{
  "deployable_pareto_front": [
    "shape_visible_out_less_cols_all_first_visible",
    "prefix1_first_visible",
    "direct_only"
  ],
  "greedy_trace": [
    {
      "added": "verified_structural",
      "selected": [
        "verified_structural"
      ],
      "step": 1,
      "train_metrics": {
        "accuracy": 0.36666666666666664,
        "avg_forward_tokens": 2597.733333333333,
        "direct_accuracy": 0.3,
        "direct_correct_losses": 0,
        "direct_exact": 9,
        "direct_miss_recoveries": 2,
        "exact": 11,
        "generated_variant_counts": {
          "verified_structural": 30
        },
        "n": 30,
        "policy": "trial_verified_structural",
        "program_commit_correct": 5,
        "program_commit_precision": 0.4166666666666667,
        "program_commits": 12,
        "total_forward_tokens": 77932,
        "triggered_tasks": 0
      }
    },
    {
      "added": "cell_parser",
      "selected": [
        "verified_structural",
        "cell_parser"
      ],
      "step": 2,
      "train_metrics": {
        "accuracy": 0.36666666666666664,
        "avg_forward_tokens": 4491.3,
        "direct_accuracy": 0.3,
        "direct_correct_losses": 1,
        "direct_exact": 9,
        "direct_miss_recoveries": 3,
        "exact": 11,
        "generated_variant_counts": {
          "cell_parser": 30,
          "verified_structural": 30
        },
        "n": 30,
        "policy": "trial_cell_parser",
        "program_commit_correct": 8,
        "program_commit_precision": 0.4444444444444444,
        "program_commits": 18,
        "total_forward_tokens": 134739,
        "triggered_tasks": 0
      }
    },
    {
      "added": "row_column_rule",
      "selected": [
        "verified_structural",
        "cell_parser",
        "row_column_rule"
      ],
      "step": 3,
      "train_metrics": {
        "accuracy": 0.36666666666666664,
        "avg_forward_tokens": 6706.666666666667,
        "direct_accuracy": 0.3,
        "direct_correct_losses": 1,
        "direct_exact": 9,
        "direct_miss_recoveries": 3,
        "exact": 11,
        "generated_variant_counts": {
          "cell_parser": 30,
          "row_column_rule": 30,
          "verified_structural": 30
        },
        "n": 30,
        "policy": "trial_row_column_rule",
        "program_commit_correct": 9,
        "program_commit_precision": 0.47368421052631576,
        "program_commits": 19,
        "total_forward_tokens": 201200,
        "triggered_tasks": 0
      }
    },
    {
      "added": "header_aware",
      "selected": [
        "verified_structural",
        "cell_parser",
        "row_column_rule",
        "header_aware"
      ],
      "step": 4,
      "train_metrics": {
        "accuracy": 0.36666666666666664,
        "avg_forward_tokens": 9332.666666666666,
        "direct_accuracy": 0.3,
        "direct_correct_losses": 1,
        "direct_exact": 9,
        "direct_miss_recoveries": 3,
        "exact": 11,
        "generated_variant_counts": {
          "cell_parser": 30,
          "header_aware": 30,
          "row_column_rule": 30,
          "verified_structural": 30
        },
        "n": 30,
        "policy": "trial_header_aware",
        "program_commit_correct": 10,
        "program_commit_precision": 0.47619047619047616,
        "program_commits": 21,
        "total_forward_tokens": 279980,
        "triggered_tasks": 0
      }
    },
    {
      "added": "split_fold_unpivot",
      "selected": [
        "verified_structural",
        "cell_parser",
        "row_column_rule",
        "header_aware",
        "split_fold_unpivot"
      ],
      "step": 5,
      "train_metrics": {
        "accuracy": 0.36666666666666664,
        "avg_forward_tokens": 11381.7,
        "direct_accuracy": 0.3,
        "direct_correct_losses": 1,
        "direct_exact": 9,
        "direct_miss_recoveries": 3,
        "exact": 11,
        "generated_variant_counts": {
          "cell_parser": 30,
          "header_aware": 30,
          "row_column_rule": 30,
          "split_fold_unpivot": 30,
          "verified_structural": 30
        },
        "n": 30,
        "policy": "trial_split_fold_unpivot",
        "program_commit_correct": 10,
        "program_commit_precision": 0.45454545454545453,
        "program_commits": 22,
        "total_forward_tokens": 341451,
        "triggered_tasks": 0
      }
    }
  ],
  "selected_policy": "shape_direct_out_cols_less_new_in_all_first_visible",
  "shape_trigger_selection": {
    "candidates": [
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 2163.2,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 3,
            "header_aware": 3,
            "row_column_rule": 3,
            "split_fold_unpivot": 3,
            "verified_structural": 3
          },
          "n": 10,
          "policy": "shape:direct_out_cols_less_new_in",
          "program_commit_correct": 3,
          "program_commit_precision": 1.0,
          "program_commits": 3,
          "total_forward_tokens": 21632,
          "triggered_tasks": 3
        },
        "score": [
          5,
          0,
          1.0,
          -21632,
          8,
          -1
        ],
        "train": {
          "accuracy": 0.26666666666666666,
          "avg_forward_tokens": 3516.3,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 1,
          "direct_exact": 9,
          "direct_miss_recoveries": 0,
          "exact": 8,
          "generated_variant_counts": {
            "cell_parser": 6,
            "header_aware": 6,
            "row_column_rule": 6,
            "split_fold_unpivot": 6,
            "verified_structural": 6
          },
          "n": 30,
          "policy": "shape:direct_out_cols_less_new_in",
          "program_commit_correct": 1,
          "program_commit_precision": 0.2,
          "program_commits": 5,
          "total_forward_tokens": 105489,
          "triggered_tasks": 6
        },
        "trigger": "direct_out_cols_less_new_in"
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 3488.7,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 4,
            "header_aware": 4,
            "row_column_rule": 4,
            "split_fold_unpivot": 4,
            "verified_structural": 4
          },
          "n": 10,
          "policy": "shape:visible_out_less_cols",
          "program_commit_correct": 3,
          "program_commit_precision": 1.0,
          "program_commits": 3,
          "total_forward_tokens": 34887,
          "triggered_tasks": 4
        },
        "score": [
