{
  "args": {
    "model_id": "hf-internal-testing/tiny-random-GPT2LMHeadModel",
    "load_in_4bit": false,
    "torch_dtype": "fp32",
    "device_map": "auto",
    "use_lora": false,
    "lora_r": 8,
    "lora_alpha": 16,
    "lora_dropout": 0.05,
    "lora_target_modules": "all-linear",
    "gradient_checkpointing": false,
    "optimizer": "adamw",
    "modulus": 97,
    "max_steps": 3,
    "train_min_len": 1,
    "train_max_len": 12,
    "curriculum_stages": "short:1:2:2,long:2:3:2",
    "answer_train_min_len": 1,
    "answer_train_max_len": 24,
    "eval_lengths": "2,3",
    "train_template_mode": "mixed",
    "paired_train": false,
    "paired_batches": true,
    "paired_template_modes": "standard,paraphrase",
    "pair_size": 2,
    "answer_train_template_mode": "mixed",
    "eval_template_modes": "standard,paraphrase",
    "paired_eval": true,
    "paired_eval_template_modes": "standard,paraphrase",
    "train_size": 16,
    "answer_train_size": 16,
    "eval_size": 4,
    "train_batch_size": 2,
    "eval_batch_size": 2,
    "bootstrap_steps": 0,
    "answer_steps": 0,
    "eval_every": 2,
    "stage_eval_every": 2,
    "lr": 0.0002,
    "answer_lr_scale": 0.1,
    "weight_decay": 0.0,
    "grad_clip": 1.0,
    "head_width": 64,
    "rank_temperature": 1.0,
    "arg_reader_mode": "after_op",
    "arg_window": 8,
    "arg_distance_temperature": 2.0,
    "trace_loss_weight": 1.0,
    "init_trace_loss_weight": 4.0,
    "op_trace_loss_weight": 1.0,
    "arg_trace_loss_weight": 4.0,
    "selection_loss_weight": 1.0,
    "role_pos_weight": 20.0,
    "init_selection_loss_weight": 1.0,
    "op_selection_loss_weight": 1.0,
    "arg_selection_loss_weight": 4.0,
    "executor_loss_weight": 1.0,
    "state_loss_weight": 0.25,
    "state_loss_schedule": "constant",
    "save_eval_checkpoints": true,
    "selection_metric": "paired_len3_repair_executor_accuracy",
    "variants": "copy_trace_state_repair",
    "repair_eval": true,
    "repair_verifier_mode": "state",
    "repair_topk": 3,
    "repair_max_edits": 2,
    "repair_max_pair_arg_slots": 24,
    "repair_include_init": true,
    "repair_include_ops": true,
    "repair_include_args": true,
    "repair_same_step_op_arg": true,
    "max_length": 192,
    "seed": 7,
    "train_seed": -1,
    "output_dir": "experiments/qwen_verifier_guided_slot_repair/runs/smoke_tiny_repair",
    "checkpoint_dir": "",
    "state_loss_weight_by_stage": {
      "short": 0.25,
      "long": 0.25
    }
  },
  "metadata": {
    "python": "3.12.3",
    "platform": "Linux-6.8.0-62-generic-x86_64-with-glibc2.39",
    "torch": "2.8.0+cu128",
    "cuda_available": true,
    "transformers_loader": "AutoModelForCausalLM",
    "peft_installed": true,
    "use_lora": false,
    "lora_r": 8,
    "gpu_name": "NVIDIA RTX 6000 Ada Generation",
    "gpu_vram_gb": 47.383
  },
  "hidden_dim": 32,
  "dataset": {
    "train_size": 16,
    "train_examples": 32,
    "curriculum_stages": [
      {
        "name": "short",
        "min_len": 1,
        "max_len": 2,
        "steps": 2
      },
      {
        "name": "long",
        "min_len": 2,
        "max_len": 3,
        "steps": 2
      }
    ],
    "state_loss_weight_by_stage": {
      "short": 0.25,
      "long": 0.25
    },
    "answer_train_size": 16,
    "eval_size": 4,
    "train_lengths": [
      1,
      12
    ],
    "answer_train_lengths": [
      1,
      24
    ],
    "eval_lengths": [
      2,
      3
    ],
    "train_template_mode": "mixed",
    "paired_train": false,
    "paired_batches": true,
    "paired_template_modes": [
      "standard",
      "paraphrase"
    ],
    "answer_train_template_mode": "mixed",
    "eval_template_modes": [
      "standard",
      "paraphrase"
    ],
    "paired_eval": true,
    "paired_eval_template_modes": [
      "standard",
      "paraphrase"
    ]
  },
  "variants": {
    "copy_trace_state_repair": {
      "variant": "copy_trace_state_repair",
      "train_seconds": 1.050353765487671,
      "checkpoints": [
        {
          "variant": "copy_trace_state_repair",
          "tag": "eval_trace_state_repair_short",
          "step": 1,
          "checkpoint_dir": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_short_step00001",
