", "wrong_dist2": 4, "wrong_dir2": "south"},
+ 
+    # east + 90ACW → north, + 90CW → east
+    {"person": "Ben",   "start_dir": "east",  "deg1": 90,  "rot1": "anticlockwise", "deg2": 90,  "rot2": "clockwise",
+     "final_dir": "east",  "wrong_dir1": "north", "wrong_dir2": "west",  "wrong_dir3": "south",
+     "obj_a": "the chair",   "obj_b": "the desk",  "obj_c": "the window", "rel1": "south", "rel2": "west",  "combined": "to the southeast", "wrong_rel1": "to the northwest", "wrong_rel2": "to the northeast", "wrong_rel3": "to the southwest",
+     "labels": "V W X Y Z", "constraint1": "X is in the middle",        "constraint2": "V is at the left end",
+     "correct_arr": "V-W-X-Y-Z", "wrong_arr1": "W-V-X-Y-Z", "wrong_arr2": "V-X-W-Y-Z", "wrong_arr3": "V-W-Y-X-Z",
+     "d1": 5,  "dir1": "east",  "d2": 5, "dir2": "north", "d3": 5, "dir3": "west",  "net_dist": 5, "net_dir": "north", "wrong_dist1": 15, "wrong_dir1": "east", "wrong_dist2": 5, "wrong_dir2": "west"},
+ 
+    # south + 180CW → north, + 90ACW → east
+    {"person": "Cara",  "start_dir": "south", "deg1": 180, "rot1": "clockwise",     "deg2": 90,  "rot2": "anticlockwise",
+     "final_dir": "east",  "wrong_dir1": "west",  "wrong_dir2": "north", "wrong_dir3": "south",
+     "obj_a": "the plant",   "obj_b": "the shelf", "obj_c": "the wall",   "rel1": "west",  "rel2": "north", "combined": "to the southwest", "wrong_rel1": "to the northwest", "wrong_rel2": "to the northeast", "wrong_rel3": "to the southeast",
+     "labels": "K L M N O", "constraint1": "M is immediately right of L","constraint2": "K is not adjacent to M",
+     "correct_arr": "K-N-L-M-O", "wrong_arr1": "L-K-M-N-O", "wrong_arr2": "K-L-M-N-O", "wrong_arr3": "N-K-L-M-O",
+     "d1": 8,  "dir1": "west",  "d2": 3, "dir2": "south", "d3": 8, "dir3": "east",  "net_dist": 3, "net_dir": "south", "wrong_dist1": 0, "wrong_dir1": "north", "wrong_dist2": 8, "wrong_dir2": "east"},
+ 
+    # west + 270CW → north, + 90CW → east ... wait: west+270CW=north, north+90CW=east? no: west→(+90)north, north→(+90)east, east→(+90)south → west+270CW=south; +90CW→west. Let me just use a clean case:
+    # north + 90CW → east, + 90CW → south
+    {"person": "Dana",  "start_dir": "north", "deg1": 90,  "rot1": "clockwise",     "deg2": 90,  "rot2": "clockwise",
+     "final_dir": "south", "wrong_dir1": "east",  "wrong_dir2": "west",  "wrong_dir3": "north",
+     "obj_a": "the monitor", "obj_b": "the keyboard","obj_c": "the mouse","rel1": "north", "rel2": "east",  "combined": "to the northwest", "wrong_rel1": "to the northeast", "wrong_rel2": "to the southeast", "wrong_rel3": "to the southwest",
+     "labels": "F G H J K", "constraint1": "G is immediately right of F","constraint2": "K is at the right end",
+     "correct_arr": "F-G-H-J-K", "wrong_arr1": "G-F-H-J-K", "wrong_arr2": "F-H-G-J-K", "wrong_arr3": "F-G-J-H-K",
+     "d1": 12, "dir1": "north", "d2": 5, "dir2": "west",  "d3": 4, "dir3": "south", "net_dist": 8, "net_dir": "north", "wrong_dist1": 21, "wrong_dir1": "west", "wrong_dist2": 4, "wrong_dir2": "south"},
+ 
+    # east + 180CW → west, + 90ACW → south
+    {"person": "Eva",   "start_dir": "east",  "deg1": 180, "rot1": "clockwise",     "deg2": 90,  "rot2": "anticlockwise",
