dataset_description.json

{
  "@context": {
    "@vocab": "https://schema.org/",
    "dct": "http://purl.org/dc/terms/",
    "conformsTo": "dct:conformsTo"
  },
  "@type": "Dataset",
  "name": "WebDS",
  "description": "WebDS is a reproducible benchmark dataset for evaluating LLM agents on complex web-based data science tasks. It comprises 870 tasks across 29 containerized websites and 10 domains. Each task simulates realistic end-to-end workflows including data acquisition, tool use, analysis, reasoning, and reporting. Tasks span multiple modalities, structured and unstructured data, and multihop reasoning, and are annotated with fine-grained attributes and difficulty levels.",
  "conformsTo": "http://mlcommons.org/croissant/1.0",
  "license": "https://creativecommons.org/licenses/by/4.0/",
  "keywords": [
    "web-based data science",
    "benchmark",
    "LLM agents",
    "multi-hop reasoning",
    "tool use",
    "agent evaluation"
  ],
  "creator": {
    "@type": "Organization",
    "name": "Stanford University, UC Berkeley, SUTD, USC"
  },
  "includedInDataCatalog": {
    "@type": "DataCatalog",
    "name": "MLCommons Croissant Benchmarks"
  },
  "spatialCoverage": {
    "@type": "Place",
    "name": "Web-based / Global"
  },
  "temporalCoverage": "2025",
  "datePublished": "2025-05-15",
  "version": "1.0",
  "isAccessibleForFree": true
}