{
  "benchmark": "SciFact test retrieval",
  "bootstrap": {
    "method_interval_seeds": {
      "late": 20260718,
      "naive": 20260717,
      "whole_document": 20260719
    },
    "paired_difference_seed": 20260717,
    "percentile_interpolation": "linear interpolation at zero-based position (B - 1) * p",
    "prng": "SplitMix64 with rejection-sampled bounded integers",
    "replicates": 20000,
    "seed": 20260717,
    "unit": "query"
  },
  "methods": {
    "late": {
      "mrr_at_10": {
        "mean": 0.6337248677248677
      },
      "ndcg_at_10": {
        "mean": 0.6609776646723947,
        "query_bootstrap_95_percent_interval": [
          0.6145399649247424,
          0.7070844743964247
        ]
      },
      "recall_at_10": {
        "mean": 0.7775555555555556
      }
    },
    "naive": {
      "mrr_at_10": {
        "mean": 0.6123492063492062
      },
      "ndcg_at_10": {
        "mean": 0.6414254895397593,
        "query_bootstrap_95_percent_interval": [
          0.5941801288149041,
          0.6875954822650707
        ]
      },
      "recall_at_10": {
        "mean": 0.7625555555555555
      }
    },
    "whole_document": {
      "mrr_at_10": {
        "mean": 0.6104973544973544
      },
      "ndcg_at_10": {
        "mean": 0.6388903593656915,
        "query_bootstrap_95_percent_interval": [
          0.5917522695931684,
          0.6852170695308275
        ]
      },
      "recall_at_10": {
        "mean": 0.7592222222222221
      }
    }
  },
  "metric_unit": "nDCG, Recall, and reciprocal rank on a 0-1 scale",
  "paired_late_minus_naive_ndcg_at_10": {
    "improved_queries": 49,
    "mean_difference": 0.019552175132635588,
    "query_bootstrap_95_percent_interval": [
      0.0032721920785872076,
      0.03584544477819499
    ],
    "tied_queries": 229,
    "uncertainty_semantics": "Percentile interval from 20,000 paired resamples of the 300 fixed SciFact test queries. It describes query-level stability for this benchmark and is not a population confidence interval.",
    "worse_queries": 22
  },
  "query_count": 300,
  "schema_version": 1
}
