{
  "args": {
    "runs": "main_state_w025_repair_s900",
    "eval_size": 256,
    "eval_lengths": "24",
    "eval_seed": 93001,
    "eval_batch_size": 8,
    "output_csv": "experiments/qwen_verifier_guided_slot_repair/analysis/selected_retest_metrics.csv",
    "output_json": "experiments/qwen_verifier_guided_slot_repair/analysis/selected_retest_results.json",
    "append_existing": false
  },
  "rows": [
    {
      "run": "main_state_w025_repair_s900",
      "variant": "copy_trace_state_repair",
      "selected_step": 800,
      "selected_tag": "eval_trace_state_repair_long",
      "state_loss_schedule": "constant",
      "repair_verifier_mode": "state",
      "repair_topk": 3,
      "repair_max_edits": 2,
      "selection_metric": "paired_len24_repair_executor_accuracy",
      "selection_value": 0.8828125,
      "split": "fresh_standard_len24",
      "n": 256.0,
      "direct_accuracy": NaN,
      "direct_target_mass": NaN,
      "executor_accuracy": 0.2734375,
      "executor_target_mass": 0.01030926572275348,
      "init_accuracy": 1.0,
      "init_pos_accuracy": 1.0,
      "op_accuracy": 1.0,
      "arg_accuracy": 0.94873046875,
      "op_pos_accuracy": 1.0,
      "arg_pos_accuracy": 0.9508463541666666,
      "program_exact": 0.26953125,
      "state_accuracy": 0.5576171875,
      "state_all_exact": 0.2734375,
      "state_prefix_fraction": 0.5576171875,
      "state_mean_correct_prefix": 13.3828125,
      "repair_executor_accuracy": 0.8828125,
      "repair_program_exact": 0.87890625,
      "repair_state_accuracy": 0.9065755208333334,
      "repair_state_all_exact": 0.8828125,
      "repair_state_prefix_fraction": 0.9065755208333334,
      "repair_state_mean_correct_prefix": 21.7578125,
      "repair_changed_fraction": 0.609375,
      "repair_found_fraction": 0.8828125,
      "repair_avg_candidates": 1299.0,
      "repair_avg_verified_candidates": 1.30859375
    },
    {
      "run": "main_state_w025_repair_s900",
      "variant": "copy_trace_state_repair",
      "selected_step": 800,
      "selected_tag": "eval_trace_state_repair_long",
      "state_loss_schedule": "constant",
      "repair_verifier_mode": "state",
      "repair_topk": 3,
      "repair_max_edits": 2,
      "selection_metric": "paired_len24_repair_executor_accuracy",
      "selection_value": 0.8828125,
      "split": "fresh_paraphrase_len24",
      "n": 256.0,
      "direct_accuracy": NaN,
      "direct_target_mass": NaN,
      "executor_accuracy": 0.23828125,
      "executor_target_mass": 0.010309285658877343,
      "init_accuracy": 1.0,
      "init_pos_accuracy": 1.0,
      "op_accuracy": 1.0,
      "arg_accuracy": 0.9422200520833334,
      "op_pos_accuracy": 1.0,
      "arg_pos_accuracy": 0.9423828125,
      "program_exact": 0.23828125,
      "state_accuracy": 0.5384114583333334,
      "state_all_exact": 0.23828125,
      "state_prefix_fraction": 0.5367838541666666,
      "state_mean_correct_prefix": 12.8828125,
      "repair_executor_accuracy": 0.86328125,
      "repair_program_exact": 0.85546875,
      "repair_state_accuracy": 0.8873697916666666,
      "repair_state_all_exact": 0.86328125,
      "repair_state_prefix_fraction": 0.8857421875,
      "repair_state_mean_correct_prefix": 21.2578125,
      "repair_changed_fraction": 0.625,
      "repair_found_fraction": 0.86328125,
      "repair_avg_candidates": 1299.0,
      "repair_avg_verified_candidates": 1.1796875
    },
    {
      "run": "main_state_w025_repair_s900",
      "variant": "copy_trace_state_repair",
      "selected_step": 800,
      "selected_tag": "eval_trace_state_repair_long",
      "state_loss_schedule": "constant",
      "repair_verifier_mode": "state",
      "repair_topk": 3,
      "repair_max_edits": 2,
      "selection_metric": "paired_len24_repair_executor_accuracy",
      "selection_value": 0.8828125,
      "split": "fresh_paired_len24",
      "n": 512.0,
      "direct_accuracy": NaN,
      "direct_target_mass": NaN,
      "executor_accuracy": 0.275390625,
      "executor_target_mass": 0.010309299934306182,
      "init_accuracy": 1.0,
      "init_pos_accuracy": 1.0,
      "op_accuracy": 1.0,
      "arg_accuracy": 0.9532063802083334,
      "op_pos_accuracy": 1.0,
      "arg_pos_accuracy": 0.9535319010416666,
      "program_exact": 0.275390625,
      "state_accuracy": 0.5511067708333334,
      "state_all_exact": 0.275390625,
      "state_prefix_fraction": 0.5511067708333334,
      "state_mean_correct_prefix": 13.2265625,
      "repair_executor_accuracy": 0.91015625,
      "repair_program_exact": 0.900390625,
      "repair_state_accuracy": 0.9341634114583334,
      "repair_state_all_exact": 0.91015625,
      "repair_state_prefix_fraction": 0.9341634114583334,
      "repair_state_mean_correct_prefix": 22.419921875,
      "repair_changed_fraction": 0.634765625,
      "repair_found_fraction": 0.91015625,
      "repair_avg_candidates": 1299.0,
      "repair_avg_verified_candidates": 1.35546875,
      "pair_true_answer_consistency": 1.0,
      "pair_true_length_consistency": 1.0,
      "executor_pair_answer_consistency": 0.69140625,
      "executor_pair_both_correct": 0.25,
      "compiler_pair_program_consistency": 0.69140625,
      "compiler_pair_state_consistency": 0.69140625,
      "repair_pair_answer_consistency": 0.921875,
      "repair_pair_both_correct": 0.89453125,
      "repair_pair_program_consistency": 0.91796875,
      "repair_pair_state_consistency": 0.921875
    }
  ]
}