{
  "boolq": {
    "attribution": "BoolQ: Exploring the Surprising Difficulty of Natural Yes/No Questions. Christopher Clark, Kenton Lee, Ming-Wei Chang, Tom Kwiatkowski, Michael Collins, Kristina Toutanova. NAACL 2019.",
    "license": "CC BY-SA 3.0",
    "license_url": "https://creativecommons.org/licenses/by-sa/3.0/",
    "mirror": "https://huggingface.co/datasets/google/boolq",
    "mirror_rows_api": "https://datasets-server.huggingface.co/rows?dataset=google%2Fboolq&config=default&split=validation",
    "official_download": "https://storage.googleapis.com/boolq/dev.jsonl",
    "paper": "https://arxiv.org/abs/1905.10044",
    "raw_jsonl_sha256": "e8fb84fbf510b022e963cddf3a3aded04151afa0ea0ef1cc1bf22f260ddd2344",
    "retrieval_note": "Original Google Cloud Storage URL returned HTTP 403. Downloaded complete Google-owned Hugging Face validation mirror via rows API, then serialized to JSONL. This JSONL hash is of the local normalized mirror, not the original GCS file.",
    "sample_method": "random.Random(seed).sample(range(3270), 60); uniform without replacement; no filtering by label or model output",
    "selected_row_indices_zero_based": [
      659,
      2774,
      579,
      2217,
      1517,
      1662,
      2868,
      1447,
      383,
      2665,
      2041,
      1296,
      1114,
      1953,
      2493,
      3009,
      2504,
      3125,
      1092,
      55,
      2791,
      1842,
      2951,
      852,
      1171,
      1550,
      2116,
      341,
      2098,
      2597,
      468,
      65,
      3114,
      877,
      2630,
      448,
      3134,
      525,
      1381,
      308,
      2849,
      210,
      3115,
      587,
      2552,
      1859,
      1739,
      1066,
      3247,
      1335,
      1979,
      2426,
      3165,
      980,
      2011,
      2818,
      1603,
      1625,
      767,
      1570
    ],
    "source": "https://github.com/google-research-datasets/boolean-questions",
    "source_rows": 3270
  },
  "cases_file": "cases.jsonl",
  "cases_sha256": "60e6db60c72869936548c3c4620aa3b070cf12e516f3be42a61a306d0f592660",
  "count": 180,
  "distribution": {
    "boolq": {
      "count": 60,
      "labels": {
        "no": 24,
        "yes": 36
      },
      "target_max": 1.0,
      "target_mean": 0.6,
      "target_min": 0.0
    },
    "policy": {
      "count": 60,
      "labels": {
        "no": 30,
        "yes": 30
      },
      "target_max": 1.0,
      "target_mean": 0.5,
      "target_min": 0.0
    },
    "probability": {
      "count": 60,
      "target_max": 1.0,
      "target_mean": 0.4896041483485319,
      "target_min": 0.0
    }
  },
  "generator_version": 1,
  "limitations": [
    "This is a small, exploratory pilot, not a broad ranking of either model.",
    "BoolQ development labels are human annotations rather than a mathematical oracle; the public set may be present in model training data.",
    "Synthetic policy and probability cases measure the specified rules and probability problems, not general operational or forecasting performance.",
    "Probability cases use exact known event probabilities, not observed binary outcomes. Use MAE/RMSE and expected proper scores for these cases.",
    "Thresholds and prompts must be frozen before reading scored model outputs."
  ],
  "model_input_fields": [
    "state",
    "question",
    "criteria"
  ],
  "name": "Luna versus Jev matched binary decisions pilot",
  "never_send_to_models": [
    "id",
    "experiment",
    "target",
    "target_kind",
    "source"
  ],
  "policy_families": {
    "access": 20,
    "delivery": 20,
    "refund": 20
  },
  "probability_families": {
    "conditional_table": 20,
    "weighted_mixture": 20,
    "without_replacement": 20
  },
  "seed": 20261002
}