{
  "@context": {
    "@language": "en",
    "@vocab": "https://schema.org/",
    "arrayShape": "cr:arrayShape",
    "citeAs": "cr:citeAs",
    "column": "cr:column",
    "conformsTo": "dct:conformsTo",
    "containedIn": "cr:containedIn",
    "cr": "http://mlcommons.org/croissant/",
    "rai": "http://mlcommons.org/croissant/RAI/",
    "data": {
      "@id": "cr:data",
      "@type": "@json"
    },
    "dataType": {
      "@id": "cr:dataType",
      "@type": "@vocab"
    },
    "dct": "http://purl.org/dc/terms/",
    "equivalentProperty": "cr:equivalentProperty",
    "examples": {
      "@id": "cr:examples",
      "@type": "@json"
    },
    "extract": "cr:extract",
    "field": "cr:field",
    "fileProperty": "cr:fileProperty",
    "fileObject": "cr:fileObject",
    "fileSet": "cr:fileSet",
    "format": "cr:format",
    "includes": "cr:includes",
    "isArray": "cr:isArray",
    "isLiveDataset": "cr:isLiveDataset",
    "jsonPath": "cr:jsonPath",
    "key": "cr:key",
    "md5": "cr:md5",
    "parentField": "cr:parentField",
    "path": "cr:path",
    "prov": "http://www.w3.org/ns/prov#",
    "recordSet": "cr:recordSet",
    "references": "cr:references",
    "regex": "cr:regex",
    "repeated": "cr:repeated",
    "replace": "cr:replace",
    "samplingRate": "cr:samplingRate",
    "sc": "https://schema.org/",
    "separator": "cr:separator",
    "source": "cr:source",
    "subField": "cr:subField",
    "transform": "cr:transform"
  },
  "@type": "sc:Dataset",
  "name": "FormulaCascade_Benchmark",
  "alternateName": "FormulaCascade Benchmark",
  "description": "FormulaCascade is an automatically constructed benchmark for long-horizon spreadsheet formula-dependency and formula-family reasoning in real XLSX workbooks. The full benchmark contains 11,304 tasks across Broad, Core, and Challenge release strata. This hosted review release provides the eval-200 release version with 100 Challenge and 100 Core tasks, balanced into 100 single-sheet and 100 cross-sheet tasks.",
  "url": "https://huggingface.co/datasets/anonymity11/FormulaCascade_Benchmark",
  "license": "https://creativecommons.org/licenses/by-nc/4.0/",
  "conformsTo": "http://mlcommons.org/croissant/1.1",
  "citeAs": "Anonymous Authors. FormulaCascade: A 10K+ Automatically Constructed Benchmark and Agentic Harness for Long-Horizon Spreadsheet Formula-Dependency Reasoning. NeurIPS 2026 Evaluations and Datasets Track submission.",
  "version": "1.0.0-release-version",
  "datePublished": "2026-05-07",
  "citation": "Anonymous Authors. FormulaCascade: A 10K+ Automatically Constructed Benchmark and Agentic Harness for Long-Horizon Spreadsheet Formula-Dependency Reasoning. NeurIPS 2026 Evaluations and Datasets Track submission.",
  "keywords": [
    "spreadsheet reasoning",
    "formula-dependency reasoning",
    "formula-family reasoning",
    "benchmark",
    "agentic evaluation",
    "xlsx"
  ],
  "isAccessibleForFree": true,
  "distribution": [
    {
      "@type": "cr:FileObject",
      "@id": "eval200_release_version_archive",
      "name": "FormulaCascade_Benchmark_eval_200_Release_version.tar.gz",
      "description": "Compressed archive of the FormulaCascade eval-200 release-version split.",
      "contentUrl": "https://huggingface.co/datasets/anonymity11/FormulaCascade_Benchmark/resolve/main/FormulaCascade_Benchmark_eval_200_Release_version.tar.gz",
      "encodingFormat": "application/gzip",
      "sha256": "ca0a502ef1749d3a2432e363bd75e41398aec7e232112f7c3c19eea47fcb7965"
    },
    {
      "@type": "cr:FileObject",
      "@id": "readme",
      "name": "README.md",
      "description": "Dataset card and usage documentation.",
      "contentUrl": "https://huggingface.co/datasets/anonymity11/FormulaCascade_Benchmark/resolve/main/README.md",
      "encodingFormat": "text/markdown",
      "sha256": "3a9d8571d5971983285f20d088e6eb55baa7e81756673e016cff5678c09471fb"
    },
    {
      "@type": "cr:FileObject",
      "@id": "sha256sums",
      "name": "SHA256SUMS.txt",
      "description": "SHA256 checksums for release integrity verification.",
      "contentUrl": "https://huggingface.co/datasets/anonymity11/FormulaCascade_Benchmark/resolve/main/SHA256SUMS.txt",
      "encodingFormat": "text/plain",
      "sha256": "e0b639f7ce9aa33a585351d9b648da8beee351231e067662ec0bf5addb5a885d"
    },
    {
      "@type": "cr:FileObject",
      "@id": "croissant_metadata",
      "name": "CROISSANT_RAI_METADATA.json",
      "description": "Croissant metadata with Responsible AI fields.",
      "contentUrl": "https://huggingface.co/datasets/anonymity11/FormulaCascade_Benchmark/resolve/main/CROISSANT_RAI_METADATA.json",
