{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2605.27922", "kind": "paper", "url": "https://arxiv.org/abs/2605.27922"}], "authors": ["Yilun Yao", "Xinyu Tan", "Chao-Hsuan Liu", "Yaoming Li", "Zhengyang Wang", "Zongwei Lv", "Wenhan Yu", "Zhewen Tan", "Yuxuan Tian", "Lin Sun", "Xiangzheng Zhang", "Guangxiang Zhao", "Tong Yang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2605.27922", "source_url": "https://arxiv.org/abs/2605.27922"}, "categories": ["Agents", "General AI", "Language & Knowledge"], "description": {"en": "Harness-Bench evaluates configuration-level harness effects in agent workflows with 106 sandboxed tasks, measuring completion, process quality, efficiency, and failure behavior across model-harness pairings."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2605.27922", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2605.27922", "title": "Harness-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Peking University"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2605.27922", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2605.27922", "source_url": "https://openalex.org/W7162690279"}, "key": "claire-radar:2605.27922", "modality": null, "name": "Harness-Bench", "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_harness-bench_b6a94bb9", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "641bc396a259e637d7ff94936bb7e65b124b8580cfb040baa31bb9a9e525c1c6", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2605.27922"}, "publisher": null, "released": "2026-05-27", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2605.27922", "source_url": "https://arxiv.org/abs/2605.27922"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2605-27922", "source": "claire_radar", "source_benchmark_id": "2605.27922", "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": 1, "hfPaperUrl": "https://huggingface.co/papers/2605.27922", "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": "2026-09-30", "state": "stale"}}, "status": "stale"}, "capabilities": [], "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": "9ebfc8b12156538dfa2d5e20681e9fc36cd36c0044c08f436d2d6be2156ac499", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Harness-Bench evaluates configuration-level harness effects in agent workflows with 106 sandboxed tasks, measuring completion, process quality, efficiency, and failure behavior across model-harness pairings.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "We introduce Harness-Bench, a diagnostic benchmark for evaluating configuration-level harness effects in realistic agent workflows."}, "familyId": "bmf_5d1a38309c97", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_harness-bench_b6a94bb9", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2605.27922", "paper": "https://arxiv.org/abs/2605.27922", "pdf": "https://arxiv.org/pdf/2605.27922", "project": null, "report": "https://arxiv.org/abs/2605.27922"}, "metrics": [], "motivation": "LLM agents are increasingly deployed as executable systems that use tools, modify workspaces, and produce concrete artifacts.", "name": "Harness-Bench", "oneLine": "Harness-Bench evaluates configuration-level harness effects in agent workflows with 106 sandboxed tasks, measuring completion, process quality, efficiency, and failure behavior across model-harness pairings.", "paperTitle": "Harness-Bench: Measuring Harness Effects across Models in Realistic Agent Workflows", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2605.27922", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-27", "source": {"authors": ["Yilun Yao", "Xinyu Tan", "Chao-Hsuan Liu", "Yaoming Li", "Zhengyang Wang", "Wenhan Yu", "Zhewen Tan", "Yuxuan Tian", "Guangxiang Zhao", "Lin Sun", "Xiangzheng Zhang", "and Tong Yang"], "categories": ["cs.AI"], "id": "2605.27922", "title": "Harness-Bench: Measuring Harness Effects across Models in Realistic Agent Workflows", "type": "arxiv", "url": "https://arxiv.org/abs/2605.27922"}, "sourceUpdatedAt": "2026-05-27T03:47:35Z", "topics": ["Agents"], "venueAttempts": [], "whyItMatters": "Addresses a gap in agent evaluation by isolating harness configuration effects, showing that agent capability is configuration-level rather than model-only, and identifying execution-alignment failures."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
