{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2609.05079", "kind": "paper", "url": "https://arxiv.org/abs/2609.05079"}, {"id": "gh:truthinsight-stack/truthinsightbench", "kind": "repo", "url": "https://github.com/TruthInsight-stack/TruthInsightBench"}], "authors": ["Zhibo Yang", "Chen Zhang", "Yuewei Zhang", "Hao Wang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2609.05079", "source_url": "https://arxiv.org/abs/2609.05079"}, "categories": ["AI Scientist", "General AI", "Language & Knowledge"], "description": {"en": "TruthInsightBench evaluates open-ended scientific discovery agents on 40 blind tasks across 10 scientific domains. Each task presents a neutral research objective with frozen data while source conclusions are withheld. A fixed LLM-based judge scores agents' claims along six dimensions with 29 artifact-grounded items, and automated aggregation produces overall scores. Data and scoring code are publicly available."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2609.05079", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2609.05079", "title": "TruthInsightBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2609.05079", "modality": null, "name": "TruthInsightBench", "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_truthinsightbench_fec89a32", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "e774516e53a2577c824e2a4d5672664efb6c7aa1ff65f28824ca475d62ca1f58", "relation": "introduces", "review_model": "deepseek-v4-pro", "review_state": "ai-reviewed", "reviewed_at": "2026-09-08T00:56:29.478814Z", "source_url": "https://arxiv.org/abs/2609.05079"}, "publisher": null, "released": "2026-09-04", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2609.05079", "source_url": "https://arxiv.org/abs/2609.05079"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2609-05079", "source": "claire_radar", "source_benchmark_id": "2609.05079", "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": "https://github.com/TruthInsight-stack/TruthInsightBench", "githubScope": "benchmark_repo", "githubStars": 3, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": "https://huggingface.co/papers/2609.05079", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "state": "unavailable"}}, "status": "fresh"}, "attentionForecast": {"confidence": "Low", "horizon": "7d", "reason": "The topic of automated scientific discovery is broad and the benchmark provides code and data, but 7-day traction is uncertain pending community uptake.", "score": 60}, "capabilities": [], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-08T08:59:06.803818Z", "inputHash": "6cafe93c6895b0f4d7c640f49fede7e4c617625cbb162834316c8b938c3f16fa", "model": "deepseek-v4-flash"}, "curation": {"canonicalNameEvidence": "We present TruthInsightBench, a benchmark configured for discovery.", "canonicalNameSource": "abstract", "decisionReason": "Released with complete task data, scoring code, and detailed instructions for reproduction; the paper defines a clear, repeatable evaluation protocol.", "model": "deepseek-v4-pro", "reviewedAt": "2026-09-08T00:56:29.478814Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "TruthInsightBench evaluates open-ended scientific discovery agents on 40 blind tasks across 10 scientific domains. Each task presents a neutral research objective with frozen data while source conclusions are withheld. A fixed LLM-based judge scores agents' claims along six dimensions with 29 artifact-grounded items, and automated aggregation produces overall scores. Data and scoring code are publicly available.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "score_submission", "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We present TruthInsightBench, a benchmark configured for discovery."}, "familyId": "bmf_a7bb291973e4", "firstSeenAt": "2026-09-07", "heat": null, "id": "bm_truthinsightbench_fec89a32", "indexedAt": "2026-09-07T06:22:02.605340Z", "industrySectors": [], "links": {"code": "https://github.com/TruthInsight-stack/TruthInsightBench", "data": null, "paper": "https://arxiv.org/abs/2609.05079", "pdf": "https://arxiv.org/pdf/2609.05079", "project": null, "report": "https://arxiv.org/abs/2609.05079"}, "metrics": [], "motivation": "Autonomous coding agents are increasingly proposed as AI-scientist systems that conduct analyses and write research reports, but executing a prescribed analysis is not the same as making a discovery.", "name": "TruthInsightBench", "oneLine": "TruthInsightBench evaluates open-ended scientific discovery agents on 40 blind tasks across 10 scientific domains. Each task presents a neutral research objective with frozen data while source conclusions are withheld. A fixed LLM-based judge scores agents' claims along six dimensions with 29 artifact-grounded items, and automated aggregation produces overall scores. Data and scoring code are publicly available.", "paperTitle": "TruthInsightBench: An Evidence-Grounded Benchmark for Automated Evaluation of Open-Ended Scientific Discovery Agents", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2609.05079", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-09-07T06:22:02.605340Z"}, "publications": [], "publishers": [{"name": "TruthInsight-stack", "organizationType": "community", "role": "benchmark-publisher", "sourceUrl": "https://github.com/TruthInsight-stack/TruthInsightBench"}], "ranking": {"90d": {"confidence": "Low", "coverage": 0.55, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 0.18504537701418716, "value": 3}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": null, "value": null}}, "confidence": "Low", "coverage": 0.55, "rank": 222, "score": 33}, "method": "attention-ranking-v13", "rank": 222, "score": 33}}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-09-04", "source": {"authors": ["Zhibo Yang", "Chen Zhang", "Yuewei Zhang", "Hao Wang"], "categories": ["cs.AI", "cs.CL"], "id": "2609.05079", "title": "TruthInsightBench: An Evidence-Grounded Benchmark for Automated Evaluation of Open-Ended Scientific Discovery Agents", "type": "arxiv", "url": "https://arxiv.org/abs/2609.05079"}, "sourceUpdatedAt": "2026-09-04T12:37:20Z", "topics": ["AI Scientist"], "venueAttempts": [], "whyItMatters": "Existing benchmark evaluation often rewards reproducing a hidden study rather than genuine discovery. TruthInsightBench measures whether agents can derive trustworthy scientific claims without seeing expected outcomes, filling a gap in evaluating AI-scientist systems for authentic discovery and making the bottleneck of scientific judgment measurable."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