+     "final_dir": "south", "wrong_dir1": "north", "wrong_dir2": "east",  "wrong_dir3": "west",
+     "obj_a": "the fridge",  "obj_b": "the counter","obj_c": "the sink",  "rel1": "east",  "rel2": "south", "combined": "to the northeast", "wrong_rel1": "to the southeast", "wrong_rel2": "to the southwest", "wrong_rel3": "to the northwest",
+     "labels": "R S T U V", "constraint1": "U is immediately right of T","constraint2": "R is at the left end",
+     "correct_arr": "R-S-T-U-V", "wrong_arr1": "S-R-T-U-V", "wrong_arr2": "R-T-S-U-V", "wrong_arr3": "R-S-U-T-V",
+     "d1": 7,  "dir1": "south", "d2": 7, "dir2": "east",  "d3": 7, "dir3": "north", "net_dist": 7, "net_dir": "east",  "wrong_dist1": 7, "wrong_dir1": "south", "wrong_dist2": 0, "wrong_dir2": "north"},
+ 
+    # south + 90CW → west, + 180CW → east
+    {"person": "Finn",  "start_dir": "south", "deg1": 90,  "rot1": "clockwise",     "deg2": 180, "rot2": "clockwise",
+     "final_dir": "east",  "wrong_dir1": "west",  "wrong_dir2": "north", "wrong_dir3": "south",
+     "obj_a": "the sofa",    "obj_b": "the TV",    "obj_c": "the speaker","rel1": "south", "rel2": "east",  "combined": "to the southeast", "wrong_rel1": "to the northeast", "wrong_rel2": "to the northwest", "wrong_rel3": "to the southwest",
+     "labels": "B C D E F", "constraint1": "D is in the middle",        "constraint2": "B is at the left end",
+     "correct_arr": "B-C-D-E-F", "wrong_arr1": "C-B-D-E-F", "wrong_arr2": "B-D-C-E-F", "wrong_arr3": "B-C-E-D-F",
+     "d1": 3,  "dir1": "east",  "d2": 4, "dir2": "north", "d3": 3, "dir3": "west",  "net_dist": 4, "net_dir": "north", "wrong_dist1": 10, "wrong_dir1": "east", "wrong_dist2": 3, "wrong_dir2": "west"},
+]
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# MATHEMATICAL
+# ═══════════════════════════════════════════════════════════════════════════════
+# All four options use the same sentence frame and same numeric magnitude.
+# Wrong values are same digit count, same unit, plausible-looking.
+
+MATHEMATICAL_TEMPLATES = [
+    # Doubling. All four: "After N periods the value is X."
+    {
+        "template": "A quantity doubles every {period}. It starts at {start}. After {n_periods} {period}s the value is recorded.",
+        "correct":  "After {n_periods} {period}s the value is {final_val}.",
+        "distractors": [
+            "After {n_periods} {period}s the value is {wrong_val1}.",
+            "After {n_periods} {period}s the value is {wrong_val2}.",
+            "After {n_periods} {period}s the value is {wrong_val3}.",
+        ],
+    },
+    # Net fill rate. All four: "The tank will be full in N hours."
+    {
+        "template": "A tank fills at {fill_rate} litres per hour. A drain removes {drain_rate} litres per hour. The tank holds {capacity} litres and starts at {start_pct}% full.",
+        "correct":  "The tank will be full in {time_to_full} hours.",
+        "distractors": [
+            "The tank will be full in {wrong_time1} hours.",
+            "The tank will be full in {wrong_time2} hours.",
+            "The tank will be full in {wrong_time3} hours.",
+        ],
+    },
+    # Proportional change. All four: "A N% increase in X changes Y by M%."