          "model_checkpoint": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_short_step00001/frozen_backbone.txt",
          "heads_checkpoint": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_short_step00001/heads.pt",
          "selection_metric": "paired_len3_repair_executor_accuracy",
          "selection_value": 0.0
        },
        {
          "variant": "copy_trace_state_repair",
          "tag": "eval_trace_state_repair_short",
          "step": 2,
          "checkpoint_dir": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_short_step00002",
          "model_checkpoint": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_short_step00002/frozen_backbone.txt",
          "heads_checkpoint": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_short_step00002/heads.pt",
          "selection_metric": "paired_len3_repair_executor_accuracy",
          "selection_value": 0.0
        },
        {
          "variant": "copy_trace_state_repair",
          "tag": "eval_trace_state_repair_long",
          "step": 3,
          "checkpoint_dir": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_long_step00003",
          "model_checkpoint": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_long_step00003/frozen_backbone.txt",
          "heads_checkpoint": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_long_step00003/heads.pt",
          "selection_metric": "paired_len3_repair_executor_accuracy",
          "selection_value": 0.0
        },
        {
          "variant": "copy_trace_state_repair",
          "tag": "eval_trace_state_repair_long",
          "step": 4,
          "checkpoint_dir": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_long_step00004",
          "model_checkpoint": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_long_step00004/frozen_backbone.txt",
          "heads_checkpoint": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_long_step00004/heads.pt",
          "selection_metric": "paired_len3_repair_executor_accuracy",
          "selection_value": 0.0
        },
        {
          "variant": "copy_trace_state_repair",
          "tag": "final",
          "step": 4,
          "checkpoint_dir": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/final_step00004",
          "model_checkpoint": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/final_step00004/frozen_backbone.txt",
          "heads_checkpoint": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/final_step00004/heads.pt",
          "selection_metric": "paired_len3_repair_executor_accuracy",
          "selection_value": 0.0
        }
      ],
      "selected_checkpoint": {
        "variant": "copy_trace_state_repair",
        "tag": "eval_trace_state_repair_short",
        "step": 1,
        "checkpoint_dir": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_short_step00001",
        "model_checkpoint": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_short_step00001/frozen_backbone.txt",
        "heads_checkpoint": "large_artifacts/qwen_verifier_guided_slot_repair/checkpoints/smoke_tiny_repair/copy_trace_state_repair/eval_trace_state_repair_short_step00001/heads.pt",
        "selection_metric": "paired_len3_repair_executor_accuracy",
        "selection_value": 0.0
      },
      "train_log": [
        {
          "variant": "copy_trace_state_repair",
          "stage": "trace_state_repair_short",
          "local_step": 1,
          "step": 1,
          "loss": 85.06532287597656,
          "trace_loss_active": true,
          "executor_loss_active": true,
          "state_loss_active": true,
          "state_loss_weight": 0.25,
          "lr_scale": 1.0,
          "executor_loss": 7.591366767883301,
          "state_loss": 5.169837951660156,
          "state_train_accuracy": 0.0,
          "init_loss": 0.905046284198761,
          "op_loss": 0.0,
          "arg_loss": 5.739584922790527,
          "init_attn_loss": 4.4479522705078125,
          "op_attn_loss": 7.888978004455566,
          "arg_attn_loss": 9.044265747070312,
          "role_loss": 1.0889782905578613,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.0,