          5,
          0,
          1.0,
          -34887,
          9,
          -1
        ],
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 3840.3,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 1,
          "direct_exact": 9,
          "direct_miss_recoveries": 1,
          "exact": 9,
          "generated_variant_counts": {
            "cell_parser": 7,
            "header_aware": 7,
            "row_column_rule": 7,
            "split_fold_unpivot": 7,
            "verified_structural": 7
          },
          "n": 30,
          "policy": "shape:visible_out_less_cols",
          "program_commit_correct": 2,
          "program_commit_precision": 0.3333333333333333,
          "program_commits": 6,
          "total_forward_tokens": 115209,
          "triggered_tasks": 7
        },
        "trigger": "visible_out_less_cols"
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 4844.0,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 6,
            "header_aware": 6,
            "row_column_rule": 6,
            "split_fold_unpivot": 6,
            "verified_structural": 6
          },
          "n": 10,
          "policy": "shape:visible_less_cols_or_direct_mismatch",
          "program_commit_correct": 3,
          "program_commit_precision": 0.6,
          "program_commits": 5,
          "total_forward_tokens": 48440,
          "triggered_tasks": 6
        },
        "score": [
          5,
          0,
          0.6,
          -48440,
          9,
          -1
        ],
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 6616.733333333334,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 1,
          "direct_exact": 9,
          "direct_miss_recoveries": 1,
          "exact": 9,
          "generated_variant_counts": {
            "cell_parser": 14,
            "header_aware": 14,
            "row_column_rule": 14,
            "split_fold_unpivot": 14,
            "verified_structural": 14
          },
          "n": 30,
          "policy": "shape:visible_less_cols_or_direct_mismatch",
          "program_commit_correct": 2,
          "program_commit_precision": 0.2,
          "program_commits": 10,
          "total_forward_tokens": 198502,
          "triggered_tasks": 14
        },
        "trigger": "visible_less_cols_or_direct_mismatch"
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 5868.5,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 7,
            "header_aware": 7,
            "row_column_rule": 7,
            "split_fold_unpivot": 7,
            "verified_structural": 7
          },
          "n": 10,
          "policy": "shape:visible_cols_change",
          "program_commit_correct": 3,
          "program_commit_precision": 0.6,
          "program_commits": 5,
          "total_forward_tokens": 58685,
          "triggered_tasks": 7
        },
        "score": [
          5,
          0,
          0.6,
          -58685,
          11,
          -1
        ],
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 9727.666666666666,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 1,
          "direct_exact": 9,
          "direct_miss_recoveries": 3,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 23,
            "header_aware": 23,
            "row_column_rule": 23,
            "split_fold_unpivot": 23,
            "verified_structural": 23
          },
          "n": 30,
          "policy": "shape:visible_cols_change",
          "program_commit_correct": 8,
          "program_commit_precision": 0.5333333333333333,
          "program_commits": 15,
          "total_forward_tokens": 291830,
          "triggered_tasks": 23
        },
        "trigger": "visible_cols_change"
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 5868.5,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 7,
            "header_aware": 7,
            "row_column_rule": 7,
            "split_fold_unpivot": 7,
            "verified_structural": 7
          },
          "n": 10,
          "policy": "shape:visible_cols_change_or_direct_mismatch",
          "program_commit_correct": 3,
          "program_commit_precision": 0.6,
          "program_commits": 5,
          "total_forward_tokens": 58685,
          "triggered_tasks": 7
        },
        "score": [
          5,
          0,
          0.6,
          -58685,
          11,
          -1
        ],
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 9891.3,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 1,
          "direct_exact": 9,
          "direct_miss_recoveries": 3,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 24,
            "header_aware": 24,
            "row_column_rule": 24,
            "split_fold_unpivot": 24,
            "verified_structural": 24
          },
          "n": 30,
          "policy": "shape:visible_cols_change_or_direct_mismatch",
          "program_commit_correct": 8,
          "program_commit_precision": 0.5,
          "program_commits": 16,
          "total_forward_tokens": 296739,
          "triggered_tasks": 24
        },
        "trigger": "visible_cols_change_or_direct_mismatch"
      },
      {
        "dev": {
          "accuracy": 0.4,
          "avg_forward_tokens": 523.5,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 0,
          "exact": 4,
          "generated_variant_counts": {},
          "n": 10,
          "policy": "shape:never",
          "program_commit_correct": 0,
          "program_commit_precision": null,
          "program_commits": 0,
          "total_forward_tokens": 5235,
          "triggered_tasks": 0
        },
        "score": [
          4,
          0,
          1.0,
          -5235,
          9,
          0
        ],
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 784.6333333333333,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 0,
          "exact": 9,
          "generated_variant_counts": {},
          "n": 30,
          "policy": "shape:never",
          "program_commit_correct": 0,
          "program_commit_precision": null,
          "program_commits": 0,
          "total_forward_tokens": 23539,
          "triggered_tasks": 0
        },
        "trigger": "never"
      },
      {
        "dev": {
          "accuracy": 0.4,
          "avg_forward_tokens": 523.5,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 0,
          "exact": 4,
          "generated_variant_counts": {},
          "n": 10,
          "policy": "shape:direct_parse_fail",
          "program_commit_correct": 0,
