Dataset Viewer
Auto-converted to Parquet Duplicate
id
stringlengths
18
43
task
stringclasses
18 values
level
int64
0
4
state
stringlengths
71
14.8k
questions
stringlengths
145
1.98k
answers
stringlengths
48
1.88k
arithmetic:train:0
arithmetic
0
{"opening_balance": 35, "currency": "£", "transactions": [{"day": 4, "type": "deposit", "amount": 30}, {"day": 12, "type": "withdrawal", "amount": 15}, {"day": 16, "type": "withdrawal", "amount": 30}]}
{"final_balance": {"type": "choice", "instructions": "What is the closing balance?", "criteria": {"£0": null, "£10": null, "£15": null, "£20": null, "£30": null}}, "lowest_day": {"type": "choice", "instructions": "After which day's transaction is the balance lowest?", "criteria": {"day 4": null, "day 12": null, "day 16...
{"final_balance": {"type": "choice", "choice": "£20", "confidence": 1.0, "probabilities": {"£0": 0.0, "£10": 0.0, "£15": 0.0, "£20": 1.0, "£30": 0.0}}, "lowest_day": {"type": "choice", "choice": "day 16", "confidence": 1.0, "probabilities": {"day 4": 0.0, "day 12": 0.0, "day 16": 1.0}}, "net_change": {"type": "score", ...
entity_belief_tracking:train:2
entity_belief_tracking
0
{"locations": ["shelf", "desk", "vault", "studio", "lab", "archive", "office", "basement", "attic", "drawer"], "initial_locations": {"item_1": "attic"}, "events": [{"step": 1, "object": "item_1", "destination": "archive", "witnesses": ["bob"]}, {"step": 2, "object": "item_1", "destination": "office", "witnesses": ["bob...
{"agent_belief_location": {"type": "choice", "instructions": "Where does alice believe item_1 is after all events?", "criteria": {"shelf": null, "desk": null, "vault": null, "studio": null, "lab": null, "archive": null, "office": null, "basement": null, "attic": null, "drawer": null}}, "belief_matches_world": {"type": ...
{"agent_belief_location": {"type": "choice", "choice": "attic", "confidence": 1.0, "probabilities": {"shelf": 0.0, "desk": 0.0, "vault": 0.0, "studio": 0.0, "lab": 0.0, "archive": 0.0, "office": 0.0, "basement": 0.0, "attic": 1.0, "drawer": 0.0}}, "belief_matches_world": {"type": "noul", "noul": 0.0}, "world_location":...
event_state_reconstruction:train:7
event_state_reconstruction
0
{"initial_state": {"owner": "bob", "open": true, "severity": "critical"}, "events": [{"ts": 101, "kind": "assign", "owner": "bob"}, {"ts": 109, "kind": "assign", "owner": "bob"}, {"ts": 111, "kind": "severity", "severity": "routine"}], "rule": "Start from initial_state and apply events in ascending timestamp order."}
{"current_owner": {"type": "choice", "instructions": "Who owns the incident after replaying the event log?", "criteria": {"alice": null, "bob": null, "carol": null, "unassigned": null}}, "current_severity": {"type": "score", "instructions": "What is the incident's current severity after replaying the event log?", "crit...
{"current_owner": {"type": "choice", "choice": "bob", "confidence": 1.0, "probabilities": {"alice": 0.0, "bob": 1.0, "carol": 0.0, "unassigned": 0.0}}, "current_severity": {"type": "score", "score": 0.0, "legend": {"0": "routine: Routine.", "1": "degraded: Degraded service requiring attention.", "2": "critical: Critica...
event_triggered_execution:train:3
event_triggered_execution
0
{"objects": [{"id": "O1", "state": false}, {"id": "O2", "state": true}], "rules": [{"id": "R1", "event_type": "check", "target": "any", "min_value": 0, "required_state": null, "set_state": false, "priority": 1, "cooldown": 0, "enabled": true}], "events": [{"ts": 1, "kind": "check", "target": "O1", "value": 6}, {"ts": 4...
{"object_ever_targeted": {"type": "noul", "instructions": "Does any rule fire on O1 during the event sequence?"}, "object_final_state": {"type": "noul", "instructions": "After processing all events, is O1.state true?"}, "rule_enabled_at_end": {"type": "noul", "instructions": "Is R1 enabled after processing all events?"...
{"object_ever_targeted": {"type": "noul", "noul": 1.0}, "object_final_state": {"type": "noul", "noul": 0.0}, "rule_enabled_at_end": {"type": "noul", "noul": 1.0}, "rule_on_signal": {"type": "choice", "choice": "R1", "confidence": 1.0, "probabilities": {"R1": 1.0, "none": 0.0}}, "total_firings": {"type": "score", "score...
evidence_sufficiency:train:1
evidence_sufficiency
0
{"claim": "The deployment is sufficiently supported as the cause of the incident.", "rules": ["Support is sufficient only with at least two independent valid supporting origins and no valid contradictory origin. Duplicates from one origin count once.", "Evidence with valid=false is invalid.", "Break reliability ties by...
{"claim_supported": {"type": "noul", "instructions": "Following the stated rules, is the claim sufficiently supported?"}, "has_conflict": {"type": "noul", "instructions": "Is there at least one valid supporting origin and at least one valid contradictory origin?"}, "strongest_support_origin": {"type": "choice", "instru...
{"claim_supported": {"type": "noul", "noul": 1.0}, "has_conflict": {"type": "noul", "noul": 0.0}, "strongest_support_origin": {"type": "choice", "choice": "S1", "confidence": 1.0, "probabilities": {"S1": 1.0, "S2": 0.0, "S3": 0.0, "none": 0.0}}}
goal_directed_planning:train:4
goal_directed_planning
0
{"steps": [{"id": "S23", "completed": false, "available": true, "requires_all": [], "requires_any": []}, {"id": "S6", "completed": false, "available": true, "requires_all": [], "requires_any": []}, {"id": "S74", "completed": false, "available": true, "requires_all": [], "requires_any": []}], "goal": ["S74"], "semantics...
{"first_legal_action": {"type": "choice", "instructions": "Which currently executable action has the smallest numeric step ID? Choose none if no action is executable.", "criteria": {"S23": null, "S6": null, "S74": null, "none": null}}, "minimal_actions": {"type": "score", "instructions": "How many newly executed action...
{"first_legal_action": {"type": "choice", "choice": "S6", "confidence": 1.0, "probabilities": {"S23": 0.0, "S6": 1.0, "S74": 0.0, "none": 0.0}}, "minimal_actions": {"type": "score", "score": 1.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8 or more", "9": "unreachab...
graph_navigation:train:0
graph_navigation
0
{"nodes": ["N09", "N16", "N42", "N57"], "edges": [{"id": "E5", "from": "N42", "to": "N16", "cost": 1, "open": true}, {"id": "E4", "from": "N57", "to": "N16", "cost": 1, "open": true}, {"id": "E2", "from": "N42", "to": "N57", "cost": 1, "open": true}, {"id": "E3", "from": "N16", "to": "N42", "cost": 1, "open": true}, {"...
{"edge_essential": {"type": "noul", "instructions": "Is edge E4 essential to achieving the minimum route cost?"}, "first_hop": {"type": "choice", "instructions": "What is the lexicographically smallest first hop among minimum-cost routes? Choose none if unreachable.", "criteria": {"N16": null, "N42": null, "N57": null,...
{"edge_essential": {"type": "noul", "noul": 0.0}, "first_hop": {"type": "choice", "choice": "N42", "confidence": 1.0, "probabilities": {"N16": 0.0, "N42": 1.0, "N57": 0.0, "none": 0.0}}, "reachable": {"type": "noul", "noul": 1.0}, "shortest_cost": {"type": "choice", "choice": "2", "confidence": 1.0, "probabilities": {"...
instruction_execution:train:3
instruction_execution
0
{"entity_type": "ticket", "property": "escalated", "entities": [{"id": "E1", "name": "ticket 1", "site": "west", "tier": "priority", "available": true, "escalated": false}, {"id": "E2", "name": "ticket 2", "site": "central", "tier": "standard", "available": true, "escalated": true}, {"id": "E3", "name": "ticket 3", "si...
{"changed_count": {"type": "score", "instructions": "How many entity properties change immediately (net changes)?", "criteria": ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9 or more"]}, "deferred_registration": {"type": "choice", "instructions": "What is registered for later, if anything?", "criteria": {"nothing": n...
{"changed_count": {"type": "score", "score": 0.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9 or more"}, "probabilities": {"0": 1.0, "1": 0.0, "2": 0.0, "3": 0.0, "4": 0.0, "5": 0.0, "6": 0.0, "7": 0.0, "8": 0.0, "9": 0.0}, "confidence": 1.0}, "deferred_r...
multi_view_adjudication:train:10
multi_view_adjudication
0
{"ticket": {"channel": "chat"}, "billing": {"duplicate_charge": false, "invoice_mismatch": false, "refund_missing": false}, "account": {"active": true, "auth_failures": 1, "permission_mismatch": false, "tier": "team"}, "telemetry": {"integration_failures": 0, "error_rate_percent": 10}, "timeline": {"deadline_hours": nu...
{"intent": {"type": "choice", "instructions": "Following the intent rule, what is the primary operational issue?", "criteria": {"billing": "Payments, invoices, refunds, or charges.", "access": "Authentication, permissions, or account access.", "technical": "Product failures, bugs, or integrations.", "other": "None of t...
{"intent": {"type": "choice", "choice": "other", "confidence": 1.0, "probabilities": {"billing": 0.0, "access": 0.0, "technical": 0.0, "other": 1.0}}, "is_urgent": {"type": "noul", "noul": 0.0}, "workflow_impact": {"type": "score", "score": 1.0, "legend": {"0": "No affected feature is down or slow.", "1": "Some affecte...
needle_retrieval:train:14
needle_retrieval
0
id,destination S-7528,Bergen S-7743,Perth S-7473,Oslo S-0282,Sfax S-6611,Accra S-4773,Oslo S-3477,Hue S-7437,Quito S-7475,Oslo S-0760,Lima
{"value_of_id": {"type": "choice", "instructions": "Which destination is listed for shipment S-7473?", "criteria": {"Oslo": null, "Quito": null, "Braga": null, "Perth": null, "Hue": null, "Accra": null}}}
{"value_of_id": {"type": "choice", "choice": "Oslo", "confidence": 1.0, "probabilities": {"Oslo": 1.0, "Quito": 0.0, "Braga": 0.0, "Perth": 0.0, "Hue": 0.0, "Accra": 0.0}}}
partial_observation_calibration:train:8
partial_observation_calibration
0
{"prior_probability_incident": "2/3", "assumption": "Sensor observations are conditionally independent given whether the incident is real.", "sensors": [{"id": "S1", "observed": "positive", "p_positive_given_incident": "2/3", "p_positive_given_no_incident": "1/3"}]}
{"incident_real": {"type": "noul", "instructions": "After conditioning on every sensor observation, is the incident more likely real than not?"}}
{"incident_real": {"type": "noul", "noul": 0.8}}
policy_applicability:train:4
policy_applicability
0
{"request": {"subject": {"role": "engineer", "team": "gamma", "clearance": 1}, "resource": {"team": "alpha", "sensitivity": "restricted"}, "action": "write"}, "policies": [{"id": "P1", "priority": 2, "effect": "allow", "min_clearance": 1}, {"id": "P2", "priority": 1, "effect": "deny", "roles": ["analyst", "engineer"]}]...
{"access_allowed": {"type": "noul", "instructions": "Does the governing policy allow the requested action?"}, "governing_policy": {"type": "choice", "instructions": "Which policy governs the request, i.e. has the highest priority among the policies whose constraints all hold?", "criteria": {"P1": null, "P2": null}}, "r...
{"access_allowed": {"type": "noul", "noul": 1.0}, "governing_policy": {"type": "choice", "choice": "P1", "confidence": 1.0, "probabilities": {"P1": 1.0, "P2": 0.0}}, "review_risk": {"type": "score", "score": 2.0, "legend": {"0": "The resource is not restricted and the matching policies agree on effect.", "1": "The reso...
policy_under_uncertainty:train:2
policy_under_uncertainty
0
{"request": {"subject": {"role": "unknown", "team": "alpha", "clearance": 2}, "resource": {"team": "beta", "sensitivity": "restricted"}, "action": "read"}, "role_evidence": {"history": "Of the last 61 requests from this account, 3 by analysts, 29 by engineers, 29 by managers. No other role has used it.", "reports": [{"...
{"access_allowed": {"type": "noul", "instructions": "Given the uncertainty about the requester's role, is the request more likely allowed than denied?"}, "governing_policy": {"type": "choice", "instructions": "Which policy most likely governs the request?", "criteria": {"P1": null, "P2": null, "none (default deny)": nu...
{"access_allowed": {"type": "noul", "noul": 0.0}, "governing_policy": {"type": "choice", "choice": "none (default deny)", "confidence": 1.0, "probabilities": {"P1": 0.0, "P2": 0.0, "none (default deny)": 1.0}}, "requester_role": {"type": "choice", "choice": "manager", "confidence": 1.0, "probabilities": {"analyst": 0.0...
record_aggregation:train:3
record_aggregation
0
item=orange tent; category=tools; quantity=5; in_stock=True item=purple rope; category=office; quantity=8; in_stock=True item=blue drill; category=office; quantity=17; in_stock=True item=green rope; category=garden; quantity=4; in_stock=True item=blue chair; category=kitchen; quantity=4; in_stock=True
{"any_out_of_stock": {"type": "noul", "instructions": "Is any kitchen item out of stock?"}, "count_in_category": {"type": "score", "instructions": "Count the kitchen items.", "criteria": ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9 or more"]}, "largest_quantity": {"type": "choice", "instructions": "Among the items,...
