Datasets:
Modalities:
Text
Formats:
json
Languages:
English
Size:
1K - 10K
Tags:
benchmark
llm-evaluation
reasoning
knowledge-representation
open-world-assumption
closed-world-assumption
License:
| { | |
| "dataset": "closurebench_base.jsonl", | |
| "version": "closurebench-v1", | |
| "num_items": 960, | |
| "num_base_scenarios": 320, | |
| "domains": { | |
| "cloud_deployment": 120, | |
| "hospital_access": 120, | |
| "library_services": 120, | |
| "university_registration": 120, | |
| "finance_controls": 120, | |
| "procurement_review": 120, | |
| "manufacturing_quality": 120, | |
| "robotics_operations": 120 | |
| }, | |
| "families": { | |
| "entailed_open_conclusion": 96, | |
| "explicit_negative_closed": 96, | |
| "explicit_negative_open": 96, | |
| "closed_missing_with_open_distractor": 96, | |
| "open_missing_with_closed_distractor": 96, | |
| "open_missing_direct": 96, | |
| "entailed_closed_conclusion": 96, | |
| "closed_missing_direct": 96, | |
| "closed_derived_missing_antecedent": 96, | |
| "open_derived_missing_antecedent": 96 | |
| }, | |
| "subsets": { | |
| "control_entailed": 192, | |
| "control_explicit_negative": 192, | |
| "core_contrastive": 576 | |
| }, | |
| "splits": { | |
| "dev": 93, | |
| "train": 648, | |
| "test": 219 | |
| }, | |
| "semantics": { | |
| "owa": 320, | |
| "cwa": 320, | |
| "lcwa": 320 | |
| }, | |
| "gold_answers": { | |
| "true": 192, | |
| "false": 480, | |
| "unknown": 288 | |
| }, | |
| "gold_reason_types": { | |
| "entailed_by_fact_or_rule": 192, | |
| "explicit_negative_fact": 192, | |
| "global_closed_world_underivable": 192, | |
| "local_open_world_underivable": 96, | |
| "open_world_underivable": 192, | |
| "local_closed_world_underivable": 96 | |
| }, | |
| "semantic_switch_patterns": { | |
| "false|false|false": 64, | |
| "true|true|true": 64, | |
| "unknown|false|false": 96, | |
| "unknown|false|unknown": 96 | |
| }, | |
| "primary_metrics": [ | |
| "semantic_switch_accuracy", | |
| "lcwa_closed_scope_accuracy", | |
| "lcwa_open_scope_accuracy", | |
| "cwa_accuracy", | |
| "owa_accuracy" | |
| ], | |
| "notes": [ | |
| "V3 preserves pilot v1/v2 files and should be treated as the publication candidate.", | |
| "Each base scenario has OWA, CWA, and LCWA variants with identical facts, rules, and target atom.", | |
| "Each item contains natural-language text plus symbolic atoms/rules and a gold reason type.", | |
| "Closure is explicit: closed predicates make unstated and underivable atoms false after rule closure.", | |
| "Controls are retained but marked separately from core contrastive cases." | |
| ] | |
| } |