          "standard_len2_executor_target_mass": 0.008736009613130591,
          "standard_len2_init_accuracy": 0.5,
          "standard_len2_init_pos_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.625,
          "standard_len2_arg_accuracy": 0.0,
          "standard_len2_op_pos_accuracy": 0.0,
          "standard_len2_arg_pos_accuracy": 0.0,
          "standard_len2_program_exact": 0.0,
          "standard_len2_state_accuracy": 0.125,
          "standard_len2_state_all_exact": 0.0,
          "standard_len2_state_prefix_fraction": 0.125,
          "standard_len2_state_mean_correct_prefix": 0.25,
          "standard_len2_repair_executor_accuracy": 0.5,
          "standard_len2_repair_program_exact": 0.25,
          "standard_len2_repair_state_accuracy": 0.5,
          "standard_len2_repair_state_all_exact": 0.5,
          "standard_len2_repair_state_prefix_fraction": 0.5,
          "standard_len2_repair_state_mean_correct_prefix": 1.0,
          "standard_len2_repair_changed_fraction": 0.5,
          "standard_len2_repair_found_fraction": 0.5,
          "standard_len2_repair_avg_candidates": 23.0,
          "standard_len2_repair_avg_verified_candidates": 0.5,
          "standard_len3_direct_accuracy": NaN,
          "standard_len3_direct_target_mass": NaN,
          "standard_len3_executor_accuracy": 0.0,
          "standard_len3_executor_target_mass": 0.04365433054044843,
          "standard_len3_init_accuracy": 0.25,
          "standard_len3_init_pos_accuracy": 0.0,
          "standard_len3_op_accuracy": 0.6666666666666666,
          "standard_len3_arg_accuracy": 0.0,
          "standard_len3_op_pos_accuracy": 0.0,
          "standard_len3_arg_pos_accuracy": 0.0,
          "standard_len3_program_exact": 0.0,
          "standard_len3_state_accuracy": 0.16666666666666666,
          "standard_len3_state_all_exact": 0.0,
          "standard_len3_state_prefix_fraction": 0.16666666666666666,
          "standard_len3_state_mean_correct_prefix": 0.5,
          "standard_len3_repair_executor_accuracy": 0.0,
          "standard_len3_repair_program_exact": 0.0,
          "standard_len3_repair_state_accuracy": 0.16666666666666666,
          "standard_len3_repair_state_all_exact": 0.0,
          "standard_len3_repair_state_prefix_fraction": 0.16666666666666666,
          "standard_len3_repair_state_mean_correct_prefix": 0.5,
          "standard_len3_repair_changed_fraction": 0.0,
          "standard_len3_repair_found_fraction": 0.0,
          "standard_len3_repair_avg_candidates": 39.0,
          "standard_len3_repair_avg_verified_candidates": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.0,
          "paraphrase_len2_executor_target_mass": 0.000660546647850424,
          "paraphrase_len2_init_accuracy": 0.5,
          "paraphrase_len2_init_pos_accuracy": 0.0,
          "paraphrase_len2_op_accuracy": 0.625,
          "paraphrase_len2_arg_accuracy": 0.0,
          "paraphrase_len2_op_pos_accuracy": 0.0,
          "paraphrase_len2_arg_pos_accuracy": 0.0,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len2_state_accuracy": 0.25,
          "paraphrase_len2_state_all_exact": 0.0,
          "paraphrase_len2_state_prefix_fraction": 0.25,
          "paraphrase_len2_state_mean_correct_prefix": 0.5,
          "paraphrase_len2_repair_executor_accuracy": 0.75,
          "paraphrase_len2_repair_program_exact": 0.25,
          "paraphrase_len2_repair_state_accuracy": 0.75,
          "paraphrase_len2_repair_state_all_exact": 0.75,
          "paraphrase_len2_repair_state_prefix_fraction": 0.75,
          "paraphrase_len2_repair_state_mean_correct_prefix": 1.5,
          "paraphrase_len2_repair_changed_fraction": 0.75,
          "paraphrase_len2_repair_found_fraction": 0.75,
          "paraphrase_len2_repair_avg_candidates": 23.0,
          "paraphrase_len2_repair_avg_verified_candidates": 0.75,
          "paraphrase_len3_direct_accuracy": NaN,
          "paraphrase_len3_direct_target_mass": NaN,
          "paraphrase_len3_executor_accuracy": 0.0,
          "paraphrase_len3_executor_target_mass": 0.0022737649269402027,
          "paraphrase_len3_init_accuracy": 0.25,
          "paraphrase_len3_init_pos_accuracy": 0.0,