{"any_out_of_stock": {"type": "noul", "noul": 0.0}, "count_in_category": {"type": "score", "score": 1.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9 or more"}, "probabilities": {"0": 0.0, "1": 1.0, "2": 0.0, "3": 0.0, "4": 0.0, "5": 0.0, "6": 0.0, "7": 0....
record_eligibility:train:2
record_eligibility
0
{"user_request": "Select eligible workshop offers under all of the following conditions:\nC1. No mandatory deposit is permitted.", "records": [{"id": "R2", "description": "No deposit is required."}, {"id": "R1", "description": "No deposit is required."}, {"id": "R4", "description": "No information about the deposit pol...
{"decisive_constraint": {"type": "choice", "instructions": "For R3, which numbered requirement is decisive under the stated ordering rule?", "criteria": {"C1": null, "none": null}}, "eligible_count": {"type": "score", "instructions": "How many listed records are eligible?", "criteria": ["0", "1", "2", "3", "4", "5", "6...
{"decisive_constraint": {"type": "choice", "choice": "C1", "confidence": 1.0, "probabilities": {"C1": 1.0, "none": 0.0}}, "eligible_count": {"type": "score", "score": 2.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9 or more"}, "probabilities": {"0": 0.0, ...
state_perturbation:train:5
state_perturbation
0
{"before": [{"id": "R1", "authorized": false, "status": "resolved", "amount": 100, "owner": "alice", "note": "reviewed"}], "after": [{"id": "R1", "authorized": false, "status": "resolved", "amount": 100, "owner": "alice", "note": "imported"}], "materiality": "Material fields are authorized, status, amount, and owner; e...
{"changed_dimension": {"type": "choice", "instructions": "Which material dimension of record R1 changed? Choose none when only non-material fields changed.", "criteria": {"authorization": null, "status": null, "financial": null, "ownership": null, "none": null}}, "material_change": {"type": "noul", "instructions": "Did...
{"changed_dimension": {"type": "choice", "choice": "none", "confidence": 1.0, "probabilities": {"authorization": 0.0, "status": 0.0, "financial": 0.0, "ownership": 0.0, "none": 1.0}}, "material_change": {"type": "noul", "noul": 0.0}, "risk_direction": {"type": "score", "score": 1.0, "legend": {"0": "Lower operational r...
table_lookup:train:4
table_lookup
0
{"people": [{"name": "Sven", "team": "data", "city": "Cuenca", "start_year": 2010}, {"name": "Ada", "team": "billing", "city": "Sfax", "start_year": 2011}, {"name": "Ximena", "team": "support", "city": "Sfax", "start_year": 2017}, {"name": "Hana", "team": "security", "city": "Cuenca", "start_year": 2013}, {"name": "Hug...
{"count_matching": {"type": "score", "instructions": "How many people based in Osaka started in 2016 or later?", "criteria": ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9"]}, "find_person": {"type": "choice", "instructions": "Which person works in Sfax on the billing team?", "criteria": {"Ximena": null, "Sven": null...
{"count_matching": {"type": "score", "score": 0.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9"}, "probabilities": {"0": 1.0, "1": 0.0, "2": 0.0, "3": 0.0, "4": 0.0, "5": 0.0, "6": 0.0, "7": 0.0, "8": 0.0, "9": 0.0}, "confidence": 1.0}, "find_person": {"t...
taxonomy_routing:train:0
taxonomy_routing
0
Routing guide (a ticket goes to the category whose conditions it all meets; exactly one does): category=returns/review; rule=amount at most 200 category=hardware/specialists; rule=region is south category=billing/review; rule=channel is chat category=hardware/desk-a; rule=tier is plus category=onboarding/desk-b; rule=a...
{"belongs_to": {"type": "noul", "instructions": "Does this ticket belong in returns/review?"}, "conditions_met": {"type": "score", "instructions": "Count the conditions of returns/review that hold for this ticket.", "criteria": ["0", "1", "2", "3", "4"]}, "route": {"type": "choice", "instructions": "Which category shou...
{"belongs_to": {"type": "noul", "noul": 0.0}, "conditions_met": {"type": "score", "score": 0.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4"}, "probabilities": {"0": 1.0, "1": 0.0, "2": 0.0, "3": 0.0, "4": 0.0}, "confidence": 1.0}, "route": {"type": "choice", "choice": "hardware/specialists", "confidence...
arithmetic:train:4
arithmetic
1
Order lines (prices in £): item=scarf; unit_price=31; quantity=4 item=chair; unit_price=6; quantity=2 item=kettle; unit_price=31; quantity=3 item=mug; unit_price=15; quantity=3 Rule: Orders of £305 or more get £10 off; smaller orders pay £4 shipping.
{"amount_due": {"type": "choice", "instructions": "What is the final amount to pay?", "criteria": {"£185": null, "£233": null, "£264": null, "£268": null, "£278": null}}, "largest_line": {"type": "choice", "instructions": "Which item accounts for the largest share of the bill?", "criteria": {"scarf": null, "chair": nul...
{"amount_due": {"type": "choice", "choice": "£278", "confidence": 1.0, "probabilities": {"£185": 0.0, "£233": 0.0, "£264": 0.0, "£268": 0.0, "£278": 1.0}}, "largest_line": {"type": "choice", "choice": "scarf", "confidence": 1.0, "probabilities": {"scarf": 1.0, "chair": 0.0, "kettle": 0.0, "mug": 0.0}}, "lines_above": {...
entity_belief_tracking:train:1
entity_belief_tracking
1
{"locations": ["workshop", "mailroom", "cabinet", "garage", "attic", "lab", "desk", "studio", "kitchen", "locker", "drawer", "office", "basement", "archive"], "initial_locations": {"item_1": "archive", "item_2": "desk"}, "events": [{"step": 1, "object": "item_1", "destination": "attic", "witnesses": []}, {"step": 2, "o...
{"agent_belief_location": {"type": "choice", "instructions": "Where does carol believe item_1 is after all events?", "criteria": {"workshop": null, "mailroom": null, "cabinet": null, "garage": null, "attic": null, "lab": null, "desk": null, "studio": null, "kitchen": null, "locker": null, "drawer": null, "office": null...
{"agent_belief_location": {"type": "choice", "choice": "desk", "confidence": 1.0, "probabilities": {"workshop": 0.0, "mailroom": 0.0, "cabinet": 0.0, "garage": 0.0, "attic": 0.0, "lab": 0.0, "desk": 1.0, "studio": 0.0, "kitchen": 0.0, "locker": 0.0, "drawer": 0.0, "office": 0.0, "basement": 0.0, "archive": 0.0}}, "beli...
event_state_reconstruction:train:1
event_state_reconstruction
1
{"initial_state": {"owner": "bob", "open": true, "severity": "critical"}, "events": [{"ts": 107, "kind": "assign", "owner": "carol"}, {"ts": 115, "kind": "severity", "severity": "routine"}, {"ts": 120, "kind": "severity", "severity": "critical"}, {"ts": 122, "kind": "reopen"}], "rule": "Start from initial_state and app...
{"current_owner": {"type": "choice", "instructions": "Who owns the incident after replaying the event log?", "criteria": {"alice": null, "bob": null, "carol": null, "unassigned": null}}, "current_severity": {"type": "score", "instructions": "What is the incident's current severity after replaying the event log?", "crit...
{"current_owner": {"type": "choice", "choice": "carol", "confidence": 1.0, "probabilities": {"alice": 0.0, "bob": 0.0, "carol": 1.0, "unassigned": 0.0}}, "current_severity": {"type": "score", "score": 2.0, "legend": {"0": "routine: Routine.", "1": "degraded: Degraded service requiring attention.", "2": "critical: Criti...
event_triggered_execution:train:0
event_triggered_execution
1
{"objects": [{"id": "O1", "state": true}, {"id": "O2", "state": true}, {"id": "O3", "state": true}], "rules": [{"id": "R1", "event_type": "expire", "target": "any", "min_value": 0, "required_state": null, "set_state": false, "priority": 1, "cooldown": 0, "enabled": true}, {"id": "R2", "event_type": "arrive", "target": ...
{"object_ever_targeted": {"type": "noul", "instructions": "Does any rule fire on O3 during the event sequence?"}, "object_final_state": {"type": "noul", "instructions": "After processing all events, is O3.state true?"}, "rule_enabled_at_end": {"type": "noul", "instructions": "Is R2 enabled after processing all events?"...
{"object_ever_targeted": {"type": "noul", "noul": 1.0}, "object_final_state": {"type": "noul", "noul": 0.0}, "rule_enabled_at_end": {"type": "noul", "noul": 1.0}, "rule_on_signal": {"type": "choice", "choice": "none", "confidence": 1.0, "probabilities": {"R1": 0.0, "R2": 0.0, "none": 1.0}}, "total_firings": {"type": "s...
evidence_sufficiency:train:6
evidence_sufficiency
1
{"claim": "The deployment is sufficiently supported as the cause of the incident.", "rules": ["Support is sufficient only with at least two independent valid supporting origins and no valid contradictory origin. Duplicates from one origin count once.", "Evidence with valid=false is invalid.", "Break reliability ties by...
{"claim_supported": {"type": "noul", "instructions": "Following the stated rules, is the claim sufficiently supported?"}, "has_conflict": {"type": "noul", "instructions": "Is there at least one valid supporting origin and at least one valid contradictory origin?"}, "strongest_support_origin": {"type": "choice", "instru...
{"claim_supported": {"type": "noul", "noul": 0.0}, "has_conflict": {"type": "noul", "noul": 1.0}, "strongest_support_origin": {"type": "choice", "choice": "S1", "confidence": 1.0, "probabilities": {"S1": 1.0, "S2": 0.0, "S3": 0.0, "S4": 0.0, "none": 0.0}}}
goal_directed_planning:train:12
goal_directed_planning
1
{"steps": [{"id": "S79", "completed": false, "available": true, "requires_all": [], "requires_any": []}, {"id": "S6", "completed": false, "available": true, "requires_all": ["S79"], "requires_any": []}, {"id": "S50", "completed": false, "available": true, "requires_all": ["S6"], "requires_any": []}, {"id": "S83", "comp...
{"first_legal_action": {"type": "choice", "instructions": "Which currently executable action has the smallest numeric step ID? Choose none if no action is executable.", "criteria": {"S79": null, "S6": null, "S50": null, "S83": null, "S80": null, "none": null}}, "minimal_actions": {"type": "score", "instructions": "How ...
{"first_legal_action": {"type": "choice", "choice": "S79", "confidence": 1.0, "probabilities": {"S79": 1.0, "S6": 0.0, "S50": 0.0, "S83": 0.0, "S80": 0.0, "none": 0.0}}, "minimal_actions": {"type": "score", "score": 3.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8 ...
graph_navigation:train:1
graph_navigation
1
{"nodes": ["N08", "N17", "N30", "N44", "N59", "N66"], "edges": [{"id": "E10", "from": "N17", "to": "N08", "cost": 1, "open": true}, {"id": "E1", "from": "N08", "to": "N30", "cost": 1, "open": true}, {"id": "E6", "from": "N08", "to": "N66", "cost": 1, "open": true}, {"id": "E4", "from": "N08", "to": "N17", "cost": 1, "o...
{"edge_essential": {"type": "noul", "instructions": "Is edge E6 essential to achieving the minimum route cost?"}, "first_hop": {"type": "choice", "instructions": "What is the lexicographically smallest first hop among minimum-cost routes? Choose none if unreachable.", "criteria": {"N17": null, "N30": null, "N44": null,...
{"edge_essential": {"type": "noul", "noul": 1.0}, "first_hop": {"type": "choice", "choice": "N66", "confidence": 1.0, "probabilities": {"N17": 0.0, "N30": 0.0, "N44": 0.0, "N59": 0.0, "N66": 1.0, "none": 0.0}}, "reachable": {"type": "noul", "noul": 1.0}, "shortest_cost": {"type": "choice", "choice": "1", "confidence": ...
instruction_execution:train:1
instruction_execution
1
{"entity_type": "shipment", "property": "held", "entities": [{"id": "E1", "name": "shipment 1", "site": "north", "tier": "standard", "available": true, "held": true}, {"id": "E2", "name": "shipment 2", "site": "north", "tier": "standard", "available": true, "held": false}, {"id": "E3", "name": "shipment 3", "site": "ce...
{"changed_count": {"type": "score", "instructions": "How many entity properties change immediately (net changes)?", "criteria": ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9 or more"]}, "deferred_registration": {"type": "choice", "instructions": "What is registered for later, if anything?", "criteria": {"nothing": n...
{"changed_count": {"type": "score", "score": 1.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9 or more"}, "probabilities": {"0": 0.0, "1": 1.0, "2": 0.0, "3": 0.0, "4": 0.0, "5": 0.0, "6": 0.0, "7": 0.0, "8": 0.0, "9": 0.0}, "confidence": 1.0}, "deferred_r...
multi_view_adjudication:train:0
multi_view_adjudication
1
{"ticket": {"channel": "email"}, "billing": {"duplicate_charge": false, "invoice_mismatch": false, "refund_missing": false}, "account": {"active": true, "auth_failures": 1, "permission_mismatch": true, "tier": "enterprise"}, "telemetry": {"integration_failures": 1, "error_rate_percent": 35}, "timeline": {"deadline_hour...
{"intent": {"type": "choice", "instructions": "Following the intent rule, what is the primary operational issue?", "criteria": {"billing": "Payments, invoices, refunds, or charges.", "access": "Authentication, permissions, or account access.", "technical": "Product failures, bugs, or integrations.", "other": "None of t...
{"intent": {"type": "choice", "choice": "access", "confidence": 1.0, "probabilities": {"billing": 0.0, "access": 1.0, "technical": 0.0, "other": 0.0}}, "is_urgent": {"type": "noul", "noul": 1.0}, "workflow_impact": {"type": "score", "score": 1.0, "legend": {"0": "No affected feature is down or slow.", "1": "Some affect...
