{"record": {"aliases": ["HANDBOOK.md: A Benchmark for Long-Context Agentic Instruction Following"], "artifacts": [{"id": "arxiv:2607.25398", "kind": "paper", "url": "https://arxiv.org/abs/2607.25398"}, {"id": "gh:surge-ai/handbook", "kind": "repo", "url": "https://github.com/surge-ai/handbook"}], "authors": ["Liudas Panavas", "Sebastian Minus", "Bradley Monton", "Derek Ray", "Suhaas Garre", "Sushant Mehta", "Edwin Chen"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.25398", "source_url": "https://arxiv.org/abs/2607.25398"}, "categories": ["Biology & Drug Discovery", "Finance", "Language & Knowledge", "Logistics & Operations", "Long Context"], "description": {"en": "HANDBOOK.md evaluates whether tool-using agents consistently obey lengthy standing procedures while completing workplace tasks in self-contained company environments."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.25398", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.25398", "title": "HANDBOOK.md"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2607.25398", "modality": null, "name": "HANDBOOK.md", "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": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_handbook-md-a-benchmark-for-long-context-a_6a90f2ca", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "2ee9fdf1be71d16562f0e38e7e66af2b2b5ecdfaadbb395cb2d25a11052f8015", "relation": "introduces", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T17:32:36.949824Z", "source_url": "https://arxiv.org/abs/2607.25398"}, "publisher": null, "released": "2026-07-28", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2607.25398", "source_url": "https://arxiv.org/abs/2607.25398"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-25398", "source": "claire_radar", "source_benchmark_id": "2607.25398", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["HANDBOOK.md: A Benchmark for Long-Context Agentic Instruction Following"], "annotation": "Unknown", "applicationDomains": ["Biology & Drug Discovery", "Logistics & Operations", "Finance"], "area": "Language & Knowledge", "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "Long-context agent compliance, deterministic grading, and a released MCP-based harness are likely to draw near-term evaluation interest.", "score": 76}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T17:32:36.949824Z", "inputHash": "06e479f10a7d4bb7d9978b44da08482ce660ff01505905640454a39b69d6f7a4", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "HANDBOOK.md: A Benchmark for Long-Context Agentic Instruction Following", "canonicalNameSource": "paper_title", "decisionReason": "The supplied evidence establishes a formal benchmark name, defined tasks, deterministic strict-pass grading, and a public release of the tasks, environments, and harness.", "evaluationEvidence": {"reuse": "We release the tasks, environments, and evaluation harness.", "scoring": "a trial passes only if every criterion is satisfied", "task": "a benchmark of 65 agentic tasks modeled on how employees follow company handbooks"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T17:32:36.949824Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "HANDBOOK.md evaluates whether tool-using agents consistently obey lengthy standing procedures while completing workplace tasks in self-contained company environments.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We present HANDBOOK_md, a benchmark of 65 agentic tasks modeled on how employees follow company handbooks."}, "familyId": "bmf_338866fccaa9", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_handbook-md-a-benchmark-for-long-context-a_6a90f2ca", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Pharma & Biotech", "Logistics", "Financial Services"], "links": {"code": "https://github.com/surge-ai/handbook", "data": null, "paper": "https://arxiv.org/abs/2607.25398", "pdf": "https://arxiv.org/pdf/2607.25398", "project": null, "report": "https://arxiv.org/abs/2607.25398"}, "metrics": [], "motivation": "Language-model agents are increasingly deployed under standing instructions: a system prompt, a policy file, or a skills document is placed in context, and the agent is trusted to let that document govern every action that follows.", "name": "HANDBOOK.md", "oneLine": "HANDBOOK.md evaluates whether tool-using agents consistently obey lengthy standing procedures while completing workplace tasks in self-contained company environments.", "paperTitle": "HANDBOOK.md: A Benchmark for Long-Context Agentic Instruction Following", "primaryDomain": "Biology & Drug Discovery", "publication": {"evidence": "16 pages, 3 figures, 5 tables. Accepted to the Workshop on Agent Behavior (WAB) at COLM 2026. Benchmark, environments, and evaluation harness: https://github.com/surge-ai/handbook", "evidenceLevel": "author-claim", "evidenceUrl": "https://arxiv.org/abs/2607.25398", "source": "arxiv-comments", "status": "acceptance_claimed", "venue": "Workshop on Agent Behavior (WAB) at COLM 2026", "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "publishers": [{"name": "Surge AI", "organizationType": "company-research-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/surge-ai/handbook"}], "readiness": "Runnable", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-07-28", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Liudas Panavas", "Sebastian Minus", "Bradley Monton", "Derek Ray", "Suhaas Garre", "Sushant Mehta", "Edwin Chen"], "categories": ["cs.AI", "cs.CL"], "id": "2607.25398", "title": "HANDBOOK.md: A Benchmark for Long-Context Agentic Instruction Following", "type": "arxiv", "url": "https://arxiv.org/abs/2607.25398"}, "sourceUpdatedAt": "2026-08-03T20:36:31Z", "topics": ["Long Context"], "venueAttempts": [{"decisionRaw": "16 pages, 3 figures, 5 tables. Accepted to the Workshop on Agent Behavior (WAB) at COLM 2026. Benchmark, environments, and evaluation harness: https://github.com/surge-ai/handbook", "evidence": [{"level": "author-claim", "observedAt": "2026-08-19T11:05:13.395391Z", "rawValue": "16 pages, 3 figures, 5 tables. Accepted to the Workshop on Agent Behavior (WAB) at COLM 2026. Benchmark, environments, and evaluation harness: https://github.com/surge-ai/handbook", "sourceType": "arxiv-comments", "sourceUrl": "https://arxiv.org/abs/2607.25398"}], "reviewStatus": "accepted", "venueName": "Workshop on Agent Behavior (WAB) at COLM 2026"}], "whyItMatters": "It tests persistent policy compliance over long tool-use horizons and uses deterministic criteria covering both required and prohibited actions."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
