{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2608.19799", "kind": "paper", "url": "https://arxiv.org/abs/2608.19799"}], "authors": ["Zhipeng Xu", "Jiahao Lu", "Yining Zheng", "Yuxin Wang", "Xipeng Qiu"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2608.19799", "source_url": "https://arxiv.org/abs/2608.19799"}, "categories": ["Code", "Code & Software", "General AI", "Software & AI Compute"], "description": {"en": "A repository-level benchmark with 119 tasks from 98 GitHub repositories across 20 scientific domains, organized into issue-driven, expert-exploratory, and engineering-integration paradigms."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2608.19799", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2608.19799", "title": "SWE-bench Science"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Fudan University", "Shanghai Innovation Institute"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2608.19799", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2608.19799", "source_url": "https://openalex.org/W7203880176"}, "key": "claire-radar:2608.19799", "modality": null, "name": "SWE-bench Science", "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": "true", "origin_record_id": "bm_swe-bench_536d758e", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "c85c04ae4adc8b66cb3732230ca1ad7e2bff4fdb9b675d0531c6937f50078f25", "relation": "introduces", "review_model": "deepseek-v4-pro", "review_state": "ai-reviewed", "reviewed_at": "2026-08-25T17:43:56.718736Z", "source_url": "https://arxiv.org/abs/2608.19799"}, "publisher": null, "released": "2026-08-20", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2608.19799", "source_url": "https://arxiv.org/abs/2608.19799"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2608-19799", "source": "claire_radar", "source_benchmark_id": "2608.19799", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Mixed", "applicationDomains": ["General AI", "Software & AI Compute"], "area": "Code & Software", "attention": {"asOf": "2026-08-26", "attemptedAt": "2026-08-25T16:48:55.714492Z", "githubRepo": "https://github.com/OpenMOSS/SWE-bench-Science", "githubScope": "benchmark_repo", "githubStars": 61, "hfDailySubmittedAt": "2026-08-21T00:00:00.000Z", "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 62, "hfPaperUrl": "https://huggingface.co/papers/2608.19799", "observedAt": "2026-08-25T16:48:55.714492Z", "signalStatus": {"githubStars": {"observedAt": "2026-08-25T16:48:55.714492Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"observedAt": "2026-08-25T16:48:55.714492Z", "state": "fresh"}}, "status": "fresh"}, "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "The benchmark extends the popular SWE-bench framework to scientific domains, addressing a critical gap with strong empirical findings.", "score": 85}, "capabilities": [], "confidence": "High", "construction": "Aggregate Existing", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-25T17:43:56.718736Z", "inputHash": "18d9f5c702f9c554acac0e9461d9466d8de63a44923c0faca5e4025fdb22d3c8", "model": "deepseek-v4-pro"}, "curation": {"canonicalNameEvidence": "SWE-bench Science: Can Coding Agents Resolve Engineering Tasks in Science?", "canonicalNameSource": "paper_title", "decisionReason": "The paper introduces a formal benchmark with a clear task structure and public repository data, enabling community reuse for evaluating coding agents in science.", "model": "deepseek-v4-pro", "reviewedAt": "2026-08-25T17:43:56.718736Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A repository-level benchmark with 119 tasks from 98 GitHub repositories across 20 scientific domains, organized into issue-driven, expert-exploratory, and engineering-integration paradigms.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "We introduce \\textbf{SWE-bench Science}, a repository-level benchmark for scientific software engineering comprising 119 tasks from 98 GitHub repositories across 20 scientific domains."}, "familyId": "bmf_c0124d8dc037", "firstSeenAt": "2026-08-21", "heat": null, "id": "bm_swe-bench_536d758e", "indexedAt": "2026-08-21T04:30:09.569171Z", "industrySectors": ["Software & Cloud"], "links": {"code": null, "data": null, "paper": "https://arxiv.org/abs/2608.19799", "pdf": "https://arxiv.org/pdf/2608.19799", "project": null, "report": "https://arxiv.org/abs/2608.19799"}, "metrics": [], "motivation": "Software increasingly functions as part of the scientific instrument itself, making failures in scientific code capable of compromising not only program behavior but also the evidence underlying scientific conclusions.", "name": "SWE-bench Science", "oneLine": "A repository-level benchmark with 119 tasks from 98 GitHub repositories across 20 scientific domains, organized into issue-driven, expert-exploratory, and engineering-integration paradigms.", "paperTitle": "SWE-bench Science: Can Coding Agents Resolve Engineering Tasks in Science?", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2608.19799", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-21T04:30:09.569171Z"}, "publications": [], "ranking": {"30d": {"confidence": "High", "coverage": 0.85, "growth": {"components": {"githubStars": {"percentile": null, "value": null}, "hfDatasetDownloads": {"percentile": null, "value": null}, "hfPaperUpvotes": {"percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 30}, "level": {"components": {"githubStars": {"percentile": 0.9548872180451128, "value": 61}, "hfDatasetDownloads": {"percentile": null, "value": null}, "hfPaperUpvotes": {"percentile": 0.9747474747474747, "value": 62}}, "confidence": "High", "coverage": 0.85, "rank": 5, "score": 89}, "method": "attention-ranking-v11", "rank": 5, "score": 89}, "90d": {"confidence": "Medium", "coverage": 0.7, "growth": {"components": {"githubStars": {"percentile": null, "value": null}, "hfDatasetDownloads": {"percentile": null, "value": null}, "hfPaperUpvotes": {"percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"percentile": 0.9025069637883009, "value": 61}, "hfDatasetDownloads": {"percentile": null, "value": null}, "hfPaperUpvotes": {"percentile": 0.9630952380952381, "value": 62}}, "confidence": "Medium", "coverage": 0.7, "rank": 24, "score": 79}, "method": "attention-ranking-v11", "rank": 24, "score": 79}}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-08-20", "source": {"authors": ["Zhipeng Xu", "Jiahao Lu", "Yining Zheng", "Yuxin Wang", "Xipeng Qiu"], "categories": ["cs.CL", "cs.SE"], "id": "2608.19799", "title": "SWE-bench Science: Can Coding Agents Resolve Engineering Tasks in Science?", "type": "arxiv", "url": "https://arxiv.org/abs/2608.19799"}, "sourceUpdatedAt": "2026-08-20T08:53:15Z", "topics": ["Code"], "venueAttempts": [], "whyItMatters": "Evaluates coding agents on scientific software failures, identifying failure modes and the nuanced role of scientific knowledge in repair tasks."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