needle_retrieval:train:1
needle_retrieval
1
[{"id": "A-9063", "branch": "Oslo"}, {"id": "A-8506", "branch": "Dakar"}, {"id": "A-0808", "branch": "Lagos"}, {"id": "A-7088", "branch": "Hue"}, {"id": "A-9367", "branch": "Thies"}, {"id": "A-7963", "branch": "Quito"}, {"id": "A-0698", "branch": "Oslo"}, {"id": "A-3867", "branch": "Perth"}, {"id": "A-0152", "branch": ...
{"id_has_value": {"type": "noul", "instructions": "Does account A-3967 have branch Kobe?"}, "id_listed": {"type": "noul", "instructions": "Does the list include account A-3767?"}, "value_of_id": {"type": "choice", "instructions": "Look up account A-3967. Which branch does it have?", "criteria": {"Cusco": null, "Hue": n...
{"id_has_value": {"type": "noul", "noul": 1.0}, "id_listed": {"type": "noul", "noul": 1.0}, "value_of_id": {"type": "choice", "choice": "Kobe", "confidence": 1.0, "probabilities": {"Cusco": 0.0, "Hue": 0.0, "Lagos": 0.0, "Accra": 0.0, "Cuenca": 0.0, "Quito": 0.0, "Perth": 0.0, "Kobe": 1.0, "Sfax": 0.0, "Thies": 0.0, "T...
partial_observation_calibration:train:2
partial_observation_calibration
1
{"prior_probability_incident": "1/5", "assumption": "Sensor observations are conditionally independent given whether the incident is real.", "sensors": [{"id": "S1", "observed": "positive", "p_positive_given_incident": "3/4", "p_positive_given_no_incident": "1/4"}, {"id": "S2", "observed": "positive", "p_positive_given...
{"incident_real": {"type": "noul", "instructions": "After conditioning on every sensor observation, is the incident more likely real than not?"}}
{"incident_real": {"type": "noul", "noul": 0.692308}}
policy_applicability:train:8
policy_applicability
1
{"request": {"subject": {"role": "engineer", "team": "gamma", "clearance": 0}, "resource": {"team": "beta", "sensitivity": "public"}, "action": "read"}, "policies": [{"id": "P1", "priority": 3, "effect": "deny", "max_sensitivity": "public"}, {"id": "P2", "priority": 2, "effect": "allow", "teams": ["beta", "gamma"], "mi...
{"access_allowed": {"type": "noul", "instructions": "Does the governing policy allow the requested action?"}, "governing_policy": {"type": "choice", "instructions": "Which policy governs the request, i.e. has the highest priority among the policies whose constraints all hold?", "criteria": {"P1": null, "P2": null, "P3"...
{"access_allowed": {"type": "noul", "noul": 0.0}, "governing_policy": {"type": "choice", "choice": "P1", "confidence": 1.0, "probabilities": {"P1": 1.0, "P2": 0.0, "P3": 0.0}}, "review_risk": {"type": "score", "score": 2.0, "legend": {"0": "The resource is not restricted and the matching policies agree on effect.", "1"...
policy_under_uncertainty:train:4
policy_under_uncertainty
1
{"request": {"subject": {"role": "unknown", "team": "alpha", "clearance": 1}, "resource": {"team": "beta", "sensitivity": "internal"}, "action": "write"}, "role_evidence": {"history": "Of the last 65 requests from this account, 30 by analysts, 18 by engineers, 17 by managers. No other role has used it.", "reports": [{"...
{"access_allowed": {"type": "noul", "instructions": "Given the uncertainty about the requester's role, is the request more likely allowed than denied?"}, "governing_policy": {"type": "choice", "instructions": "Which policy most likely governs the request?", "criteria": {"P1": null, "P2": null, "P3": null, "none (defaul...
{"access_allowed": {"type": "noul", "noul": 0.872727}, "governing_policy": {"type": "choice", "choice": "P3", "confidence": 1.0, "probabilities": {"P1": 0.0, "P2": 0.0, "P3": 0.872727, "none (default deny)": 0.127273}}, "requester_role": {"type": "choice", "choice": "analyst", "confidence": 1.0, "probabilities": {"anal...
record_aggregation:train:4
record_aggregation
1
item | category | quantity | in_stock --- | --- | --- | --- black mug | tools | 14 | True orange scarf | garden | 10 | False red kettle | outdoor | 18 | True red tent | outdoor | 2 | False red vase | garden | 19 | True yellow scarf | garden | 3 | True black scarf | office | 2 | False yellow mug | garden | 18 | True pur...
{"any_out_of_stock": {"type": "noul", "instructions": "Is any office item out of stock?"}, "count_in_category": {"type": "score", "instructions": "Count the office items.", "criteria": ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9 or more"]}, "largest_quantity": {"type": "choice", "instructions": "Among the items, w...
{"any_out_of_stock": {"type": "noul", "noul": 1.0}, "count_in_category": {"type": "score", "score": 1.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9 or more"}, "probabilities": {"0": 0.0, "1": 1.0, "2": 0.0, "3": 0.0, "4": 0.0, "5": 0.0, "6": 0.0, "7": 0....
record_eligibility:train:0
record_eligibility
1
{"user_request": "Select eligible service proposals under all of the following conditions:\nC1. The provider must hold a current certification.\nC2. Exclude offers that require annual membership.", "records": [{"id": "R5", "description": "No information about certification is provided. Annual membership is not required...
{"decisive_constraint": {"type": "choice", "instructions": "For R2, which numbered requirement is decisive under the stated ordering rule?", "criteria": {"C1": null, "C2": null, "none": null}}, "eligible_count": {"type": "score", "instructions": "How many listed records are eligible?", "criteria": ["0", "1", "2", "3", ...
{"decisive_constraint": {"type": "choice", "choice": "none", "confidence": 1.0, "probabilities": {"C1": 0.0, "C2": 0.0, "none": 1.0}}, "eligible_count": {"type": "score", "score": 2.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9 or more"}, "probabilities"...
state_perturbation:train:0
state_perturbation
1
{"before": [{"id": "R1", "authorized": true, "status": "blocked", "amount": 250, "owner": "alice", "note": "imported"}, {"id": "R2", "authorized": false, "status": "blocked", "amount": 500, "owner": "alice", "note": "imported"}], "after": [{"id": "R1", "authorized": true, "status": "blocked", "amount": 250, "owner": "a...
{"changed_dimension": {"type": "choice", "instructions": "Which material dimension of record R2 changed? Choose none when only non-material fields changed.", "criteria": {"authorization": null, "status": null, "financial": null, "ownership": null, "none": null}}, "material_change": {"type": "noul", "instructions": "Did...
{"changed_dimension": {"type": "choice", "choice": "ownership", "confidence": 1.0, "probabilities": {"authorization": 0.0, "status": 0.0, "financial": 0.0, "ownership": 1.0, "none": 0.0}}, "material_change": {"type": "noul", "noul": 1.0}, "risk_direction": {"type": "score", "score": 2.0, "legend": {"0": "Lower operatio...
table_lookup:train:0
table_lookup
1
{"people": [{"name": "Gil", "team": "mobile", "city": "Quito", "start_year": 2021}, {"name": "Farah", "team": "mobile", "city": "Osaka", "start_year": 2020}, {"name": "Viktor", "team": "search", "city": "Osaka", "start_year": 2013}, {"name": "Quinn", "team": "security", "city": "Accra", "start_year": 2012}, {"name": "I...
{"count_matching": {"type": "score", "instructions": "How many people based in Perth started in 2022 or later?", "criteria": ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9"]}, "find_person": {"type": "choice", "instructions": "Which person works in Quito on the mobile team?", "criteria": {"Jonas": null, "Viktor": nul...
{"count_matching": {"type": "score", "score": 0.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9"}, "probabilities": {"0": 1.0, "1": 0.0, "2": 0.0, "3": 0.0, "4": 0.0, "5": 0.0, "6": 0.0, "7": 0.0, "8": 0.0, "9": 0.0}, "confidence": 1.0}, "find_person": {"t...
taxonomy_routing:train:19
taxonomy_routing
1
Routing guide (a ticket goes to the category whose conditions it all meets; exactly one does): [{"category": "security/desk-b", "rule": "tier is plus and amount at most 100"}, {"category": "billing/specialists", "rule": "tier is plus and channel is chat"}, {"category": "onboarding/escalations", "rule": "tier is busines...
{"belongs_to": {"type": "noul", "instructions": "Should the ticket be routed to billing/desk-b?"}, "conditions_met": {"type": "score", "instructions": "How many of the conditions listed for billing/desk-b does the ticket meet?", "criteria": ["0", "1", "2", "3", "4"]}, "route": {"type": "choice", "instructions": "Where ...
{"belongs_to": {"type": "noul", "noul": 1.0}, "conditions_met": {"type": "score", "score": 2.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4"}, "probabilities": {"0": 0.0, "1": 0.0, "2": 1.0, "3": 0.0, "4": 0.0}, "confidence": 1.0}, "route": {"type": "choice", "choice": "billing/desk-b", "confidence": 1.0...
arithmetic:train:8
arithmetic
2
Opening balance: £335 day=5; type=withdrawal; amount=125 day=9; type=withdrawal; amount=10 day=14; type=withdrawal; amount=45 day=15; type=withdrawal; amount=105 day=16; type=withdrawal; amount=35 day=21; type=deposit; amount=115
{"final_balance": {"type": "choice", "instructions": "What is the closing balance?", "criteria": {"£130": null, "£175": null, "£140": null, "£540": null, "£220": null}}, "lowest_day": {"type": "choice", "instructions": "After which day's transaction is the balance lowest?", "criteria": {"day 5": null, "day 9": null, "d...
{"final_balance": {"type": "choice", "choice": "£130", "confidence": 1.0, "probabilities": {"£130": 1.0, "£175": 0.0, "£140": 0.0, "£540": 0.0, "£220": 0.0}}, "lowest_day": {"type": "choice", "choice": "day 16", "confidence": 1.0, "probabilities": {"day 5": 0.0, "day 9": 0.0, "day 14": 0.0, "day 15": 0.0, "day 16": 1.0...
entity_belief_tracking:train:7
entity_belief_tracking
2
{"locations": ["basement", "archive", "mailroom", "drawer", "shelf", "attic"], "initial_locations": {"item_1": "mailroom", "item_2": "archive"}, "events": [{"step": 1, "object": "item_1", "destination": "drawer", "witnesses": ["alice"]}, {"step": 2, "object": "item_1", "destination": "archive", "witnesses": []}, {"step...
{"agent_belief_location": {"type": "choice", "instructions": "Where does alice believe item_1 is after all events?", "criteria": {"basement": null, "archive": null, "mailroom": null, "drawer": null, "shelf": null, "attic": null}}, "belief_matches_world": {"type": "noul", "instructions": "After all events, does alice be...
{"agent_belief_location": {"type": "choice", "choice": "attic", "confidence": 1.0, "probabilities": {"basement": 0.0, "archive": 0.0, "mailroom": 0.0, "drawer": 0.0, "shelf": 0.0, "attic": 1.0}}, "belief_matches_world": {"type": "noul", "noul": 0.0}, "nested_belief_location": {"type": "choice", "choice": "attic", "conf...
event_state_reconstruction:train:2
event_state_reconstruction
2
{"initial_state": {"owner": "alice", "open": true, "severity": "degraded"}, "events": [{"ts": 122, "kind": "assign", "owner": "bob"}, {"ts": 121, "kind": "resolve"}, {"ts": 102, "kind": "reopen"}, {"ts": 112, "kind": "resolve"}, {"ts": 114, "kind": "severity", "severity": "routine"}, {"ts": 104, "kind": "resolve"}, {"t...
{"current_owner": {"type": "choice", "instructions": "Who owns the incident after replaying the event log?", "criteria": {"alice": null, "bob": null, "carol": null, "unassigned": null}}, "current_severity": {"type": "score", "instructions": "What is the incident's current severity after replaying the event log?", "crit...
{"current_owner": {"type": "choice", "choice": "bob", "confidence": 1.0, "probabilities": {"alice": 0.0, "bob": 1.0, "carol": 0.0, "unassigned": 0.0}}, "current_severity": {"type": "score", "score": 0.0, "legend": {"0": "routine: Routine.", "1": "degraded: Degraded service requiring attention.", "2": "critical: Critica...
event_triggered_execution:train:5
event_triggered_execution
2
{"objects": [{"id": "O1", "state": true}, {"id": "O2", "state": true}, {"id": "O3", "state": true}, {"id": "O4", "state": true}], "rules": [{"id": "R1", "event_type": "check", "target": "any", "min_value": 1, "required_state": null, "set_state": false, "priority": 2, "cooldown": 0, "enabled": true}, {"id": "R2", "event...
{"rule_on_signal": {"type": "choice", "instructions": "Which rule fires for the signal at timestamp 5?", "criteria": {"R1": null, "R2": null, "R3": null, "none": null}}, "total_firings": {"type": "score", "instructions": "How many rule firings occur in total (including no-op writes)?", "criteria": ["0", "1", "2", "3", ...
{"rule_on_signal": {"type": "choice", "choice": "R2", "confidence": 1.0, "probabilities": {"R1": 0.0, "R2": 1.0, "R3": 0.0, "none": 0.0}}, "total_firings": {"type": "score", "score": 7.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9 or more"}, "probabiliti...
evidence_sufficiency:train:4
evidence_sufficiency
2
{"claim": "The deployment is sufficiently supported as the cause of the incident.", "rules": ["Support is sufficient only with at least two independent valid supporting origins and no valid contradictory origin. Duplicates from one origin count once.", "Evidence with valid=false is invalid.", "A retraction makes the ev...