          "program_commit_precision": null,
          "program_commits": 0,
          "total_forward_tokens": 5235,
          "triggered_tasks": 0
        },
        "score": [
          4,
          0,
          1.0,
          -5235,
          9,
          0
        ],
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 2348.5,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 0,
          "exact": 9,
          "generated_variant_counts": {
            "cell_parser": 1,
            "header_aware": 1,
            "row_column_rule": 1,
            "split_fold_unpivot": 1,
            "verified_structural": 1
          },
          "n": 30,
          "policy": "shape:direct_parse_fail",
          "program_commit_correct": 0,
          "program_commit_precision": null,
          "program_commits": 0,
          "total_forward_tokens": 70455,
          "triggered_tasks": 1
        },
        "trigger": "direct_parse_fail"
      },
      {
        "dev": {
          "accuracy": 0.4,
          "avg_forward_tokens": 1673.7,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 0,
          "exact": 4,
          "generated_variant_counts": {
            "cell_parser": 2,
            "header_aware": 2,
            "row_column_rule": 2,
            "split_fold_unpivot": 2,
            "verified_structural": 2
          },
          "n": 10,
          "policy": "shape:visible_out_more_rows",
          "program_commit_correct": 1,
          "program_commit_precision": 1.0,
          "program_commits": 1,
          "total_forward_tokens": 16737,
          "triggered_tasks": 2
        },
        "score": [
          4,
          0,
          1.0,
          -16737,
          9,
          -1
        ],
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 3887.0666666666666,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 1,
          "direct_exact": 9,
          "direct_miss_recoveries": 1,
          "exact": 9,
          "generated_variant_counts": {
            "cell_parser": 8,
            "header_aware": 8,
            "row_column_rule": 8,
            "split_fold_unpivot": 8,
            "verified_structural": 8
          },
          "n": 30,
          "policy": "shape:visible_out_more_rows",
          "program_commit_correct": 5,
          "program_commit_precision": 0.7142857142857143,
          "program_commits": 7,
          "total_forward_tokens": 116612,
          "triggered_tasks": 8
        },
        "trigger": "visible_out_more_rows"
      },
      {
        "dev": {
          "accuracy": 0.4,
          "avg_forward_tokens": 2903.3,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 0,
          "exact": 4,
          "generated_variant_counts": {
            "cell_parser": 3,
            "header_aware": 3,
            "row_column_rule": 3,
            "split_fold_unpivot": 3,
            "verified_structural": 3
          },
          "n": 10,
          "policy": "shape:visible_out_more_cols",
          "program_commit_correct": 0,
          "program_commit_precision": 0.0,
          "program_commits": 2,
          "total_forward_tokens": 29033,
          "triggered_tasks": 3
        },
        "score": [
          4,
          0,
          0.0,
          -29033,
          11,
          0
        ],
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 6672.0,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 16,
            "header_aware": 16,
            "row_column_rule": 16,
            "split_fold_unpivot": 16,
            "verified_structural": 16
          },
          "n": 30,
          "policy": "shape:visible_out_more_cols",
          "program_commit_correct": 6,
          "program_commit_precision": 0.6666666666666666,
          "program_commits": 9,
          "total_forward_tokens": 200160,
          "triggered_tasks": 16
        },
        "trigger": "visible_out_more_cols"
      },
      {
        "dev": {
          "accuracy": 0.4,
          "avg_forward_tokens": 3204.3,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 0,
          "exact": 4,
          "generated_variant_counts": {
            "cell_parser": 3,
            "header_aware": 3,
            "row_column_rule": 3,
            "split_fold_unpivot": 3,
            "verified_structural": 3
          },
          "n": 10,
          "policy": "shape:direct_cols_mismatch_visible_out",
          "program_commit_correct": 0,
          "program_commit_precision": 0.0,
          "program_commits": 2,
          "total_forward_tokens": 32043,
          "triggered_tasks": 3
        },
        "score": [
          4,
          0,
          0.0,
          -32043,
          10,
          0
        ],
        "train": {
          "accuracy": 0.3333333333333333,
          "avg_forward_tokens": 5448.933333333333,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 1,
          "exact": 10,
          "generated_variant_counts": {
            "cell_parser": 9,
            "header_aware": 9,
            "row_column_rule": 9,
            "split_fold_unpivot": 9,
            "verified_structural": 9
          },
          "n": 30,
          "policy": "shape:direct_cols_mismatch_visible_out",
          "program_commit_correct": 1,
          "program_commit_precision": 0.2,
          "program_commits": 5,
          "total_forward_tokens": 163468,
          "triggered_tasks": 9
        },
        "trigger": "direct_cols_mismatch_visible_out"
      },
      {
        "dev": {
          "accuracy": 0.4,
          "avg_forward_tokens": 8263.3,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 1,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 4,
          "generated_variant_counts": {
            "cell_parser": 9,
            "header_aware": 9,
            "row_column_rule": 9,
            "split_fold_unpivot": 9,
            "verified_structural": 9
          },
          "n": 10,
          "policy": "shape:visible_shape_change",
          "program_commit_correct": 3,
          "program_commit_precision": 0.5,
          "program_commits": 6,
          "total_forward_tokens": 82633,
          "triggered_tasks": 9
        },
        "score": [
          4,
          -1,
          0.5,
          -82633,
          11,
          -1
        ],
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 10704.7,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 1,
          "direct_exact": 9,
          "direct_miss_recoveries": 3,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 26,
            "header_aware": 26,
            "row_column_rule": 26,
