{
  "selection": {
    "candidate_variants": [
      "verified_structural",
      "row_column_rule",
      "split_fold_unpivot",
      "header_aware",
      "cell_parser"
    ],
    "family_splits": {
      "dev": [
        "potters_wheel_fold_2",
        "synthetic_10",
        "synthetic_17",
        "synthetic_24",
        "synthetic_30",
        "synthetic_37",
        "synthetic_40",
        "synthetic_45",
        "synthetic_46",
        "synthetic_51"
      ],
      "test": [
        "agriculture",
        "crime_data_wrangler",
        "potters_wheel_merge_split",
        "potters_wheel_unfold",
        "potters_wheel_unfold2",
        "proactive_wrangling_fold",
        "synthetic_12",
        "synthetic_25",
        "synthetic_48",
        "synthetic_8"
      ],
      "train": [
        "craigslist_data_wrangler",
        "potters_wheel_divide",
        "potters_wheel_fold",
        "potters_wheel_split_fold",
        "proactive_wrangling_complex",
        "reshape_table_structure_data_wrangler",
        "synthetic_11",
        "synthetic_13",
        "synthetic_15",
        "synthetic_18",
        "synthetic_19",
        "synthetic_2",
        "synthetic_22",
        "synthetic_26",
        "synthetic_27",
        "synthetic_28",
        "synthetic_29",
        "synthetic_3",
        "synthetic_33",
        "synthetic_34",
        "synthetic_36",
        "synthetic_4",
        "synthetic_41",
        "synthetic_43",
        "synthetic_44",
        "synthetic_47",
        "synthetic_49",
        "synthetic_5",
        "synthetic_6",
        "synthetic_7"
      ]
    },
    "greedy_order": [
      "verified_structural",
      "cell_parser",
      "row_column_rule",
      "header_aware",
      "split_fold_unpivot"
    ],
    "selected_k": 5,
    "selected_policy": "consensus_2",
    "selected_variants": [
      "verified_structural",
      "cell_parser",
      "row_column_rule",
      "header_aware",
      "split_fold_unpivot"
    ],
    "selection_rule": "greedy ordered variants on train; choose prefix and selector maximizing dev exact accuracy with loss/precision/token tie-breaks"
  },
  "split_metrics": {
    "dev": {
      "direct": {
        "accuracy": 0.4,
        "direct_accuracy": 0.4,
        "direct_correct_losses": 0,
        "direct_exact": 4,
        "direct_miss_recoveries": 0,
        "exact": 4,
        "n": 10,
        "oracle_union": 4,
        "oracle_union_accuracy": 0.4,
        "oracle_visible_program": 0,
        "program_commit_correct": 0,
        "program_commit_precision": null,
        "program_commits": 0,
        "total_forward_tokens": 5235
      },
      "selected": {
        "accuracy": 0.5,
        "direct_accuracy": 0.4,
        "direct_correct_losses": 0,
        "direct_exact": 4,
        "direct_miss_recoveries": 1,
        "exact": 5,
        "n": 10,
        "oracle_union": 5,
        "oracle_union_accuracy": 0.5,
        "oracle_visible_program": 3,
        "program_commit_correct": 2,
        "program_commit_precision": 1.0,
        "program_commits": 2,
        "total_forward_tokens": 87242
      },
      "selected_consensus_2": {
        "accuracy": 0.5,
        "direct_accuracy": 0.4,
        "direct_correct_losses": 0,
        "direct_exact": 4,
        "direct_miss_recoveries": 1,
        "exact": 5,
        "n": 10,
        "oracle_union": 5,
        "oracle_union_accuracy": 0.5,
        "oracle_visible_program": 3,
        "program_commit_correct": 2,
        "program_commit_precision": 1.0,
        "program_commits": 2,
        "total_forward_tokens": 87242
      },
      "selected_first_visible": {
        "accuracy": 0.4,
        "direct_accuracy": 0.4,
        "direct_correct_losses": 1,
        "direct_exact": 4,
        "direct_miss_recoveries": 1,
        "exact": 4,
        "n": 10,
        "oracle_union": 5,
        "oracle_union_accuracy": 0.5,
        "oracle_visible_program": 3,
        "program_commit_correct": 3,
        "program_commit_precision": 0.42857142857142855,
        "program_commits": 7,
        "total_forward_tokens": 87242
      },
      "selected_oracle": {
        "accuracy": 0.4,
        "direct_accuracy": 0.4,
        "direct_correct_losses": 1,
        "direct_exact": 4,