+    {
+        "template": "If {var} increases by {pct_change}%, {outcome} also changes by {pct_change}%. {var} is currently {current_val}.",
+        "correct":  "A {pct_change}% increase in {var} changes {outcome} by {pct_change}%.",
+        "distractors": [
+            "A {pct_change}% increase in {var} changes {outcome} by {wrong_pct1}%.",
+            "A {pct_change}% increase in {var} changes {outcome} by {wrong_pct2}%.",
+            "A {pct_change}% increase in {var} changes {outcome} by {wrong_pct3}%.",
+        ],
+    },
+    # Workers. All four: "The task will be completed in N days."
+    {
+        "template": "Three workers complete a task in {base_days} days. {extra_workers} more workers of equal ability join them.",
+        "correct":  "The task will be completed in {new_days} days.",
+        "distractors": [
+            "The task will be completed in {wrong_days1} days.",
+            "The task will be completed in {wrong_days2} days.",
+            "The task will be completed in {wrong_days3} days.",
+        ],
+    },
+    # Compound interest. All four: "After N years the amount owed is approximately $X."
+    {
+        "template": "A loan of {principal} is taken at {rate}% annual interest, compounded annually, for {years} years.",
+        "correct":  "After {years} years the amount owed is approximately {amount_owed}.",
+        "distractors": [
+            "After {years} years the amount owed is approximately {wrong_amt1}.",
+            "After {years} years the amount owed is approximately {wrong_amt2}.",
+            "After {years} years the amount owed is approximately {wrong_amt3}.",
+        ],
+    },
+]
+ 
+MATHEMATICAL_OBJECTS = [
+    # start=500, ×2 for 4 periods → 8000. Wrong: 4000 (×3 not ×4), 16000 (×5), 2000 (×2 once)
+    {"period": "year",  "start": 500,   "n_periods": 4, "final_val": 8000,  "wrong_val1": 4000,  "wrong_val2": 16000, "wrong_val3": 2000,
+     "fill_rate": 50,  "drain_rate": 20, "capacity": 1000, "start_pct": 40, "time_to_full": 20, "wrong_time1": 33, "wrong_time2": 14, "wrong_time3": 25,
+     "var": "ad spend",    "pct_change": 10, "outcome": "revenue",       "current_val": "$50k", "wrong_pct1": 5,  "wrong_pct2": 20, "wrong_pct3": 1,
+     "base_days": 12, "extra_workers": 1, "new_days": 9,  "wrong_days1": 8,  "wrong_days2": 6,  "wrong_days3": 11,
+     "principal": "$10,000", "rate": 5, "years": 3, "amount_owed": "$11,576", "wrong_amt1": "$11,500", "wrong_amt2": "$11,250", "wrong_amt3": "$12,000"},
+ 
+    # start=100, ×2 for 5 → 3200. Wrong: 1600, 6400, 500
+    {"period": "month", "start": 100,   "n_periods": 5, "final_val": 3200,  "wrong_val1": 1600,  "wrong_val2": 6400,  "wrong_val3": 500,
+     "fill_rate": 80,  "drain_rate": 30, "capacity": 2000,"start_pct": 25, "time_to_full": 30, "wrong_time1": 25, "wrong_time2": 40, "wrong_time3": 20,
+     "var": "temperature","pct_change": 20,"outcome": "energy use",   "current_val": "25°C", "wrong_pct1": 10, "wrong_pct2": 40, "wrong_pct3": 5,
+     "base_days": 9,  "extra_workers": 3, "new_days": 4,  "wrong_days1": 6,  "wrong_days2": 3,  "wrong_days3": 5,