            "split_fold_unpivot": 26,
            "verified_structural": 26
          },
          "n": 30,
          "policy": "shape:visible_shape_change",
          "program_commit_correct": 10,
          "program_commit_precision": 0.5555555555555556,
          "program_commits": 18,
          "total_forward_tokens": 321141,
          "triggered_tasks": 26
        },
        "trigger": "visible_shape_change"
      },
      {
        "dev": {
          "accuracy": 0.4,
          "avg_forward_tokens": 7699.7,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 1,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 4,
          "generated_variant_counts": {
            "cell_parser": 9,
            "header_aware": 9,
            "row_column_rule": 9,
            "split_fold_unpivot": 9,
            "verified_structural": 9
          },
          "n": 10,
          "policy": "shape:direct_rows_mismatch_visible_out",
          "program_commit_correct": 3,
          "program_commit_precision": 0.42857142857142855,
          "program_commits": 7,
          "total_forward_tokens": 76997,
          "triggered_tasks": 9
        },
        "score": [
          4,
          -1,
          0.42857142857142855,
          -76997,
          10,
          -1
        ],
        "train": {
          "accuracy": 0.3333333333333333,
          "avg_forward_tokens": 9603.566666666668,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 1,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 10,
          "generated_variant_counts": {
            "cell_parser": 25,
            "header_aware": 25,
            "row_column_rule": 25,
            "split_fold_unpivot": 25,
            "verified_structural": 25
          },
          "n": 30,
          "policy": "shape:direct_rows_mismatch_visible_out",
          "program_commit_correct": 9,
          "program_commit_precision": 0.47368421052631576,
          "program_commits": 19,
          "total_forward_tokens": 288107,
          "triggered_tasks": 25
        },
        "trigger": "direct_rows_mismatch_visible_out"
      },
      {
        "dev": {
          "accuracy": 0.4,
          "avg_forward_tokens": 8724.2,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 1,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 4,
          "generated_variant_counts": {
            "cell_parser": 10,
            "header_aware": 10,
            "row_column_rule": 10,
            "split_fold_unpivot": 10,
            "verified_structural": 10
          },
          "n": 10,
          "policy": "shape:always",
          "program_commit_correct": 3,
          "program_commit_precision": 0.42857142857142855,
          "program_commits": 7,
          "total_forward_tokens": 87242,
          "triggered_tasks": 10
        },
        "score": [
          4,
          -1,
          0.42857142857142855,
          -87242,
          11,
          -1
        ],
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 11381.7,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 1,
          "direct_exact": 9,
          "direct_miss_recoveries": 3,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 30,
            "header_aware": 30,
            "row_column_rule": 30,
            "split_fold_unpivot": 30,
            "verified_structural": 30
          },
          "n": 30,
          "policy": "shape:always",
          "program_commit_correct": 10,
          "program_commit_precision": 0.45454545454545453,
          "program_commits": 22,
          "total_forward_tokens": 341451,
          "triggered_tasks": 30
        },
        "trigger": "always"
      },
      {
        "dev": {
          "accuracy": 0.3,
          "avg_forward_tokens": 5556.9,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 1,
          "direct_exact": 4,
          "direct_miss_recoveries": 0,
          "exact": 3,
          "generated_variant_counts": {
            "cell_parser": 6,
            "header_aware": 6,
            "row_column_rule": 6,
            "split_fold_unpivot": 6,
            "verified_structural": 6
          },
          "n": 10,
          "policy": "shape:visible_rows_change",
          "program_commit_correct": 1,
          "program_commit_precision": 0.25,
          "program_commits": 4,
          "total_forward_tokens": 55569,
          "triggered_tasks": 6
        },
        "score": [
          3,
          -1,
          0.25,
          -55569,
          11,
          -1
        ],
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 9882.0,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 1,
          "direct_exact": 9,
          "direct_miss_recoveries": 3,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 23,
            "header_aware": 23,
            "row_column_rule": 23,
            "split_fold_unpivot": 23,
            "verified_structural": 23
          },
          "n": 30,
          "policy": "shape:visible_rows_change",
          "program_commit_correct": 9,
          "program_commit_precision": 0.6,
          "program_commits": 15,
          "total_forward_tokens": 296460,
          "triggered_tasks": 23
        },
        "trigger": "visible_rows_change"
      },
      {
        "dev": {
          "accuracy": 0.3,
          "avg_forward_tokens": 4406.7,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 1,
          "direct_exact": 4,
          "direct_miss_recoveries": 0,
          "exact": 3,
          "generated_variant_counts": {
            "cell_parser": 4,
            "header_aware": 4,
            "row_column_rule": 4,
            "split_fold_unpivot": 4,
            "verified_structural": 4
          },
          "n": 10,
          "policy": "shape:visible_out_less_rows",
          "program_commit_correct": 0,
          "program_commit_precision": 0.0,
          "program_commits": 3,
          "total_forward_tokens": 44067,
          "triggered_tasks": 4
        },
        "score": [
          3,
          -1,
          0.0,
          -44067,
          11,
          0
        ],
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 6779.566666666667,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 15,
            "header_aware": 15,
            "row_column_rule": 15,
            "split_fold_unpivot": 15,
            "verified_structural": 15
          },
          "n": 30,
          "policy": "shape:visible_out_less_rows",
          "program_commit_correct": 4,
          "program_commit_precision": 0.5,
          "program_commits": 8,
