{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.22000", "kind": "paper", "url": "https://arxiv.org/abs/2606.22000"}], "authors": ["Rishi Srivastava"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.22000", "source_url": "https://arxiv.org/abs/2606.22000"}, "categories": ["Agents", "Agents & Tool Use", "Finance"], "description": {"en": "CFAgentBench is an executable environment for autonomous construction-finance agents, with 1,014 task specifications across 8 domains and 77 families. A subset of 40 tasks (54 with PM extension) has oracle-validated evaluators. Grading uses state diffs, forbidden-side-effect checks, and required-output regexes, with an LLM judge only for reply quality. A public split of 711 tasks is available, and a private split of 303 is reserved for remote scoring."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.22000", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.22000", "title": "CFAgentBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.22000", "modality": null, "name": "CFAgentBench", "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_cfagentbench_9ccf4efc", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "f82c19969e03978581d5e6cfb9225018d92f6cc317c176309106a425ee598b56", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.22000"}, "publisher": null, "released": "2026-06-20", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.22000", "source_url": "https://arxiv.org/abs/2606.22000"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-22000", "source": "claire_radar", "source_benchmark_id": "2606.22000", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Finance"], "area": "Agents & Tool Use", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": "https://huggingface.co/papers/2606.22000", "observedAt": "2026-10-04T03:02:42.754940Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "state": "unavailable"}}, "status": "unavailable"}, "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": "714454a11dd9436cd565558f81767e20878f9b940b2058727bb699e86916ff94", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "CFAgentBench is an executable environment for autonomous construction-finance agents, with 1,014 task specifications across 8 domains and 77 families. A subset of 40 tasks (54 with PM extension) has oracle-validated evaluators. Grading uses state diffs, forbidden-side-effect checks, and required-output regexes, with an LLM judge only for reply quality. A public split of 711 tasks is available, and a private split of 303 is reserved for remote scoring.", "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"], "snippet": "We introduce CFAgentBench, a reproducible, self-hostable environment and benchmark for autonomous construction-finance agents: a CFO/controller-class agent operating across the real software stack a US construction finance team runs - ERP, project management, email, documents, pay applications, payroll, certified payroll, lien waivers, and bank/treasury portals."}, "familyId": "bmf_b402658f28ba", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_cfagentbench_9ccf4efc", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Financial Services"], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.22000", "paper": "https://arxiv.org/abs/2606.22000", "pdf": "https://arxiv.org/pdf/2606.22000", "project": null, "report": "https://arxiv.org/abs/2606.22000"}, "metrics": [], "motivation": "We introduce CFAgentBench, a reproducible, self-hostable environment and benchmark for autonomous construction-finance agents: a CFO/controller-class agent operating across the real software stack a US construction finance team runs - ERP, project management, email, documents, pay applications, payroll, certified payroll, lien waivers, and bank/treasury portals.", "name": "CFAgentBench", "oneLine": "CFAgentBench is an executable environment for autonomous construction-finance agents, with 1,014 task specifications across 8 domains and 77 families. A subset of 40 tasks (54 with PM extension) has oracle-validated evaluators. Grading uses state diffs, forbidden-side-effect checks, and required-output regexes, with an LLM judge only for reply quality. A public split of 711 tasks is available, and a private split of 303 is reserved for remote scoring.", "paperTitle": "CFAgentBench: A Reproducible Environment and Benchmark for Autonomous Construction-Finance Agents", "primaryDomain": "Finance", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.22000", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "CFAgentBench Team", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://arxiv.org/abs/2606.22000"}], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-20", "source": {"authors": ["Rishi Srivastava"], "categories": ["cs.AI", "cs.CL", "cs.LG", "cs.SE"], "id": "2606.22000", "title": "CFAgentBench: A Reproducible Environment and Benchmark for Autonomous Construction-Finance Agents", "type": "arxiv", "url": "https://arxiv.org/abs/2606.22000"}, "sourceUpdatedAt": "2026-06-20T11:34:52Z", "topics": ["Agents"], "venueAttempts": [], "whyItMatters": "The benchmark addresses the gap in evaluating agents for finance workflows that involve real software stacks and high-stakes transactions. Its focus on functional correctness and a money-movement guard, where correct actions can fail tasks, provides a more realistic measure of deployable competence. The observed performance collapse under repeated attempts highlights the need for reliability assessment beyond single-attempt accuracy."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