{"claim_supported": {"type": "noul", "instructions": "Following the stated rules, is the claim sufficiently supported?"}, "has_conflict": {"type": "noul", "instructions": "Is there at least one valid supporting origin and at least one valid contradictory origin?"}, "strongest_support_origin": {"type": "choice", "instru...
{"claim_supported": {"type": "noul", "noul": 0.0}, "has_conflict": {"type": "noul", "noul": 1.0}, "strongest_support_origin": {"type": "choice", "choice": "S3", "confidence": 1.0, "probabilities": {"S1": 0.0, "S2": 0.0, "S3": 1.0, "S4": 0.0, "S5": 0.0, "none": 0.0}}}
goal_directed_planning:train:0
goal_directed_planning
2
{"steps": [{"id": "S88", "completed": false, "available": true, "requires_all": [], "requires_any": []}, {"id": "S67", "completed": false, "available": true, "requires_all": ["S88"], "requires_any": []}, {"id": "S76", "completed": false, "available": true, "requires_all": ["S67", "S88"], "requires_any": []}, {"id": "S8...
{"first_legal_action": {"type": "choice", "instructions": "Which currently executable action has the smallest numeric step ID? Choose none if no action is executable.", "criteria": {"S88": null, "S67": null, "S76": null, "S89": null, "S33": null, "S49": null, "S26": null, "none": null}}, "goal_reachable": {"type": "nou...
{"first_legal_action": {"type": "choice", "choice": "S88", "confidence": 1.0, "probabilities": {"S88": 1.0, "S67": 0.0, "S76": 0.0, "S89": 0.0, "S33": 0.0, "S49": 0.0, "S26": 0.0, "none": 0.0}}, "goal_reachable": {"type": "noul", "noul": 1.0}, "minimal_actions": {"type": "score", "score": 3.0, "legend": {"0": "0", "1":...
graph_navigation:train:11
graph_navigation
2
{"nodes": ["N16", "N20", "N27", "N28", "N47", "N49", "N64", "N70"], "edges": [{"id": "E6", "from": "N64", "to": "N20", "cost": 2, "open": true}, {"id": "E12", "from": "N70", "to": "N64", "cost": 1, "open": true}, {"id": "E8", "from": "N27", "to": "N64", "cost": 3, "open": true}, {"id": "E11", "from": "N27", "to": "N28"...
{"first_hop": {"type": "choice", "instructions": "What is the lexicographically smallest first hop among minimum-cost routes? Choose none if unreachable.", "criteria": {"N20": null, "N27": null, "N28": null, "N47": null, "N49": null, "N64": null, "N70": null, "none": null}}, "shortest_cost": {"type": "choice", "instruc...
{"first_hop": {"type": "choice", "choice": "N70", "confidence": 1.0, "probabilities": {"N20": 0.0, "N27": 0.0, "N28": 0.0, "N47": 0.0, "N49": 0.0, "N64": 0.0, "N70": 1.0, "none": 0.0}}, "shortest_cost": {"type": "choice", "choice": "2", "confidence": 1.0, "probabilities": {"4": 0.0, "0": 0.0, "3": 0.0, "7": 0.0, "1": 0...
instruction_execution:train:0
instruction_execution
2
{"entity_type": "ticket", "property": "escalated", "entities": [{"id": "E1", "name": "ticket 1", "site": "central", "tier": "priority", "available": true, "escalated": false}, {"id": "E2", "name": "ticket 2", "site": "south", "tier": "standard", "available": true, "escalated": true}, {"id": "E3", "name": "ticket 3", "s...
{"changed_count": {"type": "score", "instructions": "How many entity properties change immediately (net changes)?", "criteria": ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9 or more"]}, "deferred_registration": {"type": "choice", "instructions": "What is registered for later, if anything?", "criteria": {"nothing": n...
{"changed_count": {"type": "score", "score": 0.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9 or more"}, "probabilities": {"0": 1.0, "1": 0.0, "2": 0.0, "3": 0.0, "4": 0.0, "5": 0.0, "6": 0.0, "7": 0.0, "8": 0.0, "9": 0.0}, "confidence": 1.0}, "deferred_r...
multi_view_adjudication:train:1
multi_view_adjudication
2
{"ticket": {"channel": "email"}, "billing": {"duplicate_charge": false, "invoice_mismatch": false, "refund_missing": false}, "account": {"active": true, "auth_failures": 1, "permission_mismatch": false, "tier": "team"}, "telemetry": {"integration_failures": 2, "error_rate_percent": 19}, "timeline": {"deadline_hours": 2...
{"intent": {"type": "choice", "instructions": "Following the intent rule, what is the primary operational issue?", "criteria": {"billing": "Payments, invoices, refunds, or charges.", "access": "Authentication, permissions, or account access.", "technical": "Product failures, bugs, or integrations.", "other": "None of t...
{"intent": {"type": "choice", "choice": "technical", "confidence": 1.0, "probabilities": {"billing": 0.0, "access": 0.0, "technical": 1.0, "other": 0.0}}, "is_urgent": {"type": "noul", "noul": 1.0}, "workflow_impact": {"type": "score", "score": 0.0, "legend": {"0": "No affected feature is down or slow.", "1": "Some aff...
needle_retrieval:train:0
needle_retrieval
2
The account A-8166 has branch Bergen. The account A-4057 has branch Perth. The account A-1716 has branch Bergen. The account A-2402 has branch Osaka. The account A-2329 has branch Thies. The account A-9515 has branch Cusco. The account A-5903 has branch Cuenca. The account A-9549 has branch Darwin. The account A-2776 h...
{"id_has_value": {"type": "noul", "instructions": "Does account A-3457 have branch Hue?"}, "id_listed": {"type": "noul", "instructions": "Does the list include account A-9357?"}, "value_of_id": {"type": "choice", "instructions": "Which branch is listed for account A-3457?", "criteria": {"Sfax": null, "Cusco": null, "Da...
{"id_has_value": {"type": "noul", "noul": 1.0}, "id_listed": {"type": "noul", "noul": 1.0}, "value_of_id": {"type": "choice", "choice": "Hue", "confidence": 1.0, "probabilities": {"Sfax": 0.0, "Cusco": 0.0, "Dakar": 0.0, "Hanoi": 0.0, "Cuenca": 0.0, "Tunis": 0.0, "Osaka": 0.0, "Bergen": 0.0, "Braga": 0.0, "Kobe": 0.0, ...
partial_observation_calibration:train:3
partial_observation_calibration
2
{"prior_probability_incident": "1/3", "assumption": "Sensor observations are conditionally independent given whether the incident is real.", "sensors": [{"id": "S1", "observed": "negative", "p_positive_given_incident": "3/4", "p_positive_given_no_incident": "1/4"}, {"id": "S2", "observed": "negative", "p_positive_given...
{"incident_real": {"type": "noul", "instructions": "After conditioning on every sensor observation, is the incident more likely real than not?"}}
{"incident_real": {"type": "noul", "noul": 0.076923}}
policy_applicability:train:3
policy_applicability
2
{"request": {"subject": {"role": "engineer", "team": "alpha", "clearance": 2}, "resource": {"team": "gamma", "sensitivity": "restricted"}, "action": "approve"}, "policies": [{"id": "P3", "priority": 3, "effect": "deny", "teams": ["alpha", "beta", "gamma"], "actions": ["approve"]}, {"id": "P2", "priority": 4, "effect": ...
{"access_allowed": {"type": "noul", "instructions": "Does the governing policy allow the requested action?"}, "governing_policy": {"type": "choice", "instructions": "Which policy governs the request, i.e. has the highest priority among the policies whose constraints all hold?", "criteria": {"P1": null, "P2": null, "P3"...
{"access_allowed": {"type": "noul", "noul": 0.0}, "governing_policy": {"type": "choice", "choice": "P3", "confidence": 1.0, "probabilities": {"P1": 0.0, "P2": 0.0, "P3": 1.0, "P4": 0.0, "P5": 0.0}}, "review_risk": {"type": "score", "score": 2.0, "legend": {"0": "The resource is not restricted and the matching policies ...
policy_under_uncertainty:train:1
policy_under_uncertainty
2
{"request": {"subject": {"role": "unknown", "team": "alpha", "clearance": 0}, "resource": {"team": "gamma", "sensitivity": "internal"}, "action": "read"}, "role_evidence": {"history": "Of the last 37 requests from this account, 5 by analysts, 7 by engineers, 25 by managers. No other role has used it.", "reports": [{"so...
{"access_allowed": {"type": "noul", "instructions": "Given the uncertainty about the requester's role, is the request more likely allowed than denied?"}, "governing_policy": {"type": "choice", "instructions": "Which policy most likely governs the request?", "criteria": {"P1": null, "P2": null, "P3": null, "P4": null, "...
{"access_allowed": {"type": "noul", "noul": 0.997419}, "governing_policy": {"type": "choice", "choice": "P1", "confidence": 1.0, "probabilities": {"P1": 0.997419, "P2": 0.0, "P3": 0.0, "P4": 0.002581, "none (default deny)": 0.0}}, "requester_role": {"type": "choice", "choice": "manager", "confidence": 1.0, "probabiliti...
record_aggregation:train:1
record_aggregation
2
item,category,quantity,in_stock red rope,outdoor,4,False gray scarf,outdoor,4,True orange rope,outdoor,1,False purple scarf,kitchen,11,True gray mug,tools,20,False purple vase,tools,1,False yellow vase,office,2,True brown scarf,garden,6,False white scarf,tools,15,True blue mug,outdoor,9,True brown kettle,tools,20,True ...
{"any_out_of_stock": {"type": "noul", "instructions": "Is any office item out of stock?"}, "count_filtered": {"type": "score", "instructions": "How many office items have a quantity of at least 8?", "criteria": ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9 or more"]}, "count_in_category": {"type": "score", "instruct...
{"any_out_of_stock": {"type": "noul", "noul": 1.0}, "count_filtered": {"type": "score", "score": 3.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9 or more"}, "probabilities": {"0": 0.0, "1": 0.0, "2": 0.0, "3": 1.0, "4": 0.0, "5": 0.0, "6": 0.0, "7": 0.0, ...
record_eligibility:train:17
record_eligibility
2
{"user_request": "Select eligible project bids under all of the following conditions:\nC1. No mandatory deposit is permitted.\nC2. The combined base price and additional charges must be at most €250.\nC3. Support must include either a dedicated help desk, or live chat with extended hours.", "records": [{"id": "R4", "de...
{"decisive_constraint": {"type": "choice", "instructions": "For R5, which numbered requirement is decisive under the stated ordering rule?", "criteria": {"C1": null, "C2": null, "C3": null, "none": null}}, "eligible_count": {"type": "score", "instructions": "How many listed records are eligible?", "criteria": ["0", "1"...
{"decisive_constraint": {"type": "choice", "choice": "none", "confidence": 1.0, "probabilities": {"C1": 0.0, "C2": 0.0, "C3": 0.0, "none": 1.0}}, "eligible_count": {"type": "score", "score": 8.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9 or more"}, "pro...
state_perturbation:train:1
state_perturbation
2
{"before": [{"id": "R1", "authorized": true, "status": "open", "amount": 510, "owner": "alice", "note": "reviewed", "status_note": "open", "amount_quoted": 530, "previous_owner": "bob"}, {"id": "R2", "authorized": true, "status": "open", "amount": 490, "owner": "alice", "note": "routine", "status_note": "open", "amount...
{"changed_dimension": {"type": "choice", "instructions": "Which material dimension of record R3 changed? Choose none when only non-material fields changed.", "criteria": {"authorization": null, "status": null, "financial": null, "ownership": null, "none": null}}, "material_change": {"type": "noul", "instructions": "Did...
{"changed_dimension": {"type": "choice", "choice": "none", "confidence": 1.0, "probabilities": {"authorization": 0.0, "status": 0.0, "financial": 0.0, "ownership": 0.0, "none": 1.0}}, "material_change": {"type": "noul", "noul": 0.0}, "risk_direction": {"type": "score", "score": 1.0, "legend": {"0": "Lower operational r...
table_lookup:train:1
table_lookup
2
people: name,team,city,start_year Nora,mobile,Oslo,2013 Gil,data,Dakar,2016 Malik,support,Braga,2021 Rosa,data,Braga,2008 Hana,support,Oslo,2009 Viktor,security,Oslo,2012 Yusuf,support,Cusco,2015 Dmitri,support,Oslo,2011 Kenji,billing,Dakar,2015 Zoe,mobile,Braga,2012 Jonas,security,Braga,2019 teams: team,manager billi...
{"count_matching": {"type": "score", "instructions": "How many people based in Braga started in 2011 or later?", "criteria": ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9"]}, "find_person": {"type": "choice", "instructions": "Which person works in Braga on the team managed by Jae?", "criteria": {"Gil": null, "Malik"...
{"count_matching": {"type": "score", "score": 3.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9"}, "probabilities": {"0": 0.0, "1": 0.0, "2": 0.0, "3": 1.0, "4": 0.0, "5": 0.0, "6": 0.0, "7": 0.0, "8": 0.0, "9": 0.0}, "confidence": 1.0}, "find_person": {"t...
taxonomy_routing:train:4
taxonomy_routing
2
Routing guide (a ticket goes to the category whose conditions it all meets; exactly one does): category=partners/priority; rule=amount above 100 and product is camera category=accounts/specialists; rule=amount above 100 and product is laptop and age days above 14 category=security/intake; rule=amount above 100 and tier...
{"belongs_to": {"type": "noul", "instructions": "Should the ticket be routed to partners/priority?"}, "conditions_met": {"type": "score", "instructions": "How many of the conditions listed for partners/priority does the ticket meet?", "criteria": ["0", "1", "2", "3", "4"]}, "route": {"type": "choice", "instructions": "...