        "direct_miss_recoveries": 1,
        "exact": 4,
        "n": 10,
        "oracle_union": 5,
        "oracle_union_accuracy": 0.5,
        "oracle_visible_program": 3,
        "program_commit_correct": 3,
        "program_commit_precision": 0.42857142857142855,
        "program_commits": 7,
        "total_forward_tokens": 87242
      }
    },
    "test": {
      "direct": {
        "accuracy": 0.42,
        "direct_accuracy": 0.42,
        "direct_correct_losses": 0,
        "direct_exact": 21,
        "direct_miss_recoveries": 0,
        "exact": 21,
        "n": 50,
        "oracle_union": 21,
        "oracle_union_accuracy": 0.42,
        "oracle_visible_program": 0,
        "program_commit_correct": 0,
        "program_commit_precision": null,
        "program_commits": 0,
        "total_forward_tokens": 73911
      },
      "selected": {
        "accuracy": 0.48,
        "direct_accuracy": 0.42,
        "direct_correct_losses": 1,
        "direct_exact": 21,
        "direct_miss_recoveries": 4,
        "exact": 24,
        "n": 50,
        "oracle_union": 29,
        "oracle_union_accuracy": 0.58,
        "oracle_visible_program": 18,
        "program_commit_correct": 11,
        "program_commit_precision": 0.7857142857142857,
        "program_commits": 14,
        "total_forward_tokens": 890030
      },
      "selected_consensus_2": {
        "accuracy": 0.48,
        "direct_accuracy": 0.42,
        "direct_correct_losses": 1,
        "direct_exact": 21,
        "direct_miss_recoveries": 4,
        "exact": 24,
        "n": 50,
        "oracle_union": 29,
        "oracle_union_accuracy": 0.58,
        "oracle_visible_program": 18,
        "program_commit_correct": 11,
        "program_commit_precision": 0.7857142857142857,
        "program_commits": 14,
        "total_forward_tokens": 890030
      },
      "selected_first_visible": {
        "accuracy": 0.56,
        "direct_accuracy": 0.42,
        "direct_correct_losses": 1,
        "direct_exact": 21,
        "direct_miss_recoveries": 8,
        "exact": 28,
        "n": 50,
        "oracle_union": 29,
        "oracle_union_accuracy": 0.58,
        "oracle_visible_program": 18,
        "program_commit_correct": 18,
        "program_commit_precision": 0.782608695652174,
        "program_commits": 23,
        "total_forward_tokens": 890030
      },
      "selected_oracle": {
        "accuracy": 0.56,
        "direct_accuracy": 0.42,
        "direct_correct_losses": 1,
        "direct_exact": 21,
        "direct_miss_recoveries": 8,
        "exact": 28,
        "n": 50,
        "oracle_union": 29,
        "oracle_union_accuracy": 0.58,
        "oracle_visible_program": 18,
        "program_commit_correct": 18,
        "program_commit_precision": 0.782608695652174,
        "program_commits": 23,
        "total_forward_tokens": 890030
      }
    },
    "train": {
      "direct": {
        "accuracy": 0.3,
        "direct_accuracy": 0.3,
        "direct_correct_losses": 0,
        "direct_exact": 9,
        "direct_miss_recoveries": 0,
        "exact": 9,
        "n": 30,
        "oracle_union": 9,
        "oracle_union_accuracy": 0.3,
        "oracle_visible_program": 0,
        "program_commit_correct": 0,
        "program_commit_precision": null,
        "program_commits": 0,
        "total_forward_tokens": 23539
      },
      "selected": {
        "accuracy": 0.3333333333333333,
        "direct_accuracy": 0.3,
        "direct_correct_losses": 0,
        "direct_exact": 9,
        "direct_miss_recoveries": 1,
        "exact": 10,
        "n": 30,
        "oracle_union": 12,
        "oracle_union_accuracy": 0.4,
        "oracle_visible_program": 10,
        "program_commit_correct": 4,
        "program_commit_precision": 0.4,
        "program_commits": 10,
        "total_forward_tokens": 341451
      },
      "selected_consensus_2": {
        "accuracy": 0.3333333333333333,
        "direct_accuracy": 0.3,
        "direct_correct_losses": 0,
        "direct_exact": 9,
        "direct_miss_recoveries": 1,
        "exact": 10,
        "n": 30,
        "oracle_union": 12,
        "oracle_union_accuracy": 0.4,
        "oracle_visible_program": 10,
        "program_commit_correct": 4,
        "program_commit_precision": 0.4,