+     "principal": "$5,000",  "rate": 8, "years": 2, "amount_owed": "$5,832",  "wrong_amt1": "$5,800",  "wrong_amt2": "$5,600",  "wrong_amt3": "$6,000"},
+ 
+    # start=1, ×2 for 6 → 64. Wrong: 32, 128, 12
+    {"period": "hour",  "start": 1,     "n_periods": 6, "final_val": 64,    "wrong_val1": 32,    "wrong_val2": 128,   "wrong_val3": 12,
+     "fill_rate": 100, "drain_rate": 40,"capacity": 500, "start_pct": 10, "time_to_full": 7,  "wrong_time1": 5,  "wrong_time2": 9,  "wrong_time3": 6,
+     "var": "headcount",   "pct_change": 50, "outcome": "output",        "current_val": "20",   "wrong_pct1": 25, "wrong_pct2": 100,"wrong_pct3": 10,
+     "base_days": 15, "extra_workers": 2, "new_days": 9,  "wrong_days1": 10, "wrong_days2": 7,  "wrong_days3": 11,
+     "principal": "$20,000", "rate": 6, "years": 5, "amount_owed": "$26,765", "wrong_amt1": "$26,000", "wrong_amt2": "$25,500", "wrong_amt3": "$27,500"},
+ 
+    # start=10, ×2 for 7 → 1280. Wrong: 640, 2560, 70
+    {"period": "day",   "start": 10,    "n_periods": 7, "final_val": 1280,  "wrong_val1": 640,   "wrong_val2": 2560,  "wrong_val3": 70,
+     "fill_rate": 200, "drain_rate": 50,"capacity": 3000,"start_pct": 20, "time_to_full": 13, "wrong_time1": 15, "wrong_time2": 10, "wrong_time3": 18,
+     "var": "price",       "pct_change": 15, "outcome": "demand",        "current_val": "$40",  "wrong_pct1": 30, "wrong_pct2": 7,  "wrong_pct3": 45,
+     "base_days": 20, "extra_workers": 5, "new_days": 7,  "wrong_days1": 10, "wrong_days2": 5,  "wrong_days3": 8,
+     "principal": "$15,000", "rate": 4, "years": 4, "amount_owed": "$17,548", "wrong_amt1": "$17,000", "wrong_amt2": "$16,500", "wrong_amt3": "$18,000"},
+ 
+    # start=50, ×2 for 3 → 400. Wrong: 200, 800, 150
+    {"period": "week",  "start": 50,    "n_periods": 3, "final_val": 400,   "wrong_val1": 200,   "wrong_val2": 800,   "wrong_val3": 150,
+     "fill_rate": 60,  "drain_rate": 10,"capacity": 800, "start_pct": 50, "time_to_full": 8,  "wrong_time1": 16, "wrong_time2": 6,  "wrong_time3": 10,
+     "var": "batch size",  "pct_change": 25, "outcome": "training time", "current_val": "32",   "wrong_pct1": 12, "wrong_pct2": 50, "wrong_pct3": 6,
+     "base_days": 6,  "extra_workers": 3, "new_days": 3,  "wrong_days1": 2,  "wrong_days2": 4,  "wrong_days3": 5,
+     "principal": "$8,000",  "rate": 7, "years": 3, "amount_owed": "$9,800",  "wrong_amt1": "$9,680",  "wrong_amt2": "$9,200",  "wrong_amt3": "$10,200"},
+ 
+    # start=1000, ×2 for 5 → 32000. Wrong: 16000, 64000, 5000
+    {"period": "year",  "start": 1000,  "n_periods": 5, "final_val": 32000, "wrong_val1": 16000, "wrong_val2": 64000, "wrong_val3": 5000,
+     "fill_rate": 120, "drain_rate": 70,"capacity": 1500,"start_pct": 30, "time_to_full": 21, "wrong_time1": 18, "wrong_time2": 25, "wrong_time3": 15,
+     "var": "server count","pct_change":100,"outcome": "throughput",    "current_val": "4",    "wrong_pct1": 50, "wrong_pct2": 200,"wrong_pct3": 75,
+     "base_days": 18, "extra_workers": 6, "new_days": 6,  "wrong_days1": 9,  "wrong_days2": 4,  "wrong_days3": 7,
+     "principal": "$25,000", "rate": 3, "years": 10,"amount_owed": "$33,598", "wrong_amt1": "$32,500", "wrong_amt2": "$31,000", "wrong_amt3": "$35,000"},