          "total_forward_tokens": 203387,
          "triggered_tasks": 15
        },
        "trigger": "visible_out_less_rows"
      }
    ],
    "selected": {
      "dev": {
        "accuracy": 0.5,
        "avg_forward_tokens": 2163.2,
        "direct_accuracy": 0.4,
        "direct_correct_losses": 0,
        "direct_exact": 4,
        "direct_miss_recoveries": 1,
        "exact": 5,
        "generated_variant_counts": {
          "cell_parser": 3,
          "header_aware": 3,
          "row_column_rule": 3,
          "split_fold_unpivot": 3,
          "verified_structural": 3
        },
        "n": 10,
        "policy": "shape:direct_out_cols_less_new_in",
        "program_commit_correct": 3,
        "program_commit_precision": 1.0,
        "program_commits": 3,
        "total_forward_tokens": 21632,
        "triggered_tasks": 3
      },
      "score": [
        5,
        0,
        1.0,
        -21632,
        8,
        -1
      ],
      "train": {
        "accuracy": 0.26666666666666666,
        "avg_forward_tokens": 3516.3,
        "direct_accuracy": 0.3,
        "direct_correct_losses": 1,
        "direct_exact": 9,
        "direct_miss_recoveries": 0,
        "exact": 8,
        "generated_variant_counts": {
          "cell_parser": 6,
          "header_aware": 6,
          "row_column_rule": 6,
          "split_fold_unpivot": 6,
          "verified_structural": 6
        },
        "n": 30,
        "policy": "shape:direct_out_cols_less_new_in",
        "program_commit_correct": 1,
        "program_commit_precision": 0.2,
        "program_commits": 5,
        "total_forward_tokens": 105489,
        "triggered_tasks": 6
      },
      "trigger": "direct_out_cols_less_new_in"
    }
  },
  "test_family_metrics_primary": {
    "agriculture": {
      "accuracy": 0.8,
      "direct_accuracy": 0.8,
      "direct_exact": 4,
      "exact": 4,
      "losses": 0,
      "n": 5,
      "program_commits": 0,
      "recoveries": 0,
      "tokens": 5453
    },
    "crime_data_wrangler": {
      "accuracy": 0.0,
      "direct_accuracy": 0.0,
      "direct_exact": 0,
      "exact": 0,
      "losses": 0,
      "n": 5,
      "program_commits": 0,
      "recoveries": 0,
      "tokens": 233360
    },
    "potters_wheel_merge_split": {
      "accuracy": 0.8,
      "direct_accuracy": 0.0,
      "direct_exact": 0,
      "exact": 4,
      "losses": 0,
      "n": 5,
      "program_commits": 4,
      "recoveries": 4,
      "tokens": 21126
    },
    "potters_wheel_unfold": {
      "accuracy": 0.4,
      "direct_accuracy": 0.4,
      "direct_exact": 2,
      "exact": 2,
      "losses": 0,
      "n": 5,
      "program_commits": 0,
      "recoveries": 0,
      "tokens": 12913
    },
    "potters_wheel_unfold2": {
      "accuracy": 0.0,
      "direct_accuracy": 0.0,
      "direct_exact": 0,
      "exact": 0,
      "losses": 0,
      "n": 5,
      "program_commits": 0,
      "recoveries": 0,
      "tokens": 1409
    },
    "proactive_wrangling_fold": {
      "accuracy": 1.0,
      "direct_accuracy": 1.0,
      "direct_exact": 5,
      "exact": 5,
      "losses": 0,
      "n": 5,
      "program_commits": 0,
      "recoveries": 0,
      "tokens": 1708
    },
    "synthetic_12": {
      "accuracy": 1.0,
      "direct_accuracy": 1.0,
      "direct_exact": 5,
      "exact": 5,
      "losses": 0,
      "n": 5,
      "program_commits": 0,
      "recoveries": 0,
      "tokens": 11102
    },
    "synthetic_25": {
      "accuracy": 0.8,
      "direct_accuracy": 0.8,
      "direct_exact": 4,
      "exact": 4,
      "losses": 0,
      "n": 5,
      "program_commits": 0,
      "recoveries": 0,
      "tokens": 9942
    },
    "synthetic_48": {
      "accuracy": 0.2,
      "direct_accuracy": 0.2,
      "direct_exact": 1,
      "exact": 1,
      "losses": 0,
      "n": 5,
      "program_commits": 0,
      "recoveries": 0,
      "tokens": 2242
    },
    "synthetic_8": {
      "accuracy": 0.8,
      "direct_accuracy": 0.0,
      "direct_exact": 0,
      "exact": 4,
      "losses": 0,
      "n": 5,
      "program_commits": 4,
      "recoveries": 4,
      "tokens": 151598
    }
  },
  "test_metrics": [
    {
      "accuracy": 0.58,
      "avg_forward_tokens": 3959.02,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 8,
      "exact": 29,
      "generated_variant_counts": {
        "cell_parser": 8,
        "header_aware": 8,
        "row_column_rule": 8,
        "split_fold_unpivot": 8,
        "verified_structural": 8
      },
      "n": 50,
      "policy": "oracle_budget_run_all_only_on_helpful_tasks",
      "program_commit_correct": 8,
      "program_commit_precision": 1.0,
      "program_commits": 8,
      "total_forward_tokens": 197951,
      "triggered_tasks": 8
    },
    {
      "accuracy": 0.58,
      "avg_forward_tokens": 4627.3,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 8,
      "exact": 29,
      "generated_variant_counts": {
        "cell_parser": 9,
        "header_aware": 9,
        "row_column_rule": 9,
        "split_fold_unpivot": 9,
        "verified_structural": 9
      },
      "n": 50,
      "policy": "shape_visible_out_less_cols_all_first_visible",
      "program_commit_correct": 8,
      "program_commit_precision": 1.0,
      "program_commits": 8,
      "total_forward_tokens": 231365,
      "triggered_tasks": 9
    },
    {
      "accuracy": 0.58,
      "avg_forward_tokens": 9017.06,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 8,
      "exact": 29,
      "generated_variant_counts": {
        "cell_parser": 15,
        "header_aware": 15,
        "row_column_rule": 15,
        "split_fold_unpivot": 15,
        "verified_structural": 15
      },
      "n": 50,
      "policy": "shape_direct_out_cols_less_new_in_all_first_visible",
      "program_commit_correct": 8,
      "program_commit_precision": 1.0,
      "program_commits": 8,
      "total_forward_tokens": 450853,
      "triggered_tasks": 15
    },
    {
      "accuracy": 0.58,
      "avg_forward_tokens": 10330.72,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 8,
      "exact": 29,
      "generated_variant_counts": {
        "cell_parser": 22,
        "header_aware": 22,
        "row_column_rule": 22,
        "split_fold_unpivot": 22,
        "verified_structural": 22
      },
      "n": 50,
      "policy": "shape_visible_less_cols_or_direct_mismatch_all_first_visible",
      "program_commit_correct": 8,
      "program_commit_precision": 0.8,
      "program_commits": 10,
      "total_forward_tokens": 516536,
      "triggered_tasks": 22