        "program_commits": 10,
        "total_forward_tokens": 341451
      },
      "selected_first_visible": {
        "accuracy": 0.36666666666666664,
        "direct_accuracy": 0.3,
        "direct_correct_losses": 1,
        "direct_exact": 9,
        "direct_miss_recoveries": 3,
        "exact": 11,
        "n": 30,
        "oracle_union": 12,
        "oracle_union_accuracy": 0.4,
        "oracle_visible_program": 10,
        "program_commit_correct": 10,
        "program_commit_precision": 0.45454545454545453,
        "program_commits": 22,
        "total_forward_tokens": 341451
      },
      "selected_oracle": {
        "accuracy": 0.36666666666666664,
        "direct_accuracy": 0.3,
        "direct_correct_losses": 1,
        "direct_exact": 9,
        "direct_miss_recoveries": 3,
        "exact": 11,
        "n": 30,
        "oracle_union": 12,
        "oracle_union_accuracy": 0.4,
        "oracle_visible_program": 10,
        "program_commit_correct": 10,
        "program_commit_precision": 0.45454545454545453,
        "program_commits": 22,
        "total_forward_tokens": 341451
      }
    }
  },
  "test_family_metrics": {
    "agriculture": {
      "accuracy": 0.6,
      "direct_accuracy": 0.8,
      "direct_correct_losses": 1,
      "direct_exact": 4,
      "direct_miss_recoveries": 0,
      "exact": 3,
      "n": 5,
      "oracle_union": 4,
      "oracle_union_accuracy": 0.8,
      "oracle_visible_program": 0,
      "program_commit_correct": 0,
      "program_commit_precision": 0.0,
      "program_commits": 1,
      "total_forward_tokens": 64187
    },
    "crime_data_wrangler": {
      "accuracy": 0.0,
      "direct_accuracy": 0.0,
      "direct_correct_losses": 0,
      "direct_exact": 0,
      "direct_miss_recoveries": 0,
      "exact": 0,
      "n": 5,
      "oracle_union": 0,
      "oracle_union_accuracy": 0.0,
      "oracle_visible_program": 0,
      "program_commit_correct": 0,
      "program_commit_precision": null,
      "program_commits": 0,
      "total_forward_tokens": 233360
    },
    "potters_wheel_merge_split": {
      "accuracy": 0.6,
      "direct_accuracy": 0.0,
      "direct_correct_losses": 0,
      "direct_exact": 0,
      "direct_miss_recoveries": 3,
      "exact": 3,
      "n": 5,
      "oracle_union": 4,
      "oracle_union_accuracy": 0.8,
      "oracle_visible_program": 4,
      "program_commit_correct": 3,
      "program_commit_precision": 0.75,
      "program_commits": 4,
      "total_forward_tokens": 24387
    },
    "potters_wheel_unfold": {
      "accuracy": 0.4,
      "direct_accuracy": 0.4,
      "direct_correct_losses": 0,
      "direct_exact": 2,
      "direct_miss_recoveries": 0,
      "exact": 2,
      "n": 5,
      "oracle_union": 2,
      "oracle_union_accuracy": 0.4,
      "oracle_visible_program": 0,
      "program_commit_correct": 0,
      "program_commit_precision": null,
      "program_commits": 0,
      "total_forward_tokens": 56662
    },
    "potters_wheel_unfold2": {
      "accuracy": 0.0,
      "direct_accuracy": 0.0,
      "direct_correct_losses": 0,
      "direct_exact": 0,
      "direct_miss_recoveries": 0,
      "exact": 0,
      "n": 5,
      "oracle_union": 0,
      "oracle_union_accuracy": 0.0,
      "oracle_visible_program": 0,
      "program_commit_correct": 0,
      "program_commit_precision": null,
      "program_commits": 0,
      "total_forward_tokens": 41700
    },
    "proactive_wrangling_fold": {
      "accuracy": 1.0,
      "direct_accuracy": 1.0,
      "direct_correct_losses": 0,
      "direct_exact": 5,
      "direct_miss_recoveries": 0,
      "exact": 5,
      "n": 5,
      "oracle_union": 5,
      "oracle_union_accuracy": 1.0,
      "oracle_visible_program": 4,
      "program_commit_correct": 3,
      "program_commit_precision": 1.0,
      "program_commits": 3,
      "total_forward_tokens": 30184
    },
    "synthetic_12": {
      "accuracy": 1.0,
      "direct_accuracy": 1.0,
      "direct_correct_losses": 0,
      "direct_exact": 5,
      "direct_miss_recoveries": 0,
      "exact": 5,
      "n": 5,
      "oracle_union": 5,
      "oracle_union_accuracy": 1.0,
      "oracle_visible_program": 5,
      "program_commit_correct": 4,
      "program_commit_precision": 1.0,
      "program_commits": 4,
