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}
}