{
  "citation": "QRData benchmark/QRData.json at de450af45ff7101b328bb064c6b475f73414a7ed (Xiao Liu, Zirui Wu, Xueqing Wu, Pan Lu, Kai-Wei Chang, Yansong Feng. Are LLMs Capable of Data-based Statistical and Causal Reasoning? Benchmarking Advanced Quantitative Reasoning with Data. Findings of the Association for Computational Linguistics: ACL 2024, pages 9215-9235.)",
  "deviation_evidence": "Found in the scripted dry run 20261015T062545Z-suite-qrdata_socsci-scripted-0ae88c, after the spec was committed: the reference path gives 0.0314985 (infer.ols, n = 120), which rounds to 0.031 at the nearest thousandth the question asks for, outside eval.py's open interval (0.03104, 0.03296) around the gold 0.032; the unrounded value would fall inside. The rounding and the interval are left as frozen. Kind unverified (it was paper_rounding): no printed QSS figure shows the source rounding 0.0315 up, so the cause is not shown (B-join review).",
  "doi": null,
  "excerpt": "item 381 answer 0.032",
  "independent": null,
  "item_set": null,
  "kind": "benchmark_gold",
  "known_deviation": "unverified",
  "licence": "CC BY-NC 4.0",
  "locator": "QRData.json item 381 (0-based), field 'answer'",
  "metric": null,
  "n_items": null,
  "page": null,
  "read_copy_sha256": "e076d960f19434adf728aab70480ce34808c06066fcea48eaa38b145d1a111a1",
  "retrieved_at": "2026-10-15",
  "se": null,
  "se_as_published": null,
  "source_id": "qrdata.gold.381",
  "tolerance_reason": null,
  "units": null,
  "url": "https://raw.githubusercontent.com/xxxiaol/QRData/de450af45ff7101b328bb064c6b475f73414a7ed/benchmark/QRData.json",
  "value": 0.032,
  "value_as_published": null,
  "verified_by": "benchmark_file",
  "verified_note": "The printed gold is in the excerpt and index.json. It has fewer than 3 significant digits, so it is not set as value_as_published: the isolation gate searches that field verbatim, and a string this short can occur anywhere in a prompt (the gate does not search such numbers either)."
}