      "total_forward_tokens": 105112
    },
    "synthetic_25": {
      "accuracy": 0.8,
      "direct_accuracy": 0.8,
      "direct_correct_losses": 0,
      "direct_exact": 4,
      "direct_miss_recoveries": 0,
      "exact": 4,
      "n": 5,
      "oracle_union": 4,
      "oracle_union_accuracy": 0.8,
      "oracle_visible_program": 1,
      "program_commit_correct": 0,
      "program_commit_precision": null,
      "program_commits": 0,
      "total_forward_tokens": 128627
    },
    "synthetic_48": {
      "accuracy": 0.2,
      "direct_accuracy": 0.2,
      "direct_correct_losses": 0,
      "direct_exact": 1,
      "direct_miss_recoveries": 0,
      "exact": 1,
      "n": 5,
      "oracle_union": 1,
      "oracle_union_accuracy": 0.2,
      "oracle_visible_program": 0,
      "program_commit_correct": 0,
      "program_commit_precision": 0.0,
      "program_commits": 1,
      "total_forward_tokens": 54213
    },
    "synthetic_8": {
      "accuracy": 0.2,
      "direct_accuracy": 0.0,
      "direct_correct_losses": 0,
      "direct_exact": 0,
      "direct_miss_recoveries": 1,
      "exact": 1,
      "n": 5,
      "oracle_union": 4,
      "oracle_union_accuracy": 0.8,
      "oracle_visible_program": 4,
      "program_commit_correct": 1,
      "program_commit_precision": 1.0,
      "program_commits": 1,
      "total_forward_tokens": 151598
    }
  },
  "test_policy_metrics": {
    "consensus_2": {
      "accuracy": 0.48,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 4,
      "exact": 24,
      "n": 50,
      "oracle_union": 29,
      "oracle_union_accuracy": 0.58,
      "oracle_visible_program": 18,
      "program_commit_correct": 11,
      "program_commit_precision": 0.7857142857142857,
      "program_commits": 14,
      "total_forward_tokens": 890030
    },
    "consensus_3": {
      "accuracy": 0.48,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 3,
      "exact": 24,
      "n": 50,
      "oracle_union": 29,
      "oracle_union_accuracy": 0.58,
      "oracle_visible_program": 18,
      "program_commit_correct": 10,
      "program_commit_precision": 0.9090909090909091,
      "program_commits": 11,
      "total_forward_tokens": 890030
    },
    "direct": {
      "accuracy": 0.42,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 0,
      "exact": 21,
      "n": 50,
      "oracle_union": 29,
      "oracle_union_accuracy": 0.58,
      "oracle_visible_program": 18,
      "program_commit_correct": 0,
      "program_commit_precision": null,
      "program_commits": 0,
      "total_forward_tokens": 890030
    },
    "first_visible_program": {
      "accuracy": 0.56,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 8,
      "exact": 28,
      "n": 50,
      "oracle_union": 29,
      "oracle_union_accuracy": 0.58,
      "oracle_visible_program": 18,
      "program_commit_correct": 18,
      "program_commit_precision": 0.782608695652174,
      "program_commits": 23,
      "total_forward_tokens": 890030
    }
  },
  "variant_quality": [
    {
      "initial_visible_pass": 20,
      "n": 90,
      "repair_added_visible": 3,
      "variant": "cell_parser",
      "visible_pass": 23,
      "visible_pass_rate": 0.25555555555555554,
      "visible_precision": 0.6521739130434783
    },
    {
      "initial_visible_pass": 6,
      "n": 90,
      "repair_added_visible": 9,
      "variant": "header_aware",
      "visible_pass": 15,
      "visible_pass_rate": 0.16666666666666666,
      "visible_precision": 0.4666666666666667
    },
    {
      "initial_visible_pass": 16,
      "n": 90,
      "repair_added_visible": 6,
      "variant": "row_column_rule",
      "visible_pass": 22,
      "visible_pass_rate": 0.24444444444444444,
      "visible_precision": 0.6363636363636364
    },
    {
      "initial_visible_pass": 21,
      "n": 90,
      "repair_added_visible": 7,
      "variant": "split_fold_unpivot",
      "visible_pass": 28,
      "visible_pass_rate": 0.3111111111111111,
      "visible_precision": 0.5714285714285714
    },
    {
      "initial_visible_pass": 29,
      "n": 90,
      "repair_added_visible": 3,
      "variant": "verified_structural",
      "visible_pass": 32,
      "visible_pass_rate": 0.35555555555555557,
      "visible_precision": 0.625
    }
  ]
}