{"belongs_to": {"type": "noul", "noul": 0.0}, "conditions_met": {"type": "score", "score": 1.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4"}, "probabilities": {"0": 0.0, "1": 1.0, "2": 0.0, "3": 0.0, "4": 0.0}, "confidence": 1.0}, "route": {"type": "choice", "choice": "returns/specialists", "confidence"...
arithmetic:train:11
arithmetic
3
Order lines (prices in £): item,unit_price,quantity clock,10,3 rope,63,1 vase,20,3 kettle,10,3 mug,37,5 tent,25,1 Rule: Orders of £405 or more get £15 off; smaller orders pay £4 shipping.
{"amount_due": {"type": "choice", "instructions": "What is the final amount to pay?", "criteria": {"£249": null, "£372": null, "£378": null, "£387": null, "£397": null}}, "largest_line": {"type": "choice", "instructions": "Which line costs the most in total?", "criteria": {"clock": null, "rope": null, "vase": null, "ke...
{"amount_due": {"type": "choice", "choice": "£397", "confidence": 1.0, "probabilities": {"£249": 0.0, "£372": 0.0, "£378": 0.0, "£387": 0.0, "£397": 1.0}}, "largest_line": {"type": "choice", "choice": "mug", "confidence": 1.0, "probabilities": {"clock": 0.0, "rope": 0.0, "vase": 0.0, "kettle": 0.0, "mug": 1.0, "tent": ...
entity_belief_tracking:train:0
entity_belief_tracking
3
{"locations": ["vault", "workshop", "studio", "lab", "desk", "attic", "basement", "cabinet", "garage", "shelf", "locker", "archive", "office"], "initial_locations": {"item_1": "basement", "item_2": "studio", "item_3": "shelf"}, "events": [{"step": 1, "object": "item_1", "destination": "cabinet", "witnesses": []}, {"ste...
{"agent_belief_location": {"type": "choice", "instructions": "Where does alice believe item_3 is after all events?", "criteria": {"vault": null, "workshop": null, "studio": null, "lab": null, "desk": null, "attic": null, "basement": null, "cabinet": null, "garage": null, "shelf": null, "locker": null, "archive": null, ...
{"agent_belief_location": {"type": "choice", "choice": "studio", "confidence": 1.0, "probabilities": {"vault": 0.0, "workshop": 0.0, "studio": 1.0, "lab": 0.0, "desk": 0.0, "attic": 0.0, "basement": 0.0, "cabinet": 0.0, "garage": 0.0, "shelf": 0.0, "locker": 0.0, "archive": 0.0, "office": 0.0}}, "belief_matches_world":...
event_state_reconstruction:train:22
event_state_reconstruction
3
{"initial_state": {"owner": "bob", "open": true, "severity": "critical"}, "events": [{"ts": 117, "kind": "resolve"}, {"ts": 125, "kind": "reopen"}, {"ts": 151, "kind": "note", "text": "triage"}, {"ts": 143, "kind": "void", "voids_ts": 141}, {"ts": 156, "kind": "note", "text": "customer update"}, {"ts": 130, "kind": "as...
{"current_owner": {"type": "choice", "instructions": "Who owns the incident after replaying the event log?", "criteria": {"alice": null, "bob": null, "carol": null, "unassigned": null}}, "current_severity": {"type": "score", "instructions": "What is the incident's current severity after replaying the event log?", "crit...
{"current_owner": {"type": "choice", "choice": "carol", "confidence": 1.0, "probabilities": {"alice": 0.0, "bob": 0.0, "carol": 1.0, "unassigned": 0.0}}, "current_severity": {"type": "score", "score": 0.0, "legend": {"0": "routine: Routine.", "1": "degraded: Degraded service requiring attention.", "2": "critical: Criti...
event_triggered_execution:train:1
event_triggered_execution
3
{"objects": [{"id": "O1", "state": false}, {"id": "O2", "state": true}, {"id": "O3", "state": true}, {"id": "O4", "state": true}, {"id": "O5", "state": true}, {"id": "O6", "state": true}], "rules": [{"id": "R1", "event_type": "expire", "target": "any", "min_value": 0, "required_state": null, "set_state": true, "priorit...
{"object_ever_targeted": {"type": "noul", "instructions": "Does any rule fire on O6 during the event sequence?"}, "object_final_state": {"type": "noul", "instructions": "After processing all events, is O6.state true?"}, "rule_enabled_at_end": {"type": "noul", "instructions": "Is R1 enabled after processing all events?"...
{"object_ever_targeted": {"type": "noul", "noul": 0.0}, "object_final_state": {"type": "noul", "noul": 1.0}, "rule_enabled_at_end": {"type": "noul", "noul": 1.0}, "rule_on_signal": {"type": "choice", "choice": "R1", "confidence": 1.0, "probabilities": {"R1": 1.0, "R2": 0.0, "R3": 0.0, "R4": 0.0, "R5": 0.0, "none": 0.0}...
evidence_sufficiency:train:0
evidence_sufficiency
3
{"claim": "The deployment is sufficiently supported as the cause of the incident.", "rules": ["Support is sufficient only with at least two independent valid supporting origins and no valid contradictory origin. Duplicates from one origin count once.", "Evidence with valid=false is invalid.", "A retraction makes the ev...
{"claim_supported": {"type": "noul", "instructions": "Following the stated rules, is the claim sufficiently supported?"}, "has_conflict": {"type": "noul", "instructions": "Is there at least one valid supporting origin and at least one valid contradictory origin?"}, "strongest_support_origin": {"type": "choice", "instru...
{"claim_supported": {"type": "noul", "noul": 1.0}, "has_conflict": {"type": "noul", "noul": 0.0}, "strongest_support_origin": {"type": "choice", "choice": "S2", "confidence": 1.0, "probabilities": {"S1": 0.0, "S2": 1.0, "S3": 0.0, "S4": 0.0, "S5": 0.0, "S6": 0.0, "S7": 0.0, "none": 0.0}}}
goal_directed_planning:train:3
goal_directed_planning
3
{"steps": [{"id": "S60", "completed": false, "available": true, "requires_all": [], "requires_any": []}, {"id": "S26", "completed": false, "available": false, "requires_all": ["S60"], "requires_any": []}, {"id": "S82", "completed": false, "available": true, "requires_all": ["S60", "S26"], "requires_any": []}, {"id": "S...
{"first_legal_action": {"type": "choice", "instructions": "Which currently executable action has the smallest numeric step ID? Choose none if no action is executable.", "criteria": {"S60": null, "S26": null, "S82": null, "S68": null, "S73": null, "S51": null, "S20": null, "S31": null, "S89": null, "none": null}}, "goal...
{"first_legal_action": {"type": "choice", "choice": "S60", "confidence": 1.0, "probabilities": {"S60": 1.0, "S26": 0.0, "S82": 0.0, "S68": 0.0, "S73": 0.0, "S51": 0.0, "S20": 0.0, "S31": 0.0, "S89": 0.0, "none": 0.0}}, "goal_reachable": {"type": "noul", "noul": 0.0}, "minimal_actions": {"type": "score", "score": 9.0, "...
graph_navigation:train:3
graph_navigation
3
{"nodes": ["N06", "N10", "N23", "N26", "N33", "N35", "N46", "N50", "N54", "N55", "N62"], "edges": [{"id": "E2", "from": "N55", "to": "N35", "cost": 2, "open": false}, {"id": "E20", "from": "N46", "to": "N26", "cost": 2, "open": false}, {"id": "E11", "from": "N26", "to": "N06", "cost": 5, "open": true}, {"id": "E16", "f...
{"edge_essential": {"type": "noul", "instructions": "Is edge E7 essential to achieving the minimum route cost?"}, "first_hop": {"type": "choice", "instructions": "What is the lexicographically smallest first hop among minimum-cost routes? Choose none if unreachable.", "criteria": {"N10": null, "N23": null, "N26": null,...
{"edge_essential": {"type": "noul", "noul": 0.0}, "first_hop": {"type": "choice", "choice": "none", "confidence": 1.0, "probabilities": {"N10": 0.0, "N23": 0.0, "N26": 0.0, "N33": 0.0, "N35": 0.0, "N46": 0.0, "N50": 0.0, "N54": 0.0, "N55": 0.0, "N62": 0.0, "none": 1.0}}, "reachable": {"type": "noul", "noul": 0.0}, "sho...
instruction_execution:train:2
instruction_execution
3
{"entity_type": "job", "property": "running", "entities": [{"id": "E1", "name": "job 1", "site": "north", "tier": "priority", "available": true, "running": false}, {"id": "E2", "name": "job 2", "site": "north", "tier": "critical", "available": true, "running": false}, {"id": "E3", "name": "job 3", "site": "central", "t...
{"changed_count": {"type": "score", "instructions": "How many entity properties change immediately (net changes)?", "criteria": ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9 or more"]}, "deferred_registration": {"type": "choice", "instructions": "What is registered for later, if anything?", "criteria": {"nothing": n...
{"changed_count": {"type": "score", "score": 0.0, "legend": {"0": "0", "1": "1", "2": "2", "3": "3", "4": "4", "5": "5", "6": "6", "7": "7", "8": "8", "9": "9 or more"}, "probabilities": {"0": 1.0, "1": 0.0, "2": 0.0, "3": 0.0, "4": 0.0, "5": 0.0, "6": 0.0, "7": 0.0, "8": 0.0, "9": 0.0}, "confidence": 1.0}, "deferred_r...
multi_view_adjudication:train:7
multi_view_adjudication
3
{"ticket": {"channel": "api"}, "billing": {"duplicate_charge": false, "invoice_mismatch": false, "refund_missing": false}, "account": {"active": true, "permission_mismatch": false, "tier": "free"}, "telemetry": {"integration_failures": 1, "error_rate_percent": 22}, "timeline": {"deadline_hours": 23, "executive_escalati...
{"intent": {"type": "choice", "instructions": "Following the intent rule, what is the primary operational issue?", "criteria": {"billing": "Payments, invoices, refunds, or charges.", "access": "Authentication, permissions, or account access.", "technical": "Product failures, bugs, or integrations.", "other": "None of t...
{"intent": {"type": "choice", "choice": "technical", "confidence": 1.0, "probabilities": {"billing": 0.0, "access": 0.0, "technical": 1.0, "other": 0.0}}, "workflow_impact": {"type": "score", "score": 2.0, "legend": {"0": "No affected feature is down or slow.", "1": "Some affected feature is down or slow, but no core f...
needle_retrieval:train:8
needle_retrieval
3
id=B-2947; office=Cuenca id=B-3985; office=Accra id=B-2985; office=Cusco id=B-5099; office=Porto id=B-9884; office=Tunis id=B-1000; office=Quito id=B-0985; office=Accra id=B-8965; office=Cuenca id=B-4489; office=Dakar id=B-2679; office=Sfax id=B-9570; office=Cuenca id=B-9453; office=Braga id=B-1003; office=Thies id=B-4...
{"value_of_id": {"type": "choice", "instructions": "Which office is listed for badge B-8648?", "criteria": {"Lima": null, "Kobe": null, "Cuenca": null, "Sfax": null, "Darwin": null, "Accra": null, "Lagos": null, "Osaka": null, "Porto": null, "Braga": null, "Dakar": null, "Oslo": null, "Bergen": null, "Hanoi": null, "Hu...
{"value_of_id": {"type": "choice", "choice": "Porto", "confidence": 1.0, "probabilities": {"Lima": 0.0, "Kobe": 0.0, "Cuenca": 0.0, "Sfax": 0.0, "Darwin": 0.0, "Accra": 0.0, "Lagos": 0.0, "Osaka": 0.0, "Porto": 1.0, "Braga": 0.0, "Dakar": 0.0, "Oslo": 0.0, "Bergen": 0.0, "Hanoi": 0.0, "Hue": 0.0, "Thies": 0.0, "Cusco":...
partial_observation_calibration:train:0
partial_observation_calibration
3
{"prior_probability_incident": "1/3", "assumption": "Sensor observations are conditionally independent given whether the incident is real.", "sensors": [{"id": "S1", "observed": "positive", "p_positive_given_incident": "4/5", "p_positive_given_no_incident": "1/5"}, {"id": "S2", "observed": "positive", "p_positive_given...
{"incident_real": {"type": "noul", "instructions": "After conditioning on every sensor observation, is the incident more likely real than not?"}}
{"incident_real": {"type": "noul", "noul": 0.8}}
policy_applicability:train:0
policy_applicability
3
{"request": {"subject": {"role": "engineer", "team": "gamma", "clearance": 1}, "resource": {"team": "beta", "sensitivity": "public"}, "action": "write"}, "policies": [{"id": "P1", "priority": 8, "effect": "allow", "teams": ["beta", "gamma"], "min_clearance": 2, "max_sensitivity": "public"}, {"id": "P5", "priority": 4, ...
{"access_allowed": {"type": "noul", "instructions": "Does the governing policy allow the requested action?"}, "governing_policy": {"type": "choice", "instructions": "Which policy governs the request, i.e. has the highest priority among the policies whose constraints all hold?", "criteria": {"P1": null, "P2": null, "P3"...
{"access_allowed": {"type": "noul", "noul": 1.0}, "governing_policy": {"type": "choice", "choice": "P8", "confidence": 1.0, "probabilities": {"P1": 0.0, "P2": 0.0, "P3": 0.0, "P4": 0.0, "P5": 0.0, "P6": 0.0, "P7": 0.0, "P8": 1.0}}, "review_risk": {"type": "score", "score": 0.0, "legend": {"0": "The resource is not rest...
policy_under_uncertainty:train:0
policy_under_uncertainty
3
{"request": {"subject": {"role": "unknown", "team": "gamma", "clearance": 1}, "resource": {"team": "beta", "sensitivity": "public"}, "action": "read"}, "role_evidence": {"history": "Of the last 54 requests from this account, 8 by analysts, 29 by engineers, 17 by managers. No other role has used it.", "reports": [{"sour...
{"access_allowed": {"type": "noul", "instructions": "Given the uncertainty about the requester's role, is the request more likely allowed than denied?"}, "governing_policy": {"type": "choice", "instructions": "Which policy most likely governs the request?", "criteria": {"P1": null, "P2": null, "P3": null, "P4": null, "...
{"access_allowed": {"type": "noul", "noul": 0.211144}, "governing_policy": {"type": "choice", "choice": "P4", "confidence": 1.0, "probabilities": {"P1": 0.211144, "P2": 0.0, "P3": 0.0, "P4": 0.788856, "P5": 0.0, "P6": 0.0, "none (default deny)": 0.0}}, "requester_role": {"type": "choice", "choice": "manager", "confiden...
End of preview. Expand in Data Studio

