{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.06622", "kind": "paper", "url": "https://arxiv.org/abs/2606.06622"}, {"kind": "website", "url": "https://unpredictabenchmark.github.io/"}], "authors": ["Amirhossein Abaskohi", "Amirhossein Dabiriaghdam", "Liang Luo", "Ellie Dingqiao Wen", "Lele Wang", "Giuseppe Carenini", "Peter West"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.06622", "source_url": "https://arxiv.org/abs/2606.06622"}, "categories": ["General AI", "Language & Knowledge", "cs.CL"], "description": {"en": "UnpredictaBench evaluates whether language-model samples reproduce specified target distributions across statistical, program-induced, and natural-language random processes."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.06622", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.06622", "title": "UnpredictaBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["University of British Columbia"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.06622", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.06622", "source_url": "https://openalex.org/W7163874244"}, "key": "claire-radar:2606.06622", "modality": null, "name": "UnpredictaBench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_unpredictabench_09596471", "origin_source": "arxiv", "recognition_confidence": "0.75", "record_sha256": "8d7c2ac4178fd5b2a1c0f379ecccbcc7e77aee1134af41b37639e53e8d87980c", "relation": "introduces", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T16:18:42.106719Z", "source_url": "https://arxiv.org/abs/2606.06622"}, "publisher": null, "released": "2026-06-04", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.06622", "source_url": "https://arxiv.org/abs/2606.06622"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-06622", "source": "claire_radar", "source_benchmark_id": "2606.06622", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "The benchmark defines a clear, unusual capability gap and a compact comparable metric, supporting near-term discussion and reuse.", "score": 7}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T16:18:42.106719Z", "inputHash": "e7aa0000627d97125ca776d0e257fd897a18091ecb0a430fce9d4844ad1ab192", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "UnpredictaBench: A Benchmark for Evaluating Distributional Randomness in LLMs", "canonicalNameSource": "paper_title", "decisionReason": "The supplied text provides a formal name, a repeatable 448-problem task collection, the comparable KS@N metric with KS@100 identified as standard, and an available public resource path.", "evaluationEvidence": {"reuse": "Project website and resources are available at https://unpredictabenchmark.github.io/.", "scoring": "KS@N, a general-purpose evaluation metric that quantifies how well a model outputs approximate black-box target distributions via the Kolmogorov-Smirnov statistical test", "task": "sampling outcomes from individual target distributions, including canonical statistical distributions, distributions induced by stochastic programs, and natural-language scenarios that describe random processes"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T16:18:42.106719Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "UnpredictaBench evaluates whether language-model samples reproduce specified target distributions across statistical, program-induced, and natural-language random processes.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "evaluation protocol evidence", "public artifact URL"], "snippet": "We introduce UnpredictaBench, an evaluation that tests the ability of large language models (LLMs) to capture true underlying distributions."}, "familyId": "bmf_6865b30b6480", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_unpredictabench_09596471", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "paper": "https://arxiv.org/abs/2606.06622", "pdf": "https://arxiv.org/pdf/2606.06622", "project": "https://unpredictabenchmark.github.io/", "report": "https://arxiv.org/abs/2606.06622"}, "metrics": [], "motivation": "We introduce UnpredictaBench, an evaluation that tests the ability of large language models (LLMs) to capture true underlying distributions.", "name": "UnpredictaBench", "oneLine": "UnpredictaBench evaluates whether language-model samples reproduce specified target distributions across statistical, program-induced, and natural-language random processes.", "paperTitle": "UnpredictaBench: A Benchmark for Evaluating Distributional Randomness in LLMs", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.06622", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Paper only", "recognitionConfidence": 0.75, "relation": "introduces", "releasedAt": "2026-06-04", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Amirhossein Abaskohi", "Amirhossein Dabiriaghdam", "Liang Luo", "Ellie Dingqiao Wen", "Lele Wang", "Giuseppe Carenini", "Peter West"], "categories": ["cs.CL"], "id": "2606.06622", "title": "UnpredictaBench: A Benchmark for Evaluating Distributional Randomness in LLMs", "type": "arxiv", "url": "https://arxiv.org/abs/2606.06622"}, "sourceUpdatedAt": "2026-07-03T19:42:53Z", "topics": ["cs.CL"], "venueAttempts": [], "whyItMatters": "It distinguishes calibrated distributional sampling from superficial output diversity, which is important when models stand in for stochastic systems."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
