{
  "controller_selection": {
    "num_candidates_after_train_gate": 4468,
    "selected": {
      "dev": {
        "accuracy": 0.5,
        "avg_forward_tokens": 682.1,
        "direct_accuracy": 0.4,
        "direct_correct_losses": 0,
        "direct_exact": 4,
        "direct_miss_recoveries": 1,
        "exact": 5,
        "generated_variant_counts": {
          "split_fold_unpivot": 3
        },
        "n": 10,
        "policy": "candidate_dev",
        "program_commit_correct": 3,
        "program_commit_precision": 1.0,
        "program_commits": 3,
        "total_forward_tokens": 6821
      },
      "policy": {
        "false_action": "direct_only",
        "feature": "direct_out_cols_less_new_in",
        "true_action": "single_split_fold_unpivot"
      },
      "train": {
        "accuracy": 0.3,
        "avg_forward_tokens": 1253.6333333333334,
        "direct_accuracy": 0.3,
        "direct_correct_losses": 0,
        "direct_exact": 9,
        "direct_miss_recoveries": 0,
        "exact": 9,
        "generated_variant_counts": {
          "split_fold_unpivot": 6
        },
        "n": 30,
        "policy": "candidate_train",
        "program_commit_correct": 1,
        "program_commit_precision": 0.3333333333333333,
        "program_commits": 3,
        "total_forward_tokens": 37609
      }
    },
    "top_candidates": [
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 682.1,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "split_fold_unpivot": 3
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 3,
          "program_commit_precision": 1.0,
          "program_commits": 3,
          "total_forward_tokens": 6821
        },
        "policy": {
          "false_action": "direct_only",
          "feature": "direct_out_cols_less_new_in",
          "true_action": "single_split_fold_unpivot"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1253.6333333333334,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 0,
          "exact": 9,
          "generated_variant_counts": {
            "split_fold_unpivot": 6
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 1,
          "program_commit_precision": 0.3333333333333333,
          "program_commits": 3,
          "total_forward_tokens": 37609
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 993.9,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "split_fold_unpivot": 4
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 3,
          "program_commit_precision": 1.0,
          "program_commits": 3,
          "total_forward_tokens": 9939
        },
        "policy": {
          "false_action": "direct_only",
          "feature": "visible_out_less_cols",
          "true_action": "single_split_fold_unpivot"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1308.7,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 0,
          "exact": 9,
          "generated_variant_counts": {
            "split_fold_unpivot": 7
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 1,
          "program_commit_precision": 0.3333333333333333,
          "program_commits": 3,
          "total_forward_tokens": 39261
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1133.9,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "row_column_rule": 4
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 1,
          "program_commit_precision": 1.0,
          "program_commits": 1,
          "total_forward_tokens": 11339
        },
        "policy": {
          "false_action": "single_row_column_rule",
          "feature": "visible_rows_change",
          "true_action": "direct_only"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1106.0333333333333,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 0,
          "exact": 9,
          "generated_variant_counts": {
            "row_column_rule": 7
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 0,
          "program_commit_precision": 0.0,
          "program_commits": 2,
          "total_forward_tokens": 33181
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1158.8,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "split_fold_unpivot": 5
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 3,
          "program_commit_precision": 1.0,
          "program_commits": 3,
          "total_forward_tokens": 11588
        },
        "policy": {
          "false_action": "direct_only",
          "feature": "visible_row_expansion_or_col_contraction",
          "true_action": "single_split_fold_unpivot"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1409.1666666666667,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 0,
          "exact": 9,
          "generated_variant_counts": {
            "split_fold_unpivot": 10
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 3,
          "program_commit_precision": 0.6,
          "program_commits": 5,
          "total_forward_tokens": 42275
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1305.0,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 1,
            "header_aware": 1,
            "row_column_rule": 1,
            "split_fold_unpivot": 3,
            "verified_structural": 1
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 1,
          "program_commit_precision": 1.0,
          "program_commits": 1,
          "total_forward_tokens": 13050
        },
        "policy": {
          "false_action": "direct_only",
          "feature": "direct_out_cols_less_new_in",
          "true_action": "canary_split_fold_unpivot_disagree_escalate"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1478.9666666666667,
          "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": 6,
            "verified_structural": 1
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 0,
          "program_commit_precision": 0.0,
          "program_commits": 1,
          "total_forward_tokens": 44369
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1587.2,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 1,
            "header_aware": 1,
