{
  "protocol": "tab_namer_sealed_final_gate_v1",
  "protocol_sha256": "238e1df1ba77e7583d0e9f032580a3bb85705f08cd2656d0c307de32c55c3789",
  "sealed_tasks_sha256": "4694f1e4b5269b525fd8fb595f45b9ce65d5f831fdd981537da07479831b6874",
  "rows": 1000,
  "judge": {
    "model": "gemini-3.5-flash-lite",
    "blinded": true,
    "reference_labels_shown": false,
    "failures": [],
    "estimated_cost_usd": 0.2009706
  },
  "candidates": {
    "ours_locked": {
      "mean": 6.19741,
      "mean_bootstrap_95ci": [
        6.0814,
        6.3117
      ],
      "median": 6.5,
      "stdev": 1.8602,
      "ge_5": 749,
      "ge_5_rate": 0.749,
      "ge_6": 635,
      "ge_6_rate": 0.635,
      "fail_le_2": 25,
      "fail_le_2_rate": 0.025
    },
    "flan_locked": {
      "mean": 5.44564,
      "mean_bootstrap_95ci": [
        5.3284,
        5.5638
      ],
      "median": 5.5,
      "stdev": 1.8942,
      "ge_5": 606,
      "ge_5_rate": 0.606,
      "ge_6": 446,
      "ge_6_rate": 0.446,
      "fail_le_2": 39,
      "fail_le_2_rate": 0.039
    },
    "ours_bounded32": {
      "mean": 6.15466,
      "mean_bootstrap_95ci": [
        6.039,
        6.2686
      ],
      "median": 6.5,
      "stdev": 1.8499,
      "ge_5": 741,
      "ge_5_rate": 0.741,
      "ge_6": 623,
      "ge_6_rate": 0.623,
      "fail_le_2": 24,
      "fail_le_2_rate": 0.024
    },
    "flan_bounded32": {
      "mean": 5.42119,
      "mean_bootstrap_95ci": [
        5.3051,
        5.5379
      ],
      "median": 5.5,
      "stdev": 1.8777,
      "ge_5": 603,
      "ge_5_rate": 0.603,
      "ge_6": 441,
      "ge_6_rate": 0.441,
      "fail_le_2": 39,
      "fail_le_2_rate": 0.039
    }
  },
  "comparisons": {
    "primary_locked": {
      "left": "ours_locked",
      "right": "flan_locked",
      "mean_delta": 0.75177,
      "mean_delta_bootstrap_95ci": [
        0.6107,
        0.8911
      ],
      "wins": 603,
      "ties": 67,
      "losses": 330
    },
    "secondary_bounded32": {
      "left": "ours_bounded32",
      "right": "flan_bounded32",
      "mean_delta": 0.73347,
      "mean_delta_bootstrap_95ci": [
        0.5927,
        0.8726
      ],
      "wins": 601,
      "ties": 68,
      "losses": 331
    }
  },
  "independent_audit": {
    "primary_locked": {
      "model": "ours_locked",
      "baseline": "flan_locked",
      "n": 40,
      "mean_score_0_to_4": 2.3,
      "useful_3_or_4": 19,
      "useful_rate": 0.475,
      "useful_rate_wilson_95ci": [
        0.3294,
        0.625
      ],
      "wins": 18,
      "ties": 9,
      "losses": 13,
      "wins_plus_ties": 27,
      "wins_plus_ties_rate": 0.675,
      "wins_plus_ties_rate_wilson_95ci": [
        0.5202,
        0.7992
      ]
    },
    "secondary_bounded32": {
      "model": "ours_bounded32",
      "baseline": "flan_bounded32",
      "n": 40,
      "mean_score_0_to_4": 2.3,
      "useful_3_or_4": 19,
      "useful_rate": 0.475,
      "useful_rate_wilson_95ci": [
        0.3294,
        0.625
      ],
      "wins": 17,
      "ties": 10,
      "losses": 13,
      "wins_plus_ties": 27,
      "wins_plus_ties_rate": 0.675,
      "wins_plus_ties_rate_wilson_95ci": [
