{"record": {"aliases": ["failure-transparent-agents"], "artifacts": [{"id": "gh:junru-zhu/failure-transparent-agents", "kind": "repo", "url": "https://github.com/junru-zhu/failure-transparent-agents"}], "categories": ["Language & Knowledge", "Software & AI Compute", "cs.AI"], "description": {"en": "Measures whether LLM agents admit tool failures instead of claiming success across 100 tasks covering web, attachment, execution, permission, and stale-data failures under varying pressure and instruction conditions."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://github.com/junru-zhu/failure-transparent-agents", "source": "claire_radar", "source_url": "https://github.com/junru-zhu/failure-transparent-agents", "title": "Failure-Transparent Agents"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:github:junru-zhu/failure-transparent-agents", "modality": null, "name": "Failure-Transparent Agents", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-candidate", "display_eligible": "true", "origin_record_id": "bm_failure-transparent-agents_1c58abe3", "origin_source": "github", "recognition_confidence": "0.4", "record_sha256": "5278b3919c91947c94d0f3c762f8b07a5ed4a1f077b127d1530d7ecf06571eeb", "relation": "unclear", "review_model": "deepseek-v4-pro", "review_state": "ai-reviewed", "reviewed_at": "2026-09-11T09:00:09.423366Z", "source_url": "https://github.com/junru-zhu/failure-transparent-agents"}, "publisher": null, "released": "2026-09-10", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:github:junru-zhu/failure-transparent-agents", "source_url": "https://github.com/junru-zhu/failure-transparent-agents"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-github-junru-zhu-failure-transparent-agents", "source": "claire_radar", "source_benchmark_id": "github:junru-zhu/failure-transparent-agents", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["failure-transparent-agents"], "annotation": "Unknown", "applicationDomains": ["Software & AI Compute"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/junru-zhu/failure-transparent-agents", "githubScope": "benchmark_repo", "githubStars": 1, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": null, "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": {"state": "not_applicable"}}, "status": "fresh"}, "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "The benchmark addresses a focused but timely reliability gap in LLM tool use, with limited initial audience but clear utility for agent builders and safety teams.", "score": 45}, "capabilities": [], "confidence": "Medium", "construction": "Interactive Environment", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-11T09:00:09.423366Z", "inputHash": "97565fe740761fd7131728fee9cfac7291a475f72478967d0ff58fb5c716c68a", "model": "deepseek-v4-pro"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "Junru Zhu. Failure-Transparent Agents: A Reproducible Benchmark of Post-Failure Response Transparency. Version 0.2.0, 2026.", "canonicalNameSource": "official_readme", "decisionReason": "The repository includes a frozen 100-task benchmark, deterministic simulator, resumable harness, and public instructions for evaluating other models, satisfying a reusable evaluation protocol.", "evaluationEvidence": {"reuse": "The harness supports frozen provider configs, hard call caps, estimated cost, retries, resume, request provenance, token counts, and latency.", "scoring": "The evidence contract requires the model to expose four fields: STATUS EVIDENCE LIMITATION NEXT_ACTION .", "task": "Each task contains the unavailable evidence, deterministic failed observation, planted unsupported claims, feasible recovery, and safe partial help."}, "model": "deepseek-v4-pro", "reviewedAt": "2026-09-11T09:00:09.423366Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-candidate", "demo": false, "description": "Measures whether LLM agents admit tool failures instead of claiming success across 100 tasks covering web, attachment, execution, permission, and stale-data failures under varying pressure and instruction conditions.", "discoverySources": [{"id": "github:junru-zhu/failure-transparent-agents", "type": "github", "url": "https://github.com/junru-zhu/failure-transparent-agents"}], "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["discovered via github", "benchmark term in abstract", "evaluation protocol evidence", "public artifact URL", "no explicit benchmark release evidence"], "snippet": "ai-agents ai-evaluation benchmark guardrails hallucination llm reproducibility tool-use When a required tool fails, does the model admit it—or claim success anyway?"}, "familyId": "bmf_81145aeacf9d", "firstSeenAt": "2026-09-11", "heat": null, "id": "bm_failure-transparent-agents_1c58abe3", "indexedAt": "2026-09-11T06:20:17.513010Z", "industrySectors": ["Software & Cloud"], "links": {"code": "https://github.com/junru-zhu/failure-transparent-agents"}, "metrics": [], "motivation": "failure-transparent-agents Benchmarking whether LLM agents admit tool failure instead of hallucinating success — 100 tasks, 1,800 responses, deterministic simulator, reproducible harness.", "name": "Failure-Transparent Agents", "oneLine": "Measures whether LLM agents admit tool failures instead of claiming success across 100 tasks covering web, attachment, execution, permission, and stale-data failures under varying pressure and instruction conditions.", "paperTitle": "failure-transparent-agents", "primaryDomain": "Software & AI Compute", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://github.com/junru-zhu/failure-transparent-agents", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-09-11T06:20:17.513010Z"}, "publications": [], "publishers": [{"name": "Junru Zhu", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/junru-zhu/failure-transparent-agents"}], "ranking": {"30d": {"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": 30, "historyDate": "2026-09-05", "rank": null, "score": null, "windowDays": 30}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 0.10220105459060908, "value": 1}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": null, "value": null}}, "confidence": "Low", "coverage": 0.55, "rank": 108, "score": 28}, "method": "attention-ranking-v13", "rank": 108, "score": 28}, "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": 0.09252268850709358, "value": 1}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": null, "value": null}}, "confidence": "Low", "coverage": 0.55, "rank": 315, "score": 28}, "method": "attention-ranking-v13", "rank": 315, "score": 28}}, "readiness": "Runnable", "recognitionConfidence": 0.4, "relation": "unclear", "releasedAt": "2026-09-10", "source": {"authors": ["junru-zhu"], "categories": [], "id": "github:junru-zhu/failure-transparent-agents", "publicSignals": {"githubStars": 1}, "title": "failure-transparent-agents", "type": "github", "url": "https://github.com/junru-zhu/failure-transparent-agents"}, "sourceUpdatedAt": "2026-09-11T03:35:01Z", "topics": ["cs.AI"], "venueAttempts": [], "whyItMatters": "Provides a reproducible way to evaluate failure transparency, helping prevent hallucinated success claims in tool-using agents and supporting safer deployment and regression testing."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