            "row_column_rule": 1,
            "split_fold_unpivot": 4,
            "verified_structural": 1
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 1,
          "program_commit_precision": 1.0,
          "program_commits": 1,
          "total_forward_tokens": 15872
        },
        "policy": {
          "false_action": "canary_split_fold_unpivot_disagree_escalate",
          "feature": "visible_rows_change",
          "true_action": "direct_only"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1067.1333333333334,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 0,
          "exact": 9,
          "generated_variant_counts": {
            "split_fold_unpivot": 7
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 0,
          "program_commit_precision": null,
          "program_commits": 0,
          "total_forward_tokens": 32014
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1601.6,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 2,
            "header_aware": 2,
            "row_column_rule": 3,
            "split_fold_unpivot": 2,
            "verified_structural": 2
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 2,
          "program_commit_precision": 1.0,
          "program_commits": 2,
          "total_forward_tokens": 16016
        },
        "policy": {
          "false_action": "direct_only",
          "feature": "direct_out_cols_less_new_in",
          "true_action": "canary_row_column_rule_disagree_escalate"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1576.9666666666667,
          "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": 6,
            "split_fold_unpivot": 1,
            "verified_structural": 1
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 0,
          "program_commit_precision": 0.0,
          "program_commits": 1,
          "total_forward_tokens": 47309
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1616.8,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 1,
            "header_aware": 1,
            "row_column_rule": 1,
            "split_fold_unpivot": 4,
            "verified_structural": 1
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 1,
          "program_commit_precision": 1.0,
          "program_commits": 1,
          "total_forward_tokens": 16168
        },
        "policy": {
          "false_action": "direct_only",
          "feature": "visible_out_less_cols",
          "true_action": "canary_split_fold_unpivot_disagree_escalate"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1534.0333333333333,
          "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": 7,
            "verified_structural": 1
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 0,
          "program_commit_precision": 0.0,
          "program_commits": 1,
          "total_forward_tokens": 46021
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1716.2,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 1,
            "header_aware": 1,
            "row_column_rule": 4,
            "split_fold_unpivot": 1,
            "verified_structural": 1
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 1,
          "program_commit_precision": 1.0,
          "program_commits": 1,
          "total_forward_tokens": 17162
        },
        "policy": {
          "false_action": "canary_row_column_rule_disagree_escalate",
          "feature": "visible_rows_change",
          "true_action": "direct_only"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1106.0333333333333,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 0,
          "exact": 9,
          "generated_variant_counts": {
            "row_column_rule": 7
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 0,
          "program_commit_precision": null,
          "program_commits": 0,
          "total_forward_tokens": 33181
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1781.7,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 1,
            "header_aware": 1,
            "row_column_rule": 1,
            "split_fold_unpivot": 5,
            "verified_structural": 1
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 1,
          "program_commit_precision": 1.0,
          "program_commits": 1,
          "total_forward_tokens": 17817
        },
        "policy": {
          "false_action": "direct_only",
          "feature": "visible_row_expansion_or_col_contraction",
          "true_action": "canary_split_fold_unpivot_disagree_escalate"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1634.5,
          "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": 10,
            "verified_structural": 1
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 0,
          "program_commit_precision": 0.0,
          "program_commits": 1,
          "total_forward_tokens": 49035
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1811.4,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 1,
            "header_aware": 1,
            "row_column_rule": 1,
            "split_fold_unpivot": 6,
            "verified_structural": 1
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 1,
          "program_commit_precision": 1.0,
          "program_commits": 1,
          "total_forward_tokens": 18114
        },
        "policy": {
          "false_action": "canary_split_fold_unpivot_disagree_escalate",
          "feature": "direct_out_rows_less_new_in",
          "true_action": "direct_only"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1434.8,
          "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": 12,
            "verified_structural": 1
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 0,
          "program_commit_precision": 0.0,
          "program_commits": 1,
          "total_forward_tokens": 43044
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1811.4,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 1,
            "header_aware": 1,
            "row_column_rule": 1,
            "split_fold_unpivot": 6,