      "encodingFormat": "application/json",
      "sha256": "204c9c79c53bff20a9e6b14ea500c045cc54842ec9e72f36bae7404bab3d585d"
    }
  ],
  "rai:dataLimitations": "FormulaCascade evaluates long-horizon spreadsheet formula-dependency and formula-family reasoning in XLSX workbooks. It is not intended to measure unrestricted spreadsheet UI manipulation, macros, demographic fairness, or model safety outside spreadsheet settings. This hosted eval-200 release version is a balanced subset for efficient review and reproducible comparison; full benchmark conclusions should account for the broader 11,304-task release distribution.",
  "rai:dataBiases": "The dataset inherits selection bias from publicly available spreadsheet corpora and may overrepresent English-language, business-operational, finance, sales, HR, education, engineering, and general data-analysis workbooks. Pipeline filters favor executable, maskable, solvable, non-leaky formula dependency chains, so corrupted, encrypted, macro-dependent, highly visual, very small, or formula-sparse workbooks are underrepresented. The natural-language prompts are generated from visible workbook evidence and may standardize task phrasing beyond the diversity of real user requests.",
  "rai:personalSensitiveInformation": "The benchmark is derived from public workbook data. Source workbooks may contain labels or values originally authored by spreadsheet users. FormulaCascade applies automated filtering, masking, provenance separation, and release-quality gates, but users should treat workbook contents as public-data-derived artifacts and must not attempt reidentification. The benchmark is not designed to study or infer personal, medical, financial-credit, religious, political, or other legally sensitive attributes.",
  "rai:dataUseCases": "Intended uses include evaluating spreadsheet agents, testing long-horizon formula reasoning, measuring executable formula reconstruction, comparing harnesses, analyzing dependency-depth scaling, and diagnosing workbook-native tool-use errors. The benchmark is suitable for research evaluation and reproducibility studies. It is not validated for training on private business data, automated decision-making, compliance auditing, or high-stakes financial, employment, legal, or medical deployment.",
  "rai:dataSocialImpact": "Positive impacts include more transparent evaluation of spreadsheet agents, reproducible measurement of formula-dependency reasoning, and reduced reliance on costly manual task authoring. Potential negative impacts include overfitting spreadsheet agents to public workbook styles, overstating reliability in high-stakes spreadsheet workflows, or using public-derived workbook content outside the intended benchmark setting. Mitigations include solver-visible/evaluator-only separation, documented limitations, release-quality gates, and task-level provenance for error diagnosis.",
  "rai:hasSyntheticData": true,
  "prov:wasDerivedFrom": [
    "https://huggingface.co/datasets/tianzl66/Sheetpedia_xlsx"
  ],
  "prov:wasGeneratedBy": "FormulaCascade was generated by an automated reverse formula-mining pipeline over public formula-bearing workbooks: workbook profiling, dependency graph and formula-family mining, deterministic validity/solvability/uniqueness/leakage gates, bilingual task generation, release quality control, and v5 Challenge/Core/Broad stratification. The eval-200 release version contains 100 Challenge and 100 Core tasks, balanced into 100 single-sheet and 100 cross-sheet tasks. Local machine paths were removed from release metadata.",
  "fileSet": [
    {
      "@type": "cr:FileSet",
      "@id": "release_workbooks",
      "containedIn": {
        "@id": "eval200_release_version_archive"
      },
      "includes": "FormulaCascade_Benchmark_eval_200_Release_version/release/FCB-*/*.xlsx",
      "encodingFormat": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet"
    },
    {
      "@type": "cr:FileSet",
      "@id": "release_task_descriptions",
      "containedIn": {
        "@id": "eval200_release_version_archive"
      },
      "includes": "FormulaCascade_Benchmark_eval_200_Release_version/release/FCB-*/task_description.json",
      "encodingFormat": "application/json"
    },
    {
      "@type": "cr:FileSet",
      "@id": "internal_evaluator_files",
      "containedIn": {
        "@id": "eval200_release_version_archive"
      },
      "includes": "FormulaCascade_Benchmark_eval_200_Release_version/internal/FCB-*/*",
      "encodingFormat": "application/octet-stream"
    }
  ]
}