tasksource/procedural-typed-decisions(複製)

データの内容

手続き的に生成した意思決定の問題のデータ。1 行が 1 つの状態(state)で、同じ状態に対する複数の型付きの質問(questions)と答え(answers)を持つ。答えは状態が明示するルールから厳密に計算されるため、カードはラベルにノイズは無いとしている。生成器は reasoning-core の Jev タスクで、TypeSafe の公式の Jev データセットではない。

データの概要

項目 値
行数 792,000(19 構成・57 分割の合計)
列数 6〜22(構成による)
容量 456.87 MB(ファイル 62 本)
ファイルの形式 parquet 59、json 1
言語 en
公開 public

構成と分割

構成(config) 分割(split)ごとの行数 列数
all train 360,000、validation 18,000、test 18,000 6
arithmetic train 20,000、validation 1,000、test 1,000 22
entity_belief_tracking train 20,000、validation 1,000、test 1,000 9
event_state_reconstruction train 20,000、validation 1,000、test 1,000 8
event_triggered_execution train 20,000、validation 1,000、test 1,000 10
evidence_sufficiency train 20,000、validation 1,000、test 1,000 8
goal_directed_planning train 20,000、validation 1,000、test 1,000 11
graph_navigation train 20,000、validation 1,000、test 1,000 10
instruction_execution train 20,000、validation 1,000、test 1,000 11
multi_view_adjudication train 20,000、validation 1,000、test 1,000 8
needle_retrieval train 20,000、validation 1,000、test 1,000 8
partial_observation_calibration train 20,000、validation 1,000、test 1,000 6
policy_applicability train 20,000、validation 1,000、test 1,000 8
policy_under_uncertainty train 20,000、validation 1,000、test 1,000 8
record_aggregation train 20,000、validation 1,000、test 1,000 10
record_eligibility train 20,000、validation 1,000、test 1,000 12
state_perturbation train 20,000、validation 1,000、test 1,000 8
table_lookup train 20,000、validation 1,000、test 1,000 9
taxonomy_routing train 20,000、validation 1,000、test 1,000 8

列(構成 all・分割 train。構成ごとに列が違う)

列 型 先頭の行の値
id string arithmetic:train:0
task string arithmetic
level int64 0
state string {"opening_balance": 35, "currency": "£", "transa…
questions string {"final_balance": {"type": "choice", "instructio…
answers string {"final_balance": {"type": "choice", "choice": "…

行数と列は HF のデータセットビューアの集計(元のデータセットの現在の版で数えたもの)。全データセットの一覧は HYUKI5/dataset-catalog にある。

複製について

tasksource/procedural-typed-decisions をリビジョン 05ca1d0388f57461a634a7caed67145d28e2eca7 の時点でそのまま複製したもの。中身は変えていない。判断モデル Laya(convaiinnovations/laya)の微調整に使う候補として、調査メモ(Laya 学習用データセット調査)に載せたデータを集めている。

出典

論文

アノテーションの方法

(元のカードに該当の見出しが無いので、ラベルの付け方に触れた段落を抜き出した)

手続き的に生成された意思決定問題のデータセットである。各行は構造化された 1 つの状態(JSON、または算術・検索・集約の構成では表、CSV、key=value の行、散文のいずれか)であり、同じ状態に対する複数の型付きの質問を持つ。質問の形式は Jev / System One のリクエストの形に従い、choice(基準を 1 つ選ぶ)、noul([0, 1] の数値。確率、または yes/no)、score(順序付きのルーブリック)の 3 種類がある。すべての答えは、状態そのものが明示するルールによって状態から厳密に計算されるため、ラベルにノイズは含まれない。質問の答えのうち、yes/no や choice のラベルが 60% を超えることはない。タスクの質問のうち yes/no の質問は高々 3 分の 1 である。状態は、答えや yes/no の形式がすでに過剰に表れている質問を省くため、状態ごとに質問の部分集合が異なる。選択肢の数(4 から 60)を変える構成がいくつかあり、Jev の残りの大部分を占める二値の質問と 4〜6 択の質問のバランスを取る。

レベル 0 は、強力な判断モデルにとって易しく、レベル 4 は難しくなるように意図されている。下の表は、Jev の偶然を補正した正答率(カッパ = (正答率 − 偶然の正答率) / (1 − 偶然の正答率))を示す。測定はレベルごとに新たに生成した 40 個の状態について行った(各状態のすべての質問を対象とする)。測定には typesafe/jev-1.13-20260917(2026 年 9 月)を用いた。ただし event_triggered_execution、goal_directed_planning、graph_navigation、instruction_execution、partial_observation_calibration、record_eligibility については、jev-latest(2026 年 10 月)を用い、レベルごとに 30〜40 個の状態で測定した。値の 1 は完全な正答、0 は偶然と同じであることを表す。この測定は、生成器が reasoning-core に移る前に行われた。そこでは 5 つの構成(event_triggered_execution、goal_directed_planning、graph_navigation、instruction_execution、record_eligibility)が同一の状態を生成していた。残りの構成は、規則とレベルは保ったまま、描画方法と抽出を改めている。

tasksource、その標準化されたタスクのアノテーション、またはそれらから派生したデータセットを公表された研究で使う場合は、tasksource の論文(Sileo, LREC-COLING 2024)を引用すること。手続き的データは Reasoning Core で生成されているため、Reasoning Core の論文(Sileo et al., 2026)も引用すること。

微調整に使うときの注意

  • 調査メモの台帳での区分: B(継承条件・gated・モデル由来の条件・上流の規約などが付く)。区分は記載の文言で機械的に付けたもので、法的な判定ではない。学習に入れる前に元のカードとライセンスの原文で確かめる。
  • Laya の入力は 2,048 token まで(ModernBERT-large / mmBERT)。長い本文は先頭から黙って切られるので、切れる行は事前に除くか、要約・切り出しをしてから入れる。
  • 分割は記事・文書単位で行い、同じ本文から作った別の問いを学習側と較正側に分けない。温度と閾値は学習に使っていない分割で別々に当てる。

