{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.16151", "kind": "paper", "url": "https://arxiv.org/abs/2606.16151"}], "authors": ["Hoang Pham", "Dong Le", "Anh Tuan Luu"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.16151", "source_url": "https://arxiv.org/abs/2606.16151"}, "categories": ["General AI", "Language & Knowledge", "Reasoning"], "description": {"en": "GRACE evaluates step-level faithfulness of chain-of-thought reasoning in context-grounded tasks, with human annotations and a taxonomy of error categories."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.16151", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.16151", "title": "GRACE"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.16151", "modality": null, "name": "GRACE", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_grace_1b12b544", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "165f43de156eeca198746c822c32d63819b3a2a22463f3187ea08c8ba527294f", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.16151"}, "publisher": null, "released": "2026-06-15", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.16151", "source_url": "https://arxiv.org/abs/2606.16151"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-16151", "source": "claire_radar", "source_benchmark_id": "2606.16151", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2606.16151", "observedAt": "2026-10-04T03:02:42.754940Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-10-01", "state": "stale"}}, "status": "stale"}, "capabilities": ["Reasoning"], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "81a05d96ef78b0f815e196a14a5027cbad43d39f4c52185407e45feb16e05b4d", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "GRACE evaluates step-level faithfulness of chain-of-thought reasoning in context-grounded tasks, with human annotations and a taxonomy of error categories.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "We introduce GRACE, the first human-annotated step-level faithfulness benchmark with a data-driven error taxonomy for context-grounded textual reasoning."}, "familyId": "bmf_e010fd1ce1ac", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_grace_1b12b544", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.16151", "paper": "https://arxiv.org/abs/2606.16151", "pdf": "https://arxiv.org/pdf/2606.16151", "project": null, "report": "https://arxiv.org/abs/2606.16151"}, "metrics": [], "motivation": "Many reasoning tasks require models to reason over input context, from document-grounded question answering to rule-based deduction.", "name": "GRACE", "oneLine": "GRACE evaluates step-level faithfulness of chain-of-thought reasoning in context-grounded tasks, with human annotations and a taxonomy of error categories.", "paperTitle": "GRACE: Step-Level Benchmark for Faithful Reasoning over Context", "primaryDomain": "General AI", "publication": {"evidence": "Accepted at EMNLP Main 2026", "evidenceLevel": "author-claim", "evidenceUrl": "https://arxiv.org/abs/2606.16151", "source": "arxiv-comments", "status": "acceptance_claimed", "venue": "EMNLP Main 2026", "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-15", "source": {"authors": ["Hoang Pham", "Dong Le", "Anh Tuan Luu"], "categories": ["cs.CL"], "id": "2606.16151", "title": "GRACE: Step-Level Benchmark for Faithful Reasoning over Context", "type": "arxiv", "url": "https://arxiv.org/abs/2606.16151"}, "sourceUpdatedAt": "2026-06-15T03:11:23Z", "topics": ["Reasoning"], "venueAttempts": [{"decisionRaw": "Accepted at EMNLP Main 2026", "evidence": [{"level": "author-claim", "observedAt": "2026-10-14T01:46:58.058199Z", "rawValue": "Accepted at EMNLP Main 2026", "sourceType": "arxiv-comments", "sourceUrl": "https://arxiv.org/abs/2606.16151"}], "reviewStatus": "accepted", "venueName": "EMNLP Main 2026"}], "whyItMatters": "Step-level faithfulness assessment addresses the gap where response-level metrics miss localized reasoning failures, providing granular feedback for improving model reliability."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