    },
    {
      "accuracy": 0.58,
      "avg_forward_tokens": 13984.16,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 8,
      "exact": 29,
      "generated_variant_counts": {
        "cell_parser": 34,
        "header_aware": 34,
        "row_column_rule": 34,
        "split_fold_unpivot": 34,
        "verified_structural": 34
      },
      "n": 50,
      "policy": "shape_visible_cols_change_all_first_visible",
      "program_commit_correct": 9,
      "program_commit_precision": 0.6923076923076923,
      "program_commits": 13,
      "total_forward_tokens": 699208,
      "triggered_tasks": 34
    },
    {
      "accuracy": 0.58,
      "avg_forward_tokens": 13984.16,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 8,
      "exact": 29,
      "generated_variant_counts": {
        "cell_parser": 34,
        "header_aware": 34,
        "row_column_rule": 34,
        "split_fold_unpivot": 34,
        "verified_structural": 34
      },
      "n": 50,
      "policy": "shape_visible_cols_change_or_direct_mismatch_all_first_visible",
      "program_commit_correct": 9,
      "program_commit_precision": 0.6923076923076923,
      "program_commits": 13,
      "total_forward_tokens": 699208,
      "triggered_tasks": 34
    },
    {
      "accuracy": 0.58,
      "avg_forward_tokens": 16281.48,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 8,
      "exact": 29,
      "generated_variant_counts": {
        "cell_parser": 43,
        "header_aware": 43,
        "row_column_rule": 43,
        "split_fold_unpivot": 43,
        "verified_structural": 43
      },
      "n": 50,
      "policy": "shape_visible_shape_change_all_first_visible",
      "program_commit_correct": 18,
      "program_commit_precision": 0.8181818181818182,
      "program_commits": 22,
      "total_forward_tokens": 814074,
      "triggered_tasks": 43
    },
    {
      "accuracy": 0.58,
      "avg_forward_tokens": 17800.6,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 8,
      "exact": 29,
      "generated_variant_counts": {
        "cell_parser": 50,
        "header_aware": 50,
        "row_column_rule": 50,
        "split_fold_unpivot": 50,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "oracle_best_available_full_budget",
      "program_commit_correct": 8,
      "program_commit_precision": 1.0,
      "program_commits": 8,
      "total_forward_tokens": 890030,
      "triggered_tasks": 8
    },
    {
      "accuracy": 0.56,
      "avg_forward_tokens": 16295.52,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 8,
      "exact": 28,
      "generated_variant_counts": {
        "cell_parser": 43,
        "header_aware": 43,
        "row_column_rule": 43,
        "split_fold_unpivot": 43,
        "verified_structural": 43
      },
      "n": 50,
      "policy": "shape_direct_rows_mismatch_visible_out_all_first_visible",
      "program_commit_correct": 15,
      "program_commit_precision": 0.75,
      "program_commits": 20,
      "total_forward_tokens": 814776,
      "triggered_tasks": 43
    },
    {
      "accuracy": 0.56,
      "avg_forward_tokens": 17800.6,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 8,
      "exact": 28,
      "generated_variant_counts": {
        "cell_parser": 50,
        "header_aware": 50,
        "row_column_rule": 50,
        "split_fold_unpivot": 50,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "prefix5_first_visible",
      "program_commit_correct": 18,
      "program_commit_precision": 0.782608695652174,
      "program_commits": 23,
      "total_forward_tokens": 890030,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.56,
      "avg_forward_tokens": 17800.6,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 8,
      "exact": 28,
      "generated_variant_counts": {
        "cell_parser": 50,
        "header_aware": 50,
        "row_column_rule": 50,
        "split_fold_unpivot": 50,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "shape_always_all_first_visible",
      "program_commit_correct": 18,
      "program_commit_precision": 0.782608695652174,
      "program_commits": 23,
      "total_forward_tokens": 890030,
      "triggered_tasks": 50
    },
    {
      "accuracy": 0.54,
      "avg_forward_tokens": 10777.5,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 7,
      "exact": 27,
      "generated_variant_counts": {
        "cell_parser": 50,
        "row_column_rule": 50,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "prefix3_first_visible",
      "program_commit_correct": 17,
      "program_commit_precision": 0.85,
      "program_commits": 20,
      "total_forward_tokens": 538875,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.54,
      "avg_forward_tokens": 14513.14,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 7,
      "exact": 27,
      "generated_variant_counts": {
        "cell_parser": 50,
        "header_aware": 50,
        "row_column_rule": 50,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "prefix4_first_visible",
      "program_commit_correct": 17,
      "program_commit_precision": 0.85,
      "program_commits": 20,
      "total_forward_tokens": 725657,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.52,
      "avg_forward_tokens": 7392.14,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 6,
      "exact": 26,
      "generated_variant_counts": {
        "cell_parser": 50,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "prefix2_first_visible",
      "program_commit_correct": 16,
      "program_commit_precision": 0.8421052631578947,
      "program_commits": 19,
      "total_forward_tokens": 369607,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.5,
      "avg_forward_tokens": 4291.24,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 5,
      "exact": 25,
      "generated_variant_counts": {
        "verified_structural": 50
      },
      "n": 50,
      "policy": "prefix1_first_visible",
      "program_commit_correct": 14,
      "program_commit_precision": 0.8235294117647058,
      "program_commits": 17,
      "total_forward_tokens": 214562,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.5,
      "avg_forward_tokens": 4291.24,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 5,
      "exact": 25,
      "generated_variant_counts": {
        "verified_structural": 50
      },
      "n": 50,
      "policy": "single_verified_structural",
      "program_commit_correct": 14,
      "program_commit_precision": 0.8235294117647058,
      "program_commits": 17,