調査メモの記述

メモ 節 ライセンス(記載) 確認 注記
multitask_weak_synthetic.md Q1. Laya と同形式の合成・変換済み「typed decisions」データは、公開されているか (L25) apache-2.0(HF API) 確認済 原文注記: 金融ニュース・アナリスト判断とは領域が違う。needle_retrieval は最大約 300 レコードで、Laya の context(512〜1024)を超える。README が level 別の Jev の cha
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) apache-2.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] procedural_exact_rule
jev_sources_community.csv CSV(sulabhkatiyar/typed-decisions-english-training-mix) licence_tier=default 確認済 [Jev 元データ source] 上流の人手ラベル中心(約 8% が soft。label_kind = human/program/teacher)

元のカード(日本語訳)

以下は元のデータセットのカードの本文を日本語に訳したもの(機械翻訳)。訳に疑問があるときは、末尾の原文を確かめる。


procedural-typed-decisions

手続き的に生成された意思決定問題のデータセットである。各行は構造化された 1 つの状態(JSON、または算術・検索・集約の構成では表、CSV、key=value の行、散文のいずれか)であり、同じ状態に対する複数の型付きの質問を持つ。質問の形式は Jev / System One のリクエストの形に従い、choice(基準を 1 つ選ぶ)、noul([0, 1] の数値。確率、または yes/no)、score(順序付きのルーブリック)の 3 種類がある。すべての答えは、状態そのものが明示するルールによって状態から厳密に計算されるため、ラベルにノイズは含まれない。質問の答えのうち、yes/no や choice のラベルが 60% を超えることはない。タスクの質問のうち yes/no の質問は高々 3 分の 1 である。状態は、答えや yes/no の形式がすでに過剰に表れている質問を省くため、状態ごとに質問の部分集合が異なる。選択肢の数(4 から 60)を変える構成がいくつかあり、Jev の残りの大部分を占める二値の質問と 4〜6 択の質問のバランスを取る。

生成器は reasoning-core の Jev タスクである(reasoning_core/tasks/jev/。ここでは arithmetic が practical_arithmetic、evidence_sufficiency が claim_support に当たる)。コミットとタスクのバージョンは provenance.json に記録されている。

これは独立したデータセットである。TypeSafe の公式の Jev データセットではなく、TypeSafe や OpenJev が作成したものでも、両者と提携したものでもない。

構成

config questions
all(既定) 下の全構成を 1 つの表にまとめたもの。task 列と共通のフィールドだけを持つ(フラットなラベル列は持たない)。train の最初の 1,000 行はレベルとタスクを順に巡り、残りはシャッフルされている
arithmetic 割引と送料の規則つきの注文、勘定元帳、またはスケジュール。各状態は次のうち 2〜5 個を問う: amount_due / final_balance / finish_time(結果と典型的な誤答から選ぶ choice)、within_budget、went_negative、done_by_deadline(noul)、random_line_bulk、random_is_deposit、random_is_long(noul。正確な確率 k/n)、budget_use、net_change(score。記述的なレベル)、lines_above、withdrawal_count、starts_before_noon(score)、largest_line、lowest_day、longest_task(choice)
entity_belief_tracking world_location(choice)、agent_belief_location(choice)、belief_matches_world(noul)。レベル 2 からは nested_belief_location(choice。A が、B はある物がどこにあると信じていると考えているかを問う)も加わる。場所は 4〜16 個
event_state_reconstruction current_owner(choice)、is_open(noul)、current_severity(score)。ログはレベル 2 からシャッフルされ、レベル 3 からは無効化された項目を含む
event_triggered_execution rule_on_signal(choice)、object_final_state、rule_enabled_at_end、object_ever_targeted(noul)、total_firings(score)。タイムスタンプ付きの 3〜18 個のイベントに対して、1〜8 個の競合するルールがある。ルールには優先度、イベントの条件(ガード)、クールダウン、有効化と無効化の制御があり、提示順はソートされていない
evidence_sufficiency claim_supported(noul)、has_conflict(noul)、strongest_support_origin(choice)。レベル 2 からは撤回、レベル 3 からは互いに独立でない出所(鏡像の出所)、レベル 4 では収集日による有効性の判定を含む
goal_directed_planning goal_reachable、probe_applicable、probe_optimal_first、probe_required(noul)、minimal_actions(score)、first_legal_action(choice)。3〜12 ステップからなり、前提条件は連言(すべて必要)と選言(代替)の両方を含む。実行できない行動を含み、最短の計画のラベルは正確である
graph_navigation reachable、within_budget、edge_essential(noul)、shortest_cost、first_hop(choice)。4〜16 個の有向ノードからなり、辺にはコストがあり、閉鎖(通行止め)、必ず通る経由地、最適経路の代替案を含む
instruction_execution resolution と deferred_registration(choice)、target_count、changed_count(score)、probe_selected、probe_final_state(noul)。3〜26 個のエンティティにわたり、自然言語による選択、例外、即時と遅延の更新、曖昧さ、原子的(atomic)・ベストエフォート・順序付きの実行を扱う
multi_view_adjudication intent(choice)、is_urgent(noul)、workflow_impact(score)。レベル 2 からは閾値付近のシグナル、レベル 3 からはログインイベントから数える認証失敗、レベル 4 では時刻として与えられる期限を含む
needle_retrieval value_of_id(choice)、id_has_value(noul)、id_listed(noul)。最大約 300 件のレコードがあり、その ID は対象の ID と 1〜2 桁だけ異なる。レベル 2 からは、1〜3 回の ID の再発行をたどる連鎖も加わる。選択肢は 6〜20 個
partial_observation_calibration incident_real(noul。正確なベイズの事後確率)。センサーは 1〜6 個
policy_applicability access_allowed(noul)、governing_policy(choice)、review_risk(score)。レベル 0 では制約が 1 つだけのポリシーが 2 個、レベル 4 では制約が最大 5 個のポリシーが約 12 個あり、その多くは惜しいが該当しない(ニアミス)ものである
policy_under_uncertainty access_allowed(noul)、governing_policy(choice)、requester_role(choice)。履歴の件数と、明示された信頼性を持つ報告から知られる役割について、正確な事後確率を求める
record_aggregation count_in_category(score)、largest_quantity(choice)、any_out_of_stock(noul)、total_above(noul)。レベル 2 からは count_filtered(score。数量と在庫のフィルタつき)も加わる
record_eligibility 散文のみのレコードの説明と、番号付きのユーザー要件。probe_eligibility、requirement_status、first_eligible、decisive_constraint(choice)、eligible_count(score)、eligible_if_relaxed、probe_fully_eligible(noul)。資格の判定は 3 値であり、連言、否定、選言、条件付きの義務、合計コストの上限、欠けている証拠や食い違う証拠、反事実的な緩和を扱う。レコード数と制約数はレベルによって異なり、4〜27 件のレコードと 1〜6 個の制約がある。
state_perturbation material_change(noul)、changed_dimension(choice)、risk_direction(score)。1〜8 件のレコードがあり、同時に最大 4 個の変更を含む。変更のリスクへの影響は互いに打ち消し合うことがあり、見た目が似ているが重要ではないフィールドも含む
table_lookup find_person(choice。2 条件のフィルタ。レベル 2 からは上司をたどり、レベル 3 からは開始年の条件を伴う。選択肢は 6〜40 個で、表の行数が上限となる)、manager_of(choice。結合)、started_before(noul)、count_matching(score)
taxonomy_routing route(choice。状態ごとに引き直される、ルーティングガイドの 4〜60 個のカテゴリから選ぶ。多くのルールが正しいルールと同じ条件を共有する)、belongs_to(noul)、conditions_met(score。0〜4)

レコードの資格判定の生成

record_eligibility の状態に含まれるのは、自然言語のユーザーリクエスト、散文のレコードの説明、そして明示された評価の意味づけだけである。元になる事実と論理式の木は生成器のテストのメタデータに保持されており、公開されたデータセットには含まれない。不明な事実は違反とみなさない。違反が確認された要件は、不明な要件よりも優先される。最初に資格を満たすレコードは表示された順序で定まり、決定的な要件は番号付きの制約によって定まる。

難易度は 5 つのレベルにわたって上がる。レコード数(4〜6 から 25〜27 へ)と要件数(1 から 6 へ)が増え、入れ子の OR、条件付きの要件、合計された料金、閾値付近の紛らわしい情報、矛盾する報告が導入される。

スキーマ

field meaning
id task:split:index
level 難易度のレベル(0〜4)。Jev に対して較正してある(後述)。
state 状態。JSON の文字列か、検索と集約の構成では描画されたテキスト。
questions 名前付きの System One の質問からなる JSON オブジェクト(type、instructions、criteria)。
answers System One の answers の形をした、参照用の答えの JSON オブジェクト。
質問ごとに 1 列 閲覧と絞り込みのためのフラットなラベル。choice、score、yes/no の noul の質問には ClassLabel、段階的な noul(incident_real、random_*)には float、数値を開いた choice(amount_due、final_balance、finish_time)には選択肢のテキストが入る。状態がその質問を問わない場合は null となる(arithmetic と、レベルによって異なる質問)。

状態は分割の中で一意である。validation と test の状態は train に現れない。各構成で、train の最初の 1,000 行は閲覧用に、レベルを(易しいものから順に)巡る。分割の残りはシャッフルされている。

レベルごとの難易度

レベル 0 は、強力な判断モデルにとって易しく、レベル 4 は難しくなるように意図されている。下の表は、Jev の偶然を補正した正答率(カッパ = (正答率 − 偶然の正答率) / (1 − 偶然の正答率))を示す。測定はレベルごとに新たに生成した 40 個の状態について行った(各状態のすべての質問を対象とする)。測定には typesafe/jev-1.13-20260917(2026 年 9 月)を用いた。ただし event_triggered_execution、goal_directed_planning、graph_navigation、instruction_execution、partial_observation_calibration、record_eligibility については、jev-latest(2026 年 10 月)を用い、レベルごとに 30〜40 個の状態で測定した。値の 1 は完全な正答、0 は偶然と同じであることを表す。この測定は、生成器が reasoning-core に移る前に行われた。そこでは 5 つの構成(event_triggered_execution、goal_directed_planning、graph_navigation、instruction_execution、record_eligibility)が同一の状態を生成していた。残りの構成は、規則とレベルは保ったまま、描画方法と抽出を改めている。

config level 0 1 2 3 4
arithmetic 0.83 0.64 0.59 0.54 0.59
entity_belief_tracking 0.91 0.93 0.78 0.74 0.66
event_state_reconstruction 0.97 0.99 0.92 0.73 0.60
event_triggered_execution 0.91 0.75 0.63 0.44 0.60
evidence_sufficiency 0.99 0.97 0.78 0.72 0.78
goal_directed_planning 1.00 0.87 0.62 0.53 0.51
graph_navigation 0.90 0.81 0.54 0.27 0.17
instruction_execution 0.99 0.98 0.73 0.72 0.67
multi_view_adjudication 0.82 0.87 0.82 0.73 0.74
needle_retrieval 1.00 0.99 0.78 0.83 0.36
partial_observation_calibration 0.74 0.45 0.35 0.30 0.20
policy_applicability 0.73 0.60 0.53 0.64 0.35
policy_under_uncertainty 0.35 0.42 0.38 0.57 0.39
record_aggregation 0.99 0.96 0.84 0.75 0.70
record_eligibility 0.97 0.94 0.74 0.56 0.53
state_perturbation 0.95 0.89 0.44 0.63 0.50
table_lookup 1.00 0.98 0.97 0.90 0.87
taxonomy_routing 1.00 0.98 0.92 0.78 0.64

確率の答えは、上の表では yes/no に丸めて採点している。正確な確率に対する Jev の平均絶対誤差は、incident_real ではレベル 0 の 0.16 からレベル 4 の 0.29 へ増え、policy_under_uncertainty の事後確率 access_allowed では 0.33 前後で推移する。policy_under_uncertainty と partial_observation_calibration(正確な事後確率を求める)は、レベル 0 から難しい。table_lookup と multi_view_adjudication は、レベル 4 でも最も易しいままである。

2 つ目のモデル upstage/solar-decide(レベルごとに 10 個の状態。選択肢は最大 26 個までしか扱えないため、最も長いリストは除外してある)では、ほとんどの構成で同じく易しいものから難しいものへの傾きが見られる。例えば event_state_reconstruction では 0.95 → 0.53、evidence_sufficiency では 1.00 → 0.48、policy_applicability では 0.88 → 0.42 である。arithmetic、table_lookup、state_perturbation は、このモデルにとってもどのレベルでも易しいまま(おおよそ 0.8〜0.9)である。再実行は scripts/calibrate_procedural_levels.py で行う(別のモデルを使うには --model を指定する。対象は OpenRouter の decisions API)。