        0.5202,
        0.7992
      ]
    },
    "flan_locked_useful": {
      "n": 40,
      "useful_3_or_4": 13,
      "useful_rate": 0.325,
      "useful_rate_wilson_95ci": [
        0.2008,
        0.4798
      ]
    }
  },
  "structural_audit": {
    "raw": {
      "path": "output/gsg_scaleup_20260814/final_gate_20260815/ours_beam4.raw.jsonl",
      "sha256": "afc3198b622d25c6cc800c0365880c010bd67299bf887d30025a0409c1efc125",
      "rows": 1000,
      "unique": 956,
      "empty": 0,
      "top_title_mass": 0.011,
      "max_chars": 32,
      "violations": {}
    },
    "candidates": {
      "ours_locked": {
        "path": "output/gsg_scaleup_20260814/final_gate_20260815/ours_locked.jsonl",
        "sha256": "d28b3c41a596e9006b6d49f17ba8f9f1a3cc31e9cde1cc537c478681a1acd2c0",
        "rows": 1000,
        "ids_match_raw": true,
        "ids_unique": true,
        "unique": 999,
        "empty": 0,
        "top_title_mass": 0.002,
        "over_32_chars": 46,
        "max_chars": 48
      },
      "ours_bounded32": {
        "path": "output/gsg_scaleup_20260814/final_gate_20260815/ours_bounded32.jsonl",
        "sha256": "8afa62396b8fa47a3cd4a8f54ecc4f8acb606ed142c65f4ca1745d0dbfd46118",
        "rows": 1000,
        "ids_match_raw": true,
        "ids_unique": true,
        "unique": 999,
        "empty": 0,
        "top_title_mass": 0.002,
        "over_32_chars": 0,
        "max_chars": 32
      },
      "flan_locked": {
        "path": "output/gsg_scaleup_20260814/final_gate_20260815/flan_locked.jsonl",
        "sha256": "3bdf7563c6b313e8209a45d7d66def262763bbdfcc8ef757d522b21d70477f83",
        "rows": 1000,
        "ids_match_raw": true,
        "ids_unique": true,
        "unique": 1000,
        "empty": 0,
        "top_title_mass": 0.001,
        "over_32_chars": 50,
        "max_chars": 71
      },
      "flan_bounded32": {
        "path": "output/gsg_scaleup_20260814/final_gate_20260815/flan_bounded32.jsonl",
        "sha256": "386cb6fee7b7eeafb09662004cab3b9c9ea23ca0698dc504885bb006041e76b1",
        "rows": 1000,
        "ids_match_raw": true,
        "ids_unique": true,
        "unique": 1000,
        "empty": 0,
        "top_title_mass": 0.001,
        "over_32_chars": 0,
        "max_chars": 32
      }
    }
  },
  "preregistered_gate": {
    "mean_at_least_5_5": true,
    "audit_useful_rate_at_least_0_60": false,
    "audit_wins_plus_ties_rate_at_least_0_30": true,
    "empty_rate_at_most_0_01": true,
    "top_title_mass_at_most_0_10": true,
    "all_criteria_pass": false
  },
  "bootstrap": {
    "replicates": 100000,
    "seed": "tab-namer-final-gate-bootstrap-v1",
    "method": "paired row-resampling percentile interval"
  },
  "artifact_sha256": {
    "scores": "578ad7605e56e1c72677b2162fa8e8c83cab54348ba8d61c531006e6253f8201",
    "judge_summary": "e47037b5c8052fd332ec66e61d57ccd137b819e39524aa6ee442dd83a5b3eee9",
    "audit_primary": "d82f37481bf30ab7d429f097a28043b4fe8b1957393939867771c6b7086f1932",
    "audit_secondary": "2faae92d25f3a51bd34bd4011789eab235ce0824c5d21722573c2479a54df77e",
    "structural_audit": "e117232cf9c0b6ac127110c2e1dfb4724a5a712a7860705838d0ba5873b356ae"
  }
}