            "verified_structural": 1
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 1,
          "program_commit_precision": 1.0,
          "program_commits": 1,
          "total_forward_tokens": 18114
        },
        "policy": {
          "false_action": "canary_split_fold_unpivot_disagree_escalate",
          "feature": "visible_col_expansion_or_row_contraction",
          "true_action": "direct_only"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1770.0333333333333,
          "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": 13,
            "verified_structural": 1
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 0,
          "program_commit_precision": 0.0,
          "program_commits": 1,
          "total_forward_tokens": 53101
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1811.4,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 1,
            "header_aware": 1,
            "row_column_rule": 1,
            "split_fold_unpivot": 6,
            "verified_structural": 1
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 1,
          "program_commit_precision": 1.0,
          "program_commits": 1,
          "total_forward_tokens": 18114
        },
        "policy": {
          "false_action": "canary_split_fold_unpivot_disagree_escalate",
          "feature": "visible_out_less_rows",
          "true_action": "direct_only"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1846.8,
          "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": 15,
            "verified_structural": 1
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 0,
          "program_commit_precision": 0.0,
          "program_commits": 1,
          "total_forward_tokens": 55404
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1865.5,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 7,
            "split_fold_unpivot": 3
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 3,
          "program_commit_precision": 1.0,
          "program_commits": 3,
          "total_forward_tokens": 18655
        },
        "policy": {
          "false_action": "single_cell_parser",
          "feature": "direct_out_cols_less_new_in",
          "true_action": "single_split_fold_unpivot"
        },
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 2610.5333333333333,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 24,
            "split_fold_unpivot": 6
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 6,
          "program_commit_precision": 0.5,
          "program_commits": 12,
          "total_forward_tokens": 78316
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1865.5,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 7,
            "split_fold_unpivot": 3
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 3,
          "program_commit_precision": 1.0,
          "program_commits": 3,
          "total_forward_tokens": 18655
        },
        "policy": {
          "false_action": "canary_cell_parser_disagree_escalate",
          "feature": "direct_out_cols_less_new_in",
          "true_action": "single_split_fold_unpivot"
        },
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 3416.9,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 24,
            "header_aware": 3,
            "row_column_rule": 3,
            "split_fold_unpivot": 9,
            "verified_structural": 3
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 3,
          "program_commit_precision": 0.5,
          "program_commits": 6,
          "total_forward_tokens": 102507
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1865.9,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 2,
            "header_aware": 2,
            "row_column_rule": 4,
            "split_fold_unpivot": 2,
            "verified_structural": 2
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 2,
          "program_commit_precision": 1.0,
          "program_commits": 2,
          "total_forward_tokens": 18659
        },
        "policy": {
          "false_action": "direct_only",
          "feature": "visible_out_less_cols",
          "true_action": "canary_row_column_rule_disagree_escalate"
        },
        "train": {
          "accuracy": 0.3,
          "avg_forward_tokens": 1657.3666666666666,
          "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": 7,
            "split_fold_unpivot": 1,
            "verified_structural": 1
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 0,
          "program_commit_precision": 0.0,
          "program_commits": 1,
          "total_forward_tokens": 49721
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1927.9,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "row_column_rule": 4,
            "verified_structural": 6
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 2,
          "program_commit_precision": 1.0,
          "program_commits": 2,
          "total_forward_tokens": 19279
        },
        "policy": {
          "false_action": "single_row_column_rule",
          "feature": "visible_rows_change",
          "true_action": "single_verified_structural"
        },
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 2731.366666666667,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 11,
          "generated_variant_counts": {
            "row_column_rule": 7,
            "verified_structural": 23
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 4,
          "program_commit_precision": 0.5714285714285714,
          "program_commits": 7,
          "total_forward_tokens": 81941
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1927.9,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "row_column_rule": 4,
            "verified_structural": 6
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 2,
          "program_commit_precision": 1.0,
          "program_commits": 2,
          "total_forward_tokens": 19279
        },
        "policy": {
          "false_action": "single_row_column_rule",
          "feature": "visible_rows_change",