使い方

複数質問の Jev リクエストとして、{"state": row["state"], "questions": json.loads(row["questions"])} を送る(状態が JSON の場合は先に解析する)。答えは row["answers"] と比べる。 同じ行は、状態ごとにまとめた形で tasksource/tasksource-jev-typed-decisions にも含まれている。

再現

生成は決定的である(分割の行 i は task:split:i でシードされる)。tasksource のチェックアウトから、provenance.json に記された reasoning-core のコミットをインストールした上で実行する:

PYTHONPATH=.:src python scripts/build_procedural_jev.py --output build/procedural-typed-decisions --upload

引用

tasksource、その標準化されたタスクのアノテーション、またはそれらから派生したデータセットを公表された研究で使う場合は、tasksource の論文(Sileo, LREC-COLING 2024)を引用すること。手続き的データは Reasoning Core で生成されているため、Reasoning Core の論文(Sileo et al., 2026)も引用すること。

@inproceedings{sileo-2024-tasksource,
    title = "tasksource: A Large Collection of {NLP} tasks with a Structured Dataset Preprocessing Framework",
    author = "Sileo, Damien",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
    month = may,
    year = "2024",
    address = "Torino, Italia",
    publisher = "ELRA and ICCL",
    url = "https://aclanthology.org/2024.lrec-main.1361",
    doi = "10.63317/2h4pkr685mkq",
    pages = "15655--15684",
}

@article{sileo2026reasoning,
  title={Reasoning Core: Designing Broad Procedural Data for Completion-Supervised Reasoning Training},
  author={Sileo, Damien and Lacombe, Valentin and Kachler, Dimitri},
  journal={arXiv preprint arXiv:2608.05148},
  year={2026},
  url={https://arxiv.org/abs/2608.05148}
}
原文(英語)

procedural-typed-decisions

Procedurally generated decision problems. Each row is one structured state (JSON, or a table, CSV, key=value lines, or prose for the arithmetic, retrieval, and aggregation configs) with several typed questions over that same state, following the Jev / System One request shape: choice (pick one criterion), noul (a number in [0, 1]; a probability or a yes/no), and score (an ordered rubric). Every answer is computed exactly from the state by rules that the state itself spells out, so the labels are noise-free. No yes/no or choice label exceeds 60% of a question's answers: yes/no questions are at most a third of a task's questions. A state skips a question whose answer, or whose yes/no format, is already over-represented, so states ask varying subsets of their task's questions. Several configs vary the number of options (4 to 60), to balance the binary and 4–6-option questions that dominate the rest of Jev.

The generators are the Jev tasks of reasoning-core (reasoning_core/tasks/jev/; arithmetic is practical_arithmetic there and evidence_sufficiency is claim_support); provenance.json records the commit and task versions.

This is an independent dataset. It is not an official TypeSafe Jev dataset and is not produced by or affiliated with TypeSafe or OpenJev.

Configs

config questions
all (default) Every config below in one table, with a task column and the shared fields only (no flat label columns); the first 1,000 train rows cycle through levels and tasks, the rest is shuffled
arithmetic An order with a discount/shipping rule, an account ledger, or a schedule; each state asks 2–5 of: amount_due / final_balance / finish_time (choice among the result and typical slips), within_budget, went_negative, done_by_deadline (noul), random_line_bulk, random_is_deposit, random_is_long (noul, exact probability k/n), budget_use, net_change (score, descriptive levels), lines_above, withdrawal_count, starts_before_noon (score), largest_line, lowest_day, longest_task (choice)
entity_belief_tracking world_location (choice), agent_belief_location (choice), belief_matches_world (noul), from level 2 nested_belief_location (choice: where A thinks B believes an object is); 4 to 16 locations
event_state_reconstruction current_owner (choice), is_open (noul), current_severity (score); the log is shuffled from level 2 and has voided entries from level 3
event_triggered_execution rule_on_signal (choice), object_final_state, rule_enabled_at_end, object_ever_targeted (noul), total_firings (score); 1–8 competing rules over 3–18 timestamped events, with priorities, event guards, cooldowns, enable/disable controls, and unsorted presentation
evidence_sufficiency claim_supported (noul), has_conflict (noul), strongest_support_origin (choice); retractions from level 2, mirrored (non-independent) origins from level 3, validity by collection day at level 4
goal_directed_planning goal_reachable, probe_applicable, probe_optimal_first, probe_required (noul), minimal_actions (score), first_legal_action (choice); 3–12 steps with conjunctive and alternative prerequisites, unavailable actions, and exact shortest-plan labels
graph_navigation reachable, within_budget, edge_essential (noul), shortest_cost, first_hop (choice); 4–16 directed nodes with edge costs, closures, mandatory waypoints, and optimal-route alternatives
instruction_execution resolution and deferred_registration (choice), target_count, changed_count (score), probe_selected, probe_final_state (noul); natural-language selection, exceptions, immediate/deferred updates, ambiguity, and atomic/best-effort/ordered execution across 3–26 entities
multi_view_adjudication intent (choice), is_urgent (noul), workflow_impact (score); near-threshold signals from level 2, auth failures counted from login events from level 3, deadlines as clock times at level 4
needle_retrieval value_of_id (choice), id_has_value (noul), id_listed (noul); up to ~300 records whose ids differ from the target by one or two digits, and from level 2 a chain of one to three id reissues to follow; 6 to 20 options
partial_observation_calibration incident_real (noul, exact Bayesian posterior); 1 to 6 sensors
policy_applicability access_allowed (noul), governing_policy (choice), review_risk (score); 2 one-constraint policies at level 0, about 12 policies of up to 5 constraints, many of them near misses, at level 4
policy_under_uncertainty access_allowed (noul), governing_policy (choice), requester_role (choice); exact posteriors over a role known through history counts and reports of stated reliability
record_aggregation count_in_category (score), largest_quantity (choice), any_out_of_stock (noul), total_above (noul), from level 2 count_filtered (score, quantity and stock filters)
record_eligibility Prose-only record descriptions and numbered user requirements. probe_eligibility, requirement_status, first_eligible, decisive_constraint (choice); eligible_count (score), eligible_if_relaxed, probe_fully_eligible (noul). Three-valued eligibility covers conjunction, negation, disjunction, conditional obligations, combined-cost limits, missing or conflicting evidence, and counterfactual relaxation. 4–27 records and 1–6 constraints by level.
state_perturbation material_change (noul), changed_dimension (choice), risk_direction (score); 1 to 8 records with up to 4 simultaneous changes whose risk effects can offset, and look-alike non-material fields
table_lookup find_person (choice, two-condition filter, through the manager from level 2 and with a start-year condition from level 3; 6 to 40 options, capped by the table), manager_of (choice, join), started_before (noul), count_matching (score)
taxonomy_routing route (choice among the 4–60 categories of a routing guide drawn fresh per state; many rules share a condition with the right one), belongs_to (noul), conditions_met (score, 0–4)

Record eligibility generation

The record_eligibility state contains only a natural-language user request, prose record descriptions, and explicit evaluation semantics. Source facts and logical expression trees are kept in generator test metadata, not the published dataset. Unknown facts are not treated as violations; a known violated requirement dominates any unknown requirements. The first eligible record is defined by the displayed order, and the decisive requirement by the numbered constraints.

Difficulty rises across five levels by increasing record count (4–6 to 25–27), requirement count (1 to 6), and the introduction of nested OR, conditional requirements, summed charges, near-threshold distractors and contradictory reports.

Schema

field meaning
id task:split:index
level Difficulty level (0–4), calibrated against Jev (see below).
state The state: a JSON string, or rendered text for the retrieval and aggregation configs.
questions JSON object of named System One questions (type, instructions, criteria).
answers JSON object of reference answers, in the System One answers shape.
one column per question Flat label, for browsing and filtering: a ClassLabel for choice, score, and yes/no noul questions; a float for graded noul (incident_real, random_*); the option text for open numeric choices (amount_due, final_balance, finish_time). Null when the state does not ask that question (arithmetic, and level-dependent questions).

States are unique within a split, and validation/test states never occur in train. In each config, the first 1,000 train rows cycle through the levels (easiest first) for browsing; the rest of the split is shuffled.

Difficulty by level

Level 0 is meant to be easy for a strong decision model and level 4 hard. The table gives Jev's chance-adjusted accuracy, kappa = (accuracy − chance) / (1 − chance), on 40 fresh states per level (every question of each state; typesafe/jev-1.13-20260917, September 2026; jev-latest, October 2026, for event_triggered_execution, goal_directed_planning, graph_navigation, instruction_execution, partial_observation_calibration and record_eligibility, 30–40 states per level). 1 is perfect, 0 is chance. It was measured before the generators moved to reasoning-core: five configs generate identical states there (event_triggered_execution, goal_directed_planning, graph_navigation, instruction_execution, record_eligibility), the others keep their rules and levels with revised renderings and sampling.

config level 0 1 2 3 4
arithmetic 0.83 0.64 0.59 0.54 0.59
entity_belief_tracking 0.91 0.93 0.78 0.74 0.66
event_state_reconstruction 0.97 0.99 0.92 0.73 0.60
event_triggered_execution 0.91 0.75 0.63 0.44 0.60
evidence_sufficiency 0.99 0.97 0.78 0.72 0.78
goal_directed_planning 1.00 0.87 0.62 0.53 0.51
graph_navigation 0.90 0.81 0.54 0.27 0.17
instruction_execution 0.99 0.98 0.73 0.72 0.67
multi_view_adjudication 0.82 0.87 0.82 0.73 0.74
needle_retrieval 1.00 0.99 0.78 0.83 0.36
partial_observation_calibration 0.74 0.45 0.35 0.30 0.20
policy_applicability 0.73 0.60 0.53 0.64 0.35
policy_under_uncertainty 0.35 0.42 0.38 0.57 0.39
record_aggregation 0.99 0.96 0.84 0.75 0.70
record_eligibility 0.97 0.94 0.74 0.56 0.53
state_perturbation 0.95 0.89 0.44 0.63 0.50
table_lookup 1.00 0.98 0.97 0.90 0.87
taxonomy_routing 1.00 0.98 0.92 0.78 0.64

Probability answers are scored above by their rounding to yes/no; Jev's mean absolute error on the exact probability grows from 0.16 (level 0) to 0.29 (level 4) on incident_real, and stays around 0.33 on the posterior access_allowed of policy_under_uncertainty. policy_under_uncertainty and partial_observation_calibration (exact posteriors) are hard from level 0 on; table_lookup and multi_view_adjudication remain the easiest at level 4.

A second model, upstage/solar-decide (10 states per level; it takes at most 26 options, so the longest lists are left out), shows the same easy-to-hard slope on most configs, e.g. 0.95 → 0.53 on event_state_reconstruction, 1.00 → 0.48 on evidence_sufficiency, 0.88 → 0.42 on policy_applicability; arithmetic, table_lookup, and state_perturbation stay easy for it (about 0.8–0.9 at every level). Rerun with scripts/calibrate_procedural_levels.py (--model for another model of the OpenRouter decisions API).

Use

As a multi-question Jev request, send {"state": row["state"], "questions": json.loads(row["questions"])} (parsing the state first when it is JSON) and compare with row["answers"]. The same rows are included, grouped by state, in tasksource/tasksource-jev-typed-decisions.

Reproduction

Generation is deterministic (row i of a split is seeded by task:split:i). From a tasksource checkout, with the reasoning-core commit of provenance.json installed:

PYTHONPATH=.:src python scripts/build_procedural_jev.py --output build/procedural-typed-decisions --upload

Citation

If you use tasksource, its standardized task annotations, or datasets derived from them in published research, please cite the tasksource paper (Sileo, LREC-COLING 2024). The procedural data is generated with Reasoning Core; please also cite the Reasoning Core paper (Sileo et al., 2026).

@inproceedings{sileo-2024-tasksource,
    title = "tasksource: A Large Collection of {NLP} tasks with a Structured Dataset Preprocessing Framework",
    author = "Sileo, Damien",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
    month = may,
    year = "2024",
    address = "Torino, Italia",
    publisher = "ELRA and ICCL",
    url = "https://aclanthology.org/2024.lrec-main.1361",
    doi = "10.63317/2h4pkr685mkq",
    pages = "15655--15684",
}

@article{sileo2026reasoning,
  title={Reasoning Core: Designing Broad Procedural Data for Completion-Supervised Reasoning Training},
  author={Sileo, Damien and Lacombe, Valentin and Kachler, Dimitri},
  journal={arXiv preprint arXiv:2608.05148},
  year={2026},
  url={https://arxiv.org/abs/2608.05148}
}
Downloads last month
180

Paper for HYUKI5/tasksource__procedural-typed-decisions