{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.29955", "kind": "paper", "url": "https://arxiv.org/abs/2606.29955"}, {"id": "gh:ruckbreasoning/spreadsheetbench-2", "kind": "repo", "url": "https://github.com/RUCKBReasoning/SpreadsheetBench-2"}, {"kind": "website", "url": "https://spreadsheetbench.github.io/"}], "authors": ["Jian Zhu", "Yuzheng Zhang", "Zeyao Ma", "Bohan Zhang", "Armin Schoepf", "Daniel Woloch", "Peter Yiliu Wang", "Guangyu Robert Yang", "Samuel Jacob", "Siddharth Nagisetty", "Abhiram Chundru", "Jean Lin", "Spencer Mateega", "Jing Zhang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.29955", "source_url": "https://arxiv.org/abs/2606.29955"}, "categories": ["Code & Software", "Finance", "cs.SE"], "description": {"en": "SpreadsheetBench 2 evaluates spreadsheet agents on end-to-end business workflows across generation, debugging, and visualization tasks. It includes 321 tasks from authentic business data, with multi-sheet workbooks requiring cross-sheet reasoning. The benchmark provides a unified multi-turn agent scaffold and evaluation scripts for reproducibility."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.29955", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.29955", "title": "SpreadsheetBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Renmin University of China"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.29955", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.29955", "source_url": "https://openalex.org/W7166696629"}, "key": "claire-radar:2606.29955", "modality": null, "name": "SpreadsheetBench", "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_spreadsheetbench_a0b75616", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "7ef17d04b7e72f4f43bbe2866fdd26d3cd919d27641ab45bf995232f0ad98aa2", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.29955"}, "publisher": null, "released": "2026-06-29", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.29955", "source_url": "https://arxiv.org/abs/2606.29955"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-29955", "source": "claire_radar", "source_benchmark_id": "2606.29955", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Finance"], "area": "Code & Software", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/RUCKBReasoning/SpreadsheetBench-2", "githubScope": "benchmark_repo", "githubStars": 39, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2606.29955", "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", "lastSuccessfulDate": "2026-10-04", "state": "stale"}}, "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": "329d728aee104cb1c8c7d0c29699f8dd02c46b3c2b39565d5c320a36736ba71b", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "SpreadsheetBench 2 evaluates spreadsheet agents on end-to-end business workflows across generation, debugging, and visualization tasks. It includes 321 tasks from authentic business data, with multi-sheet workbooks requiring cross-sheet reasoning. The benchmark provides a unified multi-turn agent scaffold and evaluation scripts for reproducibility.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We introduce \\textsc{SpreadsheetBench 2}, a workflow-level benchmark for spreadsheet agents that covers three task categories: generation, debugging, and visualization."}, "familyId": "bmf_15499180d75b", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "linked-from-project", "linkedUrl": "https://github.com/RUCKBReasoning/SpreadsheetBench", "url": "https://spreadsheetbench.github.io/"}], "scope": "benchmark_repo", "url": "https://github.com/RUCKBReasoning/SpreadsheetBench", "verification": "source-linked"}, {"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/RUCKBReasoning/SpreadsheetBench-2", "url": "https://github.com/RUCKBReasoning/SpreadsheetBench-2"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/RUCKBReasoning/SpreadsheetBench-2", "url": "https://huggingface.co/papers/2606.29955"}, {"basis": "linked-from-project", "linkedUrl": "https://github.com/RUCKBReasoning/SpreadsheetBench-2", "url": "https://spreadsheetbench.github.io/"}], "scope": "benchmark_repo", "url": "https://github.com/RUCKBReasoning/SpreadsheetBench-2", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_spreadsheetbench_a0b75616", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Financial Services"], "links": {"code": "https://github.com/RUCKBReasoning/SpreadsheetBench-2", "data": null, "hfPaper": "https://huggingface.co/papers/2606.29955", "paper": "https://arxiv.org/abs/2606.29955", "pdf": "https://arxiv.org/pdf/2606.29955", "project": "https://spreadsheetbench.github.io/", "report": "https://arxiv.org/abs/2606.29955"}, "metrics": [], "motivation": "Spreadsheets are widely used for business analysis, financial modeling, reporting, and decision-making.", "name": "SpreadsheetBench", "oneLine": "SpreadsheetBench 2 evaluates spreadsheet agents on end-to-end business workflows across generation, debugging, and visualization tasks. It includes 321 tasks from authentic business data, with multi-sheet workbooks requiring cross-sheet reasoning. The benchmark provides a unified multi-turn agent scaffold and evaluation scripts for reproducibility.", "paperTitle": "SpreadsheetBench 2: Evaluating Agents on End-to-End Business Spreadsheet Workflows", "primaryDomain": "Finance", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.29955", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "RUCKBReasoning", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/RUCKBReasoning/SpreadsheetBench-2"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-29", "source": {"authors": ["Jian Zhu", "Yuzheng Zhang", "Zeyao Ma", "Bohan Zhang", "Armin Schoepf", "Daniel Woloch", "Peter Yiliu Wang", "Guangyu Robert Yang", "Samuel Jacob", "Siddharth Nagisetty", "Abhiram Chundru", "Jean Lin", "Spencer Mateega", "Jing Zhang"], "categories": ["cs.SE", "cs.AI"], "id": "2606.29955", "title": "SpreadsheetBench 2: Evaluating Agents on End-to-End Business Spreadsheet Workflows", "type": "arxiv", "url": "https://arxiv.org/abs/2606.29955"}, "sourceUpdatedAt": "2026-06-29T08:33:52Z", "topics": ["cs.SE"], "venueAttempts": [], "whyItMatters": "Existing spreadsheet benchmarks focus on isolated operations, failing to capture real-world workflow complexity. SpreadsheetBench 2 addresses this gap by assessing agents on tasks that require multi-step coordination, cross-sheet reasoning, and deliverable-level outcomes. It provides a challenging testbed for improving reliable spreadsheet automation, with current models achieving only 34.89% overall accuracy."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
