{
 "tool": "nulltest",
 "version": "1.4",
 "mode": "paired",
 "paired_version": "0.1",
 "probes": [
  "lexicality",
  "words",
  "length",
  "shape",
  "alpha_first",
  "alpha_last"
 ],
 "n_items": 934,
 "n_pairs": 467,
 "pairs_dropped_not_1v1": 0,
 "items_without_pair_key": 0,
 "pair_key": "id",
 "strata": [
  "_all"
 ],
 "alpha": 0.025,
 "family_size": 6,
 "correction": "westfall_young_min_p",
 "null": "labels swapped within each pair with probability 1/2 (sign flip), every probe refit",
 "statistic": "A*_pair = max(w, 1-w), w = share of pairs where label-1 outscores label-0 (ties 1/2)",
 "folds": "grouped by pair (both members held out together)",
 "nulls": {
  "lexicality": {
   "per_stratum": {
    "_all": {
     "n_pairs": 467,
     "astar_pair": 0.501,
     "threshold": 0.551,
     "raw_p": 0.9867,
     "adj_p": 1.0,
     "separates": false,
     "underpowered": false
    }
   }
  },
  "words": {
   "per_stratum": {
    "_all": {
     "n_pairs": 467,
     "astar_pair": 0.528,
     "threshold": 0.548,
     "raw_p": 0.2696,
     "adj_p": 0.8153,
     "separates": false,
     "underpowered": false
    }
   }
  },
  "length": {
   "per_stratum": {
    "_all": {
     "n_pairs": 467,
     "astar_pair": 0.543,
     "threshold": 0.543,
     "raw_p": 0.0582,
     "adj_p": 0.2962,
     "separates": false,
     "underpowered": false
    }
   }
  },
  "shape": {
   "per_stratum": {
    "_all": {
     "n_pairs": 467,
     "astar_pair": 0.502,
     "threshold": 0.546,
     "raw_p": 0.9401,
     "adj_p": 1.0,
     "separates": false,
     "underpowered": false
    }
   }
  },
  "alpha_first": {
   "per_stratum": {
    "_all": {
     "n_pairs": 467,
     "astar_pair": 0.511,
     "threshold": 0.532,
     "raw_p": 0.5591,
     "adj_p": 0.9884,
     "separates": false,
     "underpowered": false
    }
   }
  },
  "alpha_last": {
   "per_stratum": {
    "_all": {
     "n_pairs": 467,
     "astar_pair": 0.518,
     "threshold": 0.537,
     "raw_p": 0.3145,
     "adj_p": 0.8785,
     "separates": false,
     "underpowered": false
    }
   }
  }
 },
 "n_perm_requested": 600,
 "n_perm": 600,
 "worst_cell": {
  "n_pairs": 467,
  "astar_pair": 0.543,
  "threshold": 0.543,
  "raw_p": 0.0582,
  "adj_p": 0.2962,
  "separates": false,
  "underpowered": false,
  "null": "length",
  "stratum": "_all"
 },
 "n_separating_cells": 0,
 "verdict": "PASS",
 "borderline": false,
 "p_floor": 0.001664,
 "saturated": false,
 "headline": "worst cell A*_pair = 0.543 in length/_all (467 pairs), family-adjusted p = 0.2962 over 6 cells",
 "scope": "SCOPE: this measures how much of the label signal is recoverable from SURFACE FORM alone by six model-free probes, two-sided, refit inside every permutation (labels shuffled within stratum) and family-wise corrected across all cells (Westfall-Young min-p). A PASS does NOT certify the benchmark is good: it cannot see circularity (labels drawn from the same store the system under test retrieves from), contamination, or mislabelling. It scores no model.",
 "battery": "B2P_pairs.jsonl",
 "battery_sha256": "7abf19343c5ae245da370b9f4851775ec24983bd4e132db9aaea2ab3023deb2d",
 "tool_sha256": "0e103dc41a3c208d2652646f2b8acc8b701ed7799402ca1ea0cb856de5ddf46c",
 "paired_tool_sha256": "2fbbebca77abd8b494272801fa3f957e1f910d7a10148a7c6ccc10116e4f37d3",
 "keys": {
  "text": "text",
  "label": "label",
  "stratum": null,
  "pair": "id"
 }
}