{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2608.04003", "kind": "paper", "url": "https://arxiv.org/abs/2608.04003"}, {"id": "gh:gen-verse/past-bench", "kind": "repo", "url": "https://github.com/Gen-Verse/PAST-Bench"}], "authors": ["Shuhan Xue", "Zixin Ding", "Yichen Shen", "Yinjie Wang", "Zhenfei Yin", "Yingcheng Wu", "Yuxin Chen", "Mengdi Wang", "Ling Yang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2608.04003", "source_url": "https://arxiv.org/abs/2608.04003"}, "categories": ["Agents", "General AI", "Language & Knowledge", "Self-Evolution"], "description": {"en": "PAST-Bench evaluates recursive self-improvement in personal AI agents by testing whether retained experience improves performance on future tasks. It spans 26 scenarios and 204 episodes across memory, procedural reuse, information gathering, and update capabilities, with matched persistence on/off controls."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2608.04003", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2608.04003", "title": "PAST-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2608.04003", "modality": null, "name": "PAST-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_past-bench_b04dad58", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "428d4293cc26a24078f26000b732db391cf13b4d9e6f3af8c6af178c9a112718", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2608.04003"}, "publisher": null, "released": "2026-08-04", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2608.04003", "source_url": "https://arxiv.org/abs/2608.04003"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2608-04003", "source": "claire_radar", "source_benchmark_id": "2608.04003", "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/Gen-Verse/PAST-Bench", "githubScope": "benchmark_repo", "githubStars": 33, "hfDailySubmittedAt": "2026-08-05T00:00:00.000Z", "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 36, "hfPaperUrl": "https://huggingface.co/papers/2608.04003", "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": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}}, "status": "fresh"}, "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": "0ef8aa96463f20467ad56ab00c20214e33f953a38a56dd9a61d0b003234a832f", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "PAST-Bench evaluates recursive self-improvement in personal AI agents by testing whether retained experience improves performance on future tasks. It spans 26 scenarios and 204 episodes across memory, procedural reuse, information gathering, and update capabilities, with matched persistence on/off controls.", "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", "public artifact URL"], "snippet": "We introduce PAST-Bench, a benchmark designed to isolate this question."}, "familyId": "bmf_d6508586cac7", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/Gen-Verse/PAST-Bench", "url": "https://github.com/Gen-Verse/PAST-Bench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/Gen-Verse/PAST-Bench", "url": "https://huggingface.co/papers/2608.04003"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/Gen-Verse/PAST-Bench", "url": "https://arxiv.org/abs/2608.04003"}], "scope": "benchmark_repo", "url": "https://github.com/Gen-Verse/PAST-Bench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_past-bench_b04dad58", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/Gen-Verse/PAST-Bench", "data": null, "hfPaper": "https://huggingface.co/papers/2608.04003", "paper": "https://arxiv.org/abs/2608.04003", "pdf": "https://arxiv.org/pdf/2608.04003", "project": null, "report": "https://arxiv.org/abs/2608.04003"}, "metrics": [], "motivation": "Recursive self-improvement requires agents to turn accumulated experience into better future behavior.", "name": "PAST-Bench", "oneLine": "PAST-Bench evaluates recursive self-improvement in personal AI agents by testing whether retained experience improves performance on future tasks. It spans 26 scenarios and 204 episodes across memory, procedural reuse, information gathering, and update capabilities, with matched persistence on/off controls.", "paperTitle": "PAST-Bench: Benchmarking the Foundations of Recursive Self-Improvement in Personal Agents", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2608.04003", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "Gen-Verse", "organizationType": "community", "role": "benchmark-publisher", "sourceUrl": "https://github.com/Gen-Verse/PAST-Bench"}], "ranking": {"90d": {"confidence": "Medium", "coverage": 0.7, "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": 0.4707057397524185, "value": 33}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.6380297304564572, "value": 36}}, "confidence": "Medium", "coverage": 0.7, "rank": 60, "score": 50}, "method": "attention-ranking-v13", "rank": 60, "score": 50}}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-08-04", "source": {"authors": ["Shuhan Xue", "Zixin Ding", "Yichen Shen", "Yinjie Wang", "Zhenfei Yin", "Yingcheng Wu", "Yuxin Chen", "Mengdi Wang", "Ling Yang"], "categories": ["cs.CL"], "id": "2608.04003", "title": "PAST-Bench: Benchmarking the Foundations of Recursive Self-Improvement in Personal Agents", "type": "arxiv", "url": "https://arxiv.org/abs/2608.04003"}, "sourceUpdatedAt": "2026-08-04T17:58:05Z", "topics": ["Self-Evolution", "Agents"], "venueAttempts": [], "whyItMatters": "Whether personal agents actually improve from retained experience has not been systematically tested. PAST-Bench provides a controlled benchmark to measure and attribute cross-session improvement, distinguishing capability gains from pathway evidence."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