      "total_forward_tokens": 214562,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.5,
      "avg_forward_tokens": 4657.04,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 4,
      "exact": 25,
      "generated_variant_counts": {
        "cell_parser": 9,
        "header_aware": 9,
        "row_column_rule": 9,
        "split_fold_unpivot": 9,
        "verified_structural": 9
      },
      "n": 50,
      "policy": "shape_visible_out_more_rows_all_first_visible",
      "program_commit_correct": 8,
      "program_commit_precision": 1.0,
      "program_commits": 8,
      "total_forward_tokens": 232852,
      "triggered_tasks": 9
    },
    {
      "accuracy": 0.5,
      "avg_forward_tokens": 4765.68,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 4,
      "exact": 25,
      "generated_variant_counts": {
        "split_fold_unpivot": 50
      },
      "n": 50,
      "policy": "single_split_fold_unpivot",
      "program_commit_correct": 9,
      "program_commit_precision": 0.8181818181818182,
      "program_commits": 11,
      "total_forward_tokens": 238284,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.5,
      "avg_forward_tokens": 4863.58,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 4,
      "exact": 25,
      "generated_variant_counts": {
        "row_column_rule": 50
      },
      "n": 50,
      "policy": "single_row_column_rule",
      "program_commit_correct": 10,
      "program_commit_precision": 0.8333333333333334,
      "program_commits": 12,
      "total_forward_tokens": 243179,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.5,
      "avg_forward_tokens": 5604.12,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 5,
      "exact": 25,
      "generated_variant_counts": {
        "cell_parser": 8,
        "header_aware": 8,
        "row_column_rule": 8,
        "split_fold_unpivot": 8,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "canary_verified_structural_disagree_escalate",
      "program_commit_correct": 5,
      "program_commit_precision": 0.625,
      "program_commits": 8,
      "total_forward_tokens": 280206,
      "triggered_tasks": 8
    },
    {
      "accuracy": 0.5,
      "avg_forward_tokens": 5724.42,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 4,
      "exact": 25,
      "generated_variant_counts": {
        "cell_parser": 6,
        "header_aware": 6,
        "row_column_rule": 50,
        "split_fold_unpivot": 6,
        "verified_structural": 6
      },
      "n": 50,
      "policy": "canary_row_column_rule_disagree_escalate",
      "program_commit_correct": 4,
      "program_commit_precision": 0.6666666666666666,
      "program_commits": 6,
      "total_forward_tokens": 286221,
      "triggered_tasks": 6
    },
    {
      "accuracy": 0.5,
      "avg_forward_tokens": 5982.86,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 4,
      "exact": 25,
      "generated_variant_counts": {
        "cell_parser": 5,
        "header_aware": 5,
        "row_column_rule": 5,
        "split_fold_unpivot": 50,
        "verified_structural": 5
      },
      "n": 50,
      "policy": "canary_split_fold_unpivot_disagree_escalate",
      "program_commit_correct": 4,
      "program_commit_precision": 0.8,
      "program_commits": 5,
      "total_forward_tokens": 299143,
      "triggered_tasks": 5
    },
    {
      "accuracy": 0.5,
      "avg_forward_tokens": 8629.68,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 4,
      "exact": 25,
      "generated_variant_counts": {
        "cell_parser": 11,
        "header_aware": 11,
        "row_column_rule": 11,
        "split_fold_unpivot": 11,
        "verified_structural": 11
      },
      "n": 50,
      "policy": "shape_direct_parse_fail_all_first_visible",
      "program_commit_correct": 4,
      "program_commit_precision": 1.0,
      "program_commits": 4,
      "total_forward_tokens": 431484,
      "triggered_tasks": 11
    },
    {
      "accuracy": 0.5,
      "avg_forward_tokens": 9943.34,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 4,
      "exact": 25,
      "generated_variant_counts": {
        "cell_parser": 18,
        "header_aware": 18,
        "row_column_rule": 18,
        "split_fold_unpivot": 18,
        "verified_structural": 18
      },
      "n": 50,
      "policy": "shape_direct_cols_mismatch_visible_out_all_first_visible",
      "program_commit_correct": 4,
      "program_commit_precision": 0.6666666666666666,
      "program_commits": 6,
      "total_forward_tokens": 497167,
      "triggered_tasks": 18
    },
    {
      "accuracy": 0.5,
      "avg_forward_tokens": 15828.88,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 4,
      "exact": 25,
      "generated_variant_counts": {
        "cell_parser": 38,
        "header_aware": 38,
        "row_column_rule": 38,
        "split_fold_unpivot": 38,
        "verified_structural": 38
      },
      "n": 50,
      "policy": "shape_visible_rows_change_all_first_visible",
      "program_commit_correct": 14,
      "program_commit_precision": 0.8235294117647058,
      "program_commits": 17,
      "total_forward_tokens": 791444,
      "triggered_tasks": 38
    },
    {
      "accuracy": 0.48,
      "avg_forward_tokens": 17800.6,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 3,
      "exact": 24,
      "generated_variant_counts": {
        "cell_parser": 50,
        "header_aware": 50,
        "row_column_rule": 50,
        "split_fold_unpivot": 50,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "prefix5_consensus3",
      "program_commit_correct": 10,
      "program_commit_precision": 0.9090909090909091,
      "program_commits": 11,
      "total_forward_tokens": 890030,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.48,
      "avg_forward_tokens": 17800.6,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 4,
      "exact": 24,
      "generated_variant_counts": {
        "cell_parser": 50,
        "header_aware": 50,
        "row_column_rule": 50,
        "split_fold_unpivot": 50,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "prefix5_consensus2",
      "program_commit_correct": 11,
      "program_commit_precision": 0.7857142857142857,
      "program_commits": 14,
      "total_forward_tokens": 890030,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.46,
      "avg_forward_tokens": 10777.5,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 3,
      "exact": 23,
      "generated_variant_counts": {
        "cell_parser": 50,
        "row_column_rule": 50,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "prefix3_consensus2",
