{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2605.28044", "kind": "paper", "url": "https://arxiv.org/abs/2605.28044"}], "authors": ["Pin Qian", "Su Wang", "Xiaoyuan Wang", "Yihang Chen", "Wenxuan Xu", "Qiaolin Yu", "Shuhuai Lin", "Sipeng Zhang", "Junxian You", "Xinpeng Wei"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2605.28044", "source_url": "https://arxiv.org/abs/2605.28044"}, "categories": ["General AI", "Language & Knowledge", "cs.AI"], "description": {"en": "FORCEBENCH is a contrastive stress test for evidence-force calibration in cited RAG. It pairs fixed cited passages with evidence-calibrated claims and force-raised variants across five axes: relation, modality, scope, temporal validity, and numeric specificity. Evaluation is a fixed, locality-filtered set of 198 pairs."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2605.28044", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2605.28044", "title": "FORCEBENCH"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2605.28044", "modality": null, "name": "FORCEBENCH", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "Medium", "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_forcebench_de197b73", "origin_source": "arxiv", "recognition_confidence": "0.85", "record_sha256": "b646dbf1772e9a5ffbf2e83e2ec328aecec09f6e73a415c7bbccad03b8d91e82", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2605.28044"}, "publisher": null, "released": "2026-05-27", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2605.28044", "source_url": "https://arxiv.org/abs/2605.28044"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2605-28044", "source": "claire_radar", "source_benchmark_id": "2605.28044", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Mixed", "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": null, "hfPaperUrl": "https://huggingface.co/papers/2605.28044", "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", "state": "unavailable"}}, "status": "unavailable"}, "capabilities": [], "confidence": "Medium", "construction": "Aggregate Existing", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "ee77919ebc271f871f95c08b31ee3200e7432ef59af1bf20dd9ed5cf8c43053e", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "FORCEBENCH is a contrastive stress test for evidence-force calibration in cited RAG. It pairs fixed cited passages with evidence-calibrated claims and force-raised variants across five axes: relation, modality, scope, temporal validity, and numeric specificity. Evaluation is a fixed, locality-filtered set of 198 pairs.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "benchmark release stated in one sentence"], "snippet": "We introduce FORCEBENCH, a contrastive stress test for evidence-force calibration."}, "familyId": "bmf_3be86d47de12", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_forcebench_de197b73", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2605.28044", "paper": "https://arxiv.org/abs/2605.28044", "pdf": "https://arxiv.org/pdf/2605.28044", "project": null, "report": "https://arxiv.org/abs/2605.28044"}, "metrics": [], "motivation": "Cited RAG evaluation often treats visible sources as a grounding signal, but a real, topically relevant citation can still under-warrant the attached wording.", "name": "FORCEBENCH", "oneLine": "FORCEBENCH is a contrastive stress test for evidence-force calibration in cited RAG. It pairs fixed cited passages with evidence-calibrated claims and force-raised variants across five axes: relation, modality, scope, temporal validity, and numeric specificity. Evaluation is a fixed, locality-filtered set of 198 pairs.", "paperTitle": "Relevant Is Not Warranted: Evidence-Force Calibration for Cited RAG", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2605.28044", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 0.85, "relation": "introduces", "releasedAt": "2026-05-27", "source": {"authors": ["Pin Qian", "Su Wang", "Xiaoyuan Wang", "Yihang Chen", "Wenxuan Xu", "Qiaolin Yu", "Shuhuai Lin", "Sipeng Zhang", "Junxian You", "Xinpeng Wei"], "categories": ["cs.AI"], "id": "2605.28044", "title": "Relevant Is Not Warranted: Evidence-Force Calibration for Cited RAG", "type": "arxiv", "url": "https://arxiv.org/abs/2605.28044"}, "sourceUpdatedAt": "2026-05-27T06:47:44Z", "topics": ["cs.AI"], "venueAttempts": [], "whyItMatters": "Current cited RAG evaluation often treats topical relevance as sufficient grounding, missing cases where a relevant source under-warrants an over-strong claim. FORCEBENCH exposes this citation laundering failure and measures evaluator calibration via monotonicity violation rate and force sensitivity."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