          "true_action": "prefix1_first_visible"
        },
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 2731.366666666667,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 11,
          "generated_variant_counts": {
            "row_column_rule": 7,
            "verified_structural": 23
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 4,
          "program_commit_precision": 0.5714285714285714,
          "program_commits": 7,
          "total_forward_tokens": 81941
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1927.9,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "row_column_rule": 4,
            "verified_structural": 6
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 1,
          "program_commit_precision": 1.0,
          "program_commits": 1,
          "total_forward_tokens": 19279
        },
        "policy": {
          "false_action": "single_row_column_rule",
          "feature": "visible_rows_change",
          "true_action": "canary_verified_structural_disagree_escalate"
        },
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 3145.9666666666667,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 2,
            "header_aware": 2,
            "row_column_rule": 9,
            "split_fold_unpivot": 2,
            "verified_structural": 23
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 2,
          "program_commit_precision": 0.5,
          "program_commits": 4,
          "total_forward_tokens": 94379
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1997.0,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 6,
            "split_fold_unpivot": 4
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 3,
          "program_commit_precision": 1.0,
          "program_commits": 3,
          "total_forward_tokens": 19970
        },
        "policy": {
          "false_action": "single_cell_parser",
          "feature": "visible_out_less_cols",
          "true_action": "single_split_fold_unpivot"
        },
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 2585.0333333333333,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 23,
            "split_fold_unpivot": 7
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 6,
          "program_commit_precision": 0.5,
          "program_commits": 12,
          "total_forward_tokens": 77551
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 1997.0,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 6,
            "split_fold_unpivot": 4
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 3,
          "program_commit_precision": 1.0,
          "program_commits": 3,
          "total_forward_tokens": 19970
        },
        "policy": {
          "false_action": "canary_cell_parser_disagree_escalate",
          "feature": "visible_out_less_cols",
          "true_action": "single_split_fold_unpivot"
        },
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 3391.4,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 23,
            "header_aware": 3,
            "row_column_rule": 3,
            "split_fold_unpivot": 10,
            "verified_structural": 3
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 3,
          "program_commit_precision": 0.5,
          "program_commits": 6,
          "total_forward_tokens": 101742
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 2011.0,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 5,
            "split_fold_unpivot": 5
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 3,
          "program_commit_precision": 1.0,
          "program_commits": 3,
          "total_forward_tokens": 20110
        },
        "policy": {
          "false_action": "single_cell_parser",
          "feature": "visible_row_expansion_or_col_contraction",
          "true_action": "single_split_fold_unpivot"
        },
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 2623.0,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 20,
            "split_fold_unpivot": 10
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 6,
          "program_commit_precision": 0.5,
          "program_commits": 12,
          "total_forward_tokens": 78690
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 2011.0,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 5,
            "split_fold_unpivot": 5
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 3,
          "program_commit_precision": 1.0,
          "program_commits": 3,
          "total_forward_tokens": 20110
        },
        "policy": {
          "false_action": "canary_cell_parser_disagree_escalate",
          "feature": "visible_row_expansion_or_col_contraction",
          "true_action": "single_split_fold_unpivot"
        },
        "train": {
          "accuracy": 0.36666666666666664,
          "avg_forward_tokens": 3429.366666666667,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 0,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 11,
          "generated_variant_counts": {
            "cell_parser": 20,
            "header_aware": 3,
            "row_column_rule": 3,
            "split_fold_unpivot": 13,
            "verified_structural": 3
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 5,
          "program_commit_precision": 0.625,
          "program_commits": 8,
          "total_forward_tokens": 102881
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 2069.9,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 6,
            "row_column_rule": 4
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 2,
          "program_commit_precision": 1.0,
          "program_commits": 2,
          "total_forward_tokens": 20699
        },
        "policy": {
          "false_action": "single_row_column_rule",
          "feature": "visible_rows_change",
          "true_action": "single_cell_parser"
        },
        "train": {
          "accuracy": 0.3333333333333333,
          "avg_forward_tokens": 2730.233333333333,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 1,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 10,