      "program_commit_correct": 10,
      "program_commit_precision": 0.7692307692307693,
      "program_commits": 13,
      "total_forward_tokens": 538875,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.46,
      "avg_forward_tokens": 14513.14,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 3,
      "exact": 23,
      "generated_variant_counts": {
        "cell_parser": 50,
        "header_aware": 50,
        "row_column_rule": 50,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "prefix4_consensus2",
      "program_commit_correct": 10,
      "program_commit_precision": 0.7692307692307693,
      "program_commits": 13,
      "total_forward_tokens": 725657,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.44,
      "avg_forward_tokens": 5683.82,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 1,
      "exact": 22,
      "generated_variant_counts": {
        "cell_parser": 3,
        "header_aware": 50,
        "row_column_rule": 3,
        "split_fold_unpivot": 3,
        "verified_structural": 3
      },
      "n": 50,
      "policy": "canary_header_aware_disagree_escalate",
      "program_commit_correct": 2,
      "program_commit_precision": 0.6666666666666666,
      "program_commits": 3,
      "total_forward_tokens": 284191,
      "triggered_tasks": 3
    },
    {
      "accuracy": 0.44,
      "avg_forward_tokens": 14513.14,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 1,
      "exact": 22,
      "generated_variant_counts": {
        "cell_parser": 50,
        "header_aware": 50,
        "row_column_rule": 50,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "prefix4_consensus3",
      "program_commit_correct": 7,
      "program_commit_precision": 0.875,
      "program_commits": 8,
      "total_forward_tokens": 725657,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.42,
      "avg_forward_tokens": 1478.22,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 0,
      "exact": 21,
      "generated_variant_counts": {},
      "n": 50,
      "policy": "direct_only",
      "program_commit_correct": 0,
      "program_commit_precision": null,
      "program_commits": 0,
      "total_forward_tokens": 73911,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.42,
      "avg_forward_tokens": 1478.22,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 0,
      "exact": 21,
      "generated_variant_counts": {},
      "n": 50,
      "policy": "shape_never_all_first_visible",
      "program_commit_correct": 0,
      "program_commit_precision": null,
      "program_commits": 0,
      "total_forward_tokens": 73911,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.42,
      "avg_forward_tokens": 4579.12,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 1,
      "exact": 21,
      "generated_variant_counts": {
        "cell_parser": 50
      },
      "n": 50,
      "policy": "single_cell_parser",
      "program_commit_correct": 8,
      "program_commit_precision": 0.8,
      "program_commits": 10,
      "total_forward_tokens": 228956,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.42,
      "avg_forward_tokens": 5204.14,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 1,
      "exact": 21,
      "generated_variant_counts": {
        "cell_parser": 50,
        "header_aware": 3,
        "row_column_rule": 3,
        "split_fold_unpivot": 3,
        "verified_structural": 3
      },
      "n": 50,
      "policy": "canary_cell_parser_disagree_escalate",
      "program_commit_correct": 1,
      "program_commit_precision": 0.3333333333333333,
      "program_commits": 3,
      "total_forward_tokens": 260207,
      "triggered_tasks": 3
    },
    {
      "accuracy": 0.42,
      "avg_forward_tokens": 5213.86,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 1,
      "exact": 21,
      "generated_variant_counts": {
        "header_aware": 50
      },
      "n": 50,
      "policy": "single_header_aware",
      "program_commit_correct": 5,
      "program_commit_precision": 0.7142857142857143,
      "program_commits": 7,
      "total_forward_tokens": 260693,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.42,
      "avg_forward_tokens": 10777.5,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 0,
      "exact": 21,
      "generated_variant_counts": {
        "cell_parser": 50,
        "row_column_rule": 50,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "prefix3_consensus3",
      "program_commit_correct": 5,
      "program_commit_precision": 0.8333333333333334,
      "program_commits": 6,
      "total_forward_tokens": 538875,
      "triggered_tasks": 0
    },
    {
      "accuracy": 0.42,
      "avg_forward_tokens": 10835.08,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 0,
      "exact": 21,
      "generated_variant_counts": {
        "cell_parser": 25,
        "header_aware": 25,
        "row_column_rule": 25,
        "split_fold_unpivot": 25,
        "verified_structural": 25
      },
      "n": 50,
      "policy": "shape_visible_out_more_cols_all_first_visible",
      "program_commit_correct": 1,
      "program_commit_precision": 0.2,
      "program_commits": 5,
      "total_forward_tokens": 541754,
      "triggered_tasks": 25
    },
    {
      "accuracy": 0.42,
      "avg_forward_tokens": 12650.06,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 0,
      "exact": 21,
      "generated_variant_counts": {
        "cell_parser": 29,
        "header_aware": 29,
        "row_column_rule": 29,
        "split_fold_unpivot": 29,
        "verified_structural": 29
      },
      "n": 50,
      "policy": "shape_visible_out_less_rows_all_first_visible",
      "program_commit_correct": 6,
      "program_commit_precision": 0.6666666666666666,
      "program_commits": 9,
      "total_forward_tokens": 632503,
      "triggered_tasks": 29
    },
    {
      "accuracy": 0.4,
      "avg_forward_tokens": 7392.14,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 1,
      "direct_exact": 21,
      "direct_miss_recoveries": 0,
      "exact": 20,
      "generated_variant_counts": {
        "cell_parser": 50,
        "verified_structural": 50
      },
      "n": 50,
      "policy": "prefix2_consensus2",
      "program_commit_correct": 6,
      "program_commit_precision": 0.75,
      "program_commits": 8,
      "total_forward_tokens": 369607,
      "triggered_tasks": 0
    }
  ],
  "variant_order": [
    "verified_structural",
    "cell_parser",
    "row_column_rule",
    "header_aware",
    "split_fold_unpivot"
  ]
}