{"record": {"aliases": ["DeFAb: A Verifiable Benchmark for Defeasible Abduction in Foundation Models"], "artifacts": [{"id": "arxiv:2606.18557", "kind": "paper", "url": "https://arxiv.org/abs/2606.18557"}, {"id": "gh:patrickallencooper/blanc", "kind": "repo", "url": "https://github.com/PatrickAllenCooper/blanc"}, {"id": "hf:patrickallencooper/defab", "kind": "dataset", "url": "https://huggingface.co/datasets/PatrickAllenCooper/DeFAb"}], "authors": ["Patrick Cooper", "Avyay Sadhu", "Alvaro Velasquez"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.18557", "source_url": "https://arxiv.org/abs/2606.18557"}, "categories": ["Factuality", "General AI", "Language & Knowledge", "cs.AI"], "description": {"en": "A formally grounded benchmark and generation pipeline for evaluating defeasible abduction through automatically verifiable hypothesis construction."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.18557", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.18557", "title": "DeFAb"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["University of Colorado Boulder"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.18557", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.18557", "source_url": "https://openalex.org/W7165133503"}, "key": "claire-radar:2606.18557", "modality": null, "name": "DeFAb", "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_defab-a-verifiable-benchmark-for-defeasibl_0a1a0b62", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "f72d66ada9b7948cb99c2e81b9e796f36425a1e13562da795df6e98b63fbc411", "relation": "introduces", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T17:15:50.996573Z", "source_url": "https://arxiv.org/abs/2606.18557"}, "publisher": null, "released": "2026-06-17", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.18557", "source_url": "https://arxiv.org/abs/2606.18557"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-18557", "source": "claire_radar", "source_benchmark_id": "2606.18557", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["DeFAb: A Verifiable Benchmark for Defeasible Abduction in Foundation Models"], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attentionForecast": {"confidence": "Low", "horizon": "7d", "reason": "The large released dataset, verifier-based scoring, and difficult frontier-model results may generate broad reasoning-evaluation interest.", "score": 4}, "capabilities": ["Factuality"], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T17:15:50.996573Z", "inputHash": "dbfb85150eb7faf6020b03188d16a347a510b7b649f961f608b66c706c07d999", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "DeFAb: A Verifiable Benchmark for Defeasible Abduction in Foundation Models", "canonicalNameSource": "paper_title", "decisionReason": "The source supplies a formal benchmark name, a defined hypothesis-construction task, deterministic verification criteria, public data, and an evaluation harness.", "evaluationEvidence": {"reuse": "code and evaluation harness: https://github.com/PatrickAllenCooper/blanc", "scoring": "every hypothesis must pass polynomial-time checks for valid derivation, conservativity, and minimality", "task": "constructing hypotheses that explain anomalies by overriding defaults while preserving unrelated expectations"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T17:15:50.996573Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A formally grounded benchmark and generation pipeline for evaluating defeasible abduction through automatically verifiable hypothesis construction.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We introduce DeFAb (Defeasible Abduction Benchmark), a dataset and generation pipeline that converts four decades of publicly funded knowledge bases into formally grounded instances for defeasible abduction: constructing hypotheses that explain anomalies by overriding defaults while preserving unrelated expectations."}, "familyId": "bmf_2c7d40a00695", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_defab-a-verifiable-benchmark-for-defeasibl_0a1a0b62", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/PatrickAllenCooper/blanc", "data": "https://huggingface.co/datasets/PatrickAllenCooper/DeFAb", "paper": "https://arxiv.org/abs/2606.18557", "pdf": "https://arxiv.org/pdf/2606.18557", "project": null, "report": "https://arxiv.org/abs/2606.18557"}, "metrics": [], "motivation": "A rule-based logic solver resolves every instance in our benchmark in under 50 microseconds with 100% accuracy; the best frontier language model reaches 65% at best and drops to 23.5% under rendering-robust evaluation (worst case over four surface renderings).", "name": "DeFAb", "oneLine": "A formally grounded benchmark and generation pipeline for evaluating defeasible abduction through automatically verifiable hypothesis construction.", "paperTitle": "DeFAb: A Verifiable Benchmark for Defeasible Abduction in Foundation Models", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.18557", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Runnable", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-06-17", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Patrick Cooper", "Alvaro Velasquez"], "categories": ["cs.AI", "cs.LG", "cs.LO"], "id": "2606.18557", "title": "DeFAb: A Verifiable Benchmark for Defeasible Abduction in Foundation Models", "type": "arxiv", "url": "https://arxiv.org/abs/2606.18557"}, "sourceUpdatedAt": "2026-06-17T00:13:40Z", "topics": ["cs.AI"], "venueAttempts": [], "whyItMatters": "Its rule-based validity, conservativity, and minimality checks provide judge-free evaluation of structured theory revision."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