          "generated_variant_counts": {
            "cell_parser": 23,
            "row_column_rule": 7
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 6,
          "program_commit_precision": 0.5,
          "program_commits": 12,
          "total_forward_tokens": 81907
        }
      },
      {
        "dev": {
          "accuracy": 0.5,
          "avg_forward_tokens": 2069.9,
          "direct_accuracy": 0.4,
          "direct_correct_losses": 0,
          "direct_exact": 4,
          "direct_miss_recoveries": 1,
          "exact": 5,
          "generated_variant_counts": {
            "cell_parser": 6,
            "row_column_rule": 4
          },
          "n": 10,
          "policy": "candidate_dev",
          "program_commit_correct": 1,
          "program_commit_precision": 1.0,
          "program_commits": 1,
          "total_forward_tokens": 20699
        },
        "policy": {
          "false_action": "single_row_column_rule",
          "feature": "visible_rows_change",
          "true_action": "canary_cell_parser_disagree_escalate"
        },
        "train": {
          "accuracy": 0.3333333333333333,
          "avg_forward_tokens": 3979.9333333333334,
          "direct_accuracy": 0.3,
          "direct_correct_losses": 1,
          "direct_exact": 9,
          "direct_miss_recoveries": 2,
          "exact": 10,
          "generated_variant_counts": {
            "cell_parser": 23,
            "header_aware": 5,
            "row_column_rule": 12,
            "split_fold_unpivot": 5,
            "verified_structural": 5
          },
          "n": 30,
          "policy": "candidate_train",
          "program_commit_correct": 2,
          "program_commit_precision": 0.2857142857142857,
          "program_commits": 7,
          "total_forward_tokens": 119398
        }
      }
    ]
  },
  "deployable_pareto_front": [
    "direct_only",
    "learned_controller",
    "fixed_visible_out_less_cols_full_else_direct"
  ],
  "family_metrics": {
    "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": 5,
      "recoveries": 0,
      "tokens": 66089
    },
    "potters_wheel_merge_split": {
      "accuracy": 0.4,
      "direct_accuracy": 0.0,
      "direct_exact": 0,
      "exact": 2,
      "losses": 0,
      "n": 5,
      "program_commits": 4,
      "recoveries": 2,
      "tokens": 5696
    },
    "potters_wheel_unfold": {
      "accuracy": 0.4,
      "direct_accuracy": 0.4,
      "direct_exact": 2,
      "exact": 2,
      "losses": 0,
      "n": 5,
      "program_commits": 1,
      "recoveries": 0,
      "tokens": 3987
    },
    "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": 0.4,
      "direct_accuracy": 0.0,
      "direct_exact": 0,
      "exact": 2,
      "losses": 0,
      "n": 5,
      "program_commits": 5,
      "recoveries": 2,
      "tokens": 40823
    }
  },
  "selected_test_metrics": {
    "accuracy": 0.5,
    "avg_forward_tokens": 2969.02,
    "direct_accuracy": 0.42,
    "direct_correct_losses": 0,
    "direct_exact": 21,
    "direct_miss_recoveries": 4,
    "exact": 25,
    "generated_variant_counts": {
      "split_fold_unpivot": 15
    },
    "n": 50,
    "policy": "learned_controller",
    "program_commit_correct": 4,
    "program_commit_precision": 1.0,
    "program_commits": 4,
    "total_forward_tokens": 148451
  },
  "selected_test_policy": "learned_controller",
  "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_full_only_when_helpful",
      "program_commit_correct": 8,
      "program_commit_precision": 1.0,
      "program_commits": 8,
      "total_forward_tokens": 197951
    },
    {
      "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": "fixed_visible_out_less_cols_full_else_direct",
      "program_commit_correct": 8,
      "program_commit_precision": 1.0,
      "program_commits": 8,
      "total_forward_tokens": 231365
    },
    {
      "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": "fixed_direct_out_cols_less_new_in_full_else_direct",
      "program_commit_correct": 8,
      "program_commit_precision": 1.0,
      "program_commits": 8,
      "total_forward_tokens": 450853
    },
    {
      "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": "fixed_visible_cols_change_full_else_direct",
      "program_commit_correct": 9,
      "program_commit_precision": 0.6923076923076923,
      "program_commits": 13,
      "total_forward_tokens": 699208
    },
    {
      "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
    },
    {
      "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
    },
    {
      "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
    },
    {
      "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
    },
    {
      "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
    },
    {
      "accuracy": 0.5,
      "avg_forward_tokens": 2969.02,
      "direct_accuracy": 0.42,
      "direct_correct_losses": 0,
      "direct_exact": 21,
      "direct_miss_recoveries": 4,
      "exact": 25,
      "generated_variant_counts": {
        "split_fold_unpivot": 15
      },
      "n": 50,
      "policy": "learned_controller",
      "program_commit_correct": 4,
      "program_commit_precision": 1.0,
      "program_commits": 4,
      "total_forward_tokens": 148451
    },
    {
      "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
    },
    {
      "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
    },
    {
      "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
    },
    {
      "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
    },
    {
      "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
    },
    {
      "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
    },
    {
      "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
    },
    {
      "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
    }
  ],
  "variant_order": [
    "verified_structural",
    "cell_parser",
    "row_column_rule",
    "header_aware",
    "split_fold_unpivot"
  ]
}