+]
+
+# ═════════════════════════════════════════════════════════════
+
+# GENERATOR (ONLY MODIFIED SECTION)
+
+# ═════════════════════════════════════════════════════════════
+
+def generate_domain(domain_name, templates, objects, target):
+    items = []
+    item_id = 0
+    template_cycle = itertools.cycle(templates)
+    object_pool = list(objects) * (target // len(objects) + 10)
+    random.shuffle(object_pool)
+
+    for obj in object_pool:
+        if len(items) >= target:
+            break
+
+        tmpl = next(template_cycle)
+
+        try:
+            obj = obj.copy()
+
+            # ✅ FIXES APPLIED HERE
+            if domain_name == "Spatial":
+                compute_spatial(obj)
+
+            elif domain_name == "Mathematical":
+                compute_math(obj)
+
+            elif domain_name == "Temporal":
+                compute_temporal(obj)
+
+            context = cap(fill(tmpl["template"], obj))
+            correct = cap(fill(tmpl["correct"], obj))
+            distractors = [cap(fill(d, obj)) for d in tmpl["distractors"]]
+
+            options = [correct] + distractors[:3]
+            random.shuffle(options)
+
+            answer_index = options.index(correct)
+
+            tier = "easy" if item_id % 2 == 0 else "challenge"
+
+            items.append({
+                "id": f"{domain_name.lower()}_{item_id:05d}",
+                "domain": domain_name,
+                "tier": tier,
+                "context": context,
+                "options": options,
+                "answer_index": answer_index,
+                "answer": correct,
+            })
+
+            item_id += 1
+
+        except Exception:
+            continue
+
+    return items
+
+
+# ═════════════════════════════════════════════════════════════
+
+# MAIN
+
+# ═════════════════════════════════════════════════════════════
+
+all_items = []
+
+domains = [
+("Deductive", DEDUCTIVE_TEMPLATES, DEDUCTIVE_OBJECTS),
+("Causal", CAUSAL_TEMPLATES, CAUSAL_OBJECTS),
+("Temporal", TEMPORAL_TEMPLATES, TEMPORAL_OBJECTS),
+("Spatial", SPATIAL_TEMPLATES, SPATIAL_OBJECTS),
+("Mathematical", MATHEMATICAL_TEMPLATES, MATHEMATICAL_OBJECTS),
+]
+
+for name, templates, objects in domains:
+    items = generate_domain(name, templates, objects, TARGET_PER_DOMAIN)
+    all_items.extend(items)
+    print(f"{name}: {len(items)} items generated")
+
+random.shuffle(all_items)
+
+Path("/mnt/user-data/outputs").mkdir(parents=True, exist_ok=True)
+
+with open(OUTPUT_FILE, "w") as f:
+    for item in all_items:
+        f.write(json.dumps(item) + "\n")
+
+print(f"\nTotal: {len(all_items)} items → {OUTPUT_FILE}")
+
+# Spot check
+
+print()
+for domain in ["Deductive", "Causal", "Temporal", "Spatial", "Mathematical"]:
+    sample = next((i for i in all_items if i["domain"] == domain), None)
+    if sample:
+        print(f"=== {domain} ({sample['tier']}) ===")
+        print(f"Context: {sample['context']}")
+        for idx, opt in enumerate(sample["options"]):
+            marker = "✓" if idx == sample["answer_index"] else " "
+            print(f"  [{marker}] {chr(65+idx)}) {opt}")
+            print()