          "paraphrase_len3_op_accuracy": 0.5,
          "paraphrase_len3_arg_accuracy": 0.0,
          "paraphrase_len3_op_pos_accuracy": 0.0,
          "paraphrase_len3_arg_pos_accuracy": 0.0,
          "paraphrase_len3_program_exact": 0.0,
          "paraphrase_len3_state_accuracy": 0.0,
          "paraphrase_len3_state_all_exact": 0.0,
          "paraphrase_len3_state_prefix_fraction": 0.0,
          "paraphrase_len3_state_mean_correct_prefix": 0.0,
          "paraphrase_len3_repair_executor_accuracy": 0.0,
          "paraphrase_len3_repair_program_exact": 0.0,
          "paraphrase_len3_repair_state_accuracy": 0.0,
          "paraphrase_len3_repair_state_all_exact": 0.0,
          "paraphrase_len3_repair_state_prefix_fraction": 0.0,
          "paraphrase_len3_repair_state_mean_correct_prefix": 0.0,
          "paraphrase_len3_repair_changed_fraction": 0.0,
          "paraphrase_len3_repair_found_fraction": 0.0,
          "paraphrase_len3_repair_avg_candidates": 39.0,
          "paraphrase_len3_repair_avg_verified_candidates": 0.0,
          "paired_len2_direct_accuracy": NaN,
          "paired_len2_direct_target_mass": NaN,
          "paired_len2_executor_accuracy": 0.25,
          "paired_len2_executor_target_mass": 0.12585168680743664,
          "paired_len2_init_accuracy": 0.625,
          "paired_len2_init_pos_accuracy": 0.0,
          "paired_len2_op_accuracy": 0.5,
          "paired_len2_arg_accuracy": 0.0,
          "paired_len2_op_pos_accuracy": 0.0,
          "paired_len2_arg_pos_accuracy": 0.0,
          "paired_len2_program_exact": 0.0,
          "paired_len2_state_accuracy": 0.1875,
          "paired_len2_state_all_exact": 0.0,
          "paired_len2_state_prefix_fraction": 0.0625,
          "paired_len2_state_mean_correct_prefix": 0.125,
          "paired_len2_repair_executor_accuracy": 0.625,
          "paired_len2_repair_program_exact": 0.0,
          "paired_len2_repair_state_accuracy": 0.5,
          "paired_len2_repair_state_all_exact": 0.375,
          "paired_len2_repair_state_prefix_fraction": 0.375,
          "paired_len2_repair_state_mean_correct_prefix": 0.75,
          "paired_len2_repair_changed_fraction": 0.375,
          "paired_len2_repair_found_fraction": 0.375,
          "paired_len2_repair_avg_candidates": 23.0,
          "paired_len2_repair_avg_verified_candidates": 0.375,
          "paired_len2_pair_true_answer_consistency": 1.0,
          "paired_len2_pair_true_length_consistency": 1.0,
          "paired_len2_executor_pair_answer_consistency": 0.75,
          "paired_len2_executor_pair_both_correct": 0.25,
          "paired_len2_compiler_pair_program_consistency": 0.75,
          "paired_len2_compiler_pair_state_consistency": 0.75,
          "paired_len2_repair_pair_answer_consistency": 0.75,
          "paired_len2_repair_pair_both_correct": 0.5,
          "paired_len2_repair_pair_program_consistency": 0.75,
          "paired_len2_repair_pair_state_consistency": 0.75,
          "paired_len3_direct_accuracy": NaN,
          "paired_len3_direct_target_mass": NaN,
          "paired_len3_executor_accuracy": 0.0,
          "paired_len3_executor_target_mass": 0.00029581269313894154,
          "paired_len3_init_accuracy": 0.375,
          "paired_len3_init_pos_accuracy": 0.0,
          "paired_len3_op_accuracy": 0.5,
          "paired_len3_arg_accuracy": 0.0,
          "paired_len3_op_pos_accuracy": 0.0,
          "paired_len3_arg_pos_accuracy": 0.0,
          "paired_len3_program_exact": 0.0,
          "paired_len3_state_accuracy": 0.041666666666666664,
          "paired_len3_state_all_exact": 0.0,
          "paired_len3_state_prefix_fraction": 0.041666666666666664,
          "paired_len3_state_mean_correct_prefix": 0.125,
          "paired_len3_repair_executor_accuracy": 0.0,
          "paired_len3_repair_program_exact": 0.0,
          "paired_len3_repair_state_accuracy": 0.041666666666666664,
          "paired_len3_repair_state_all_exact": 0.0,
          "paired_len3_repair_state_prefix_fraction": 0.041666666666666664,
          "paired_len3_repair_state_mean_correct_prefix": 0.125,
          "paired_len3_repair_changed_fraction": 0.0,
          "paired_len3_repair_found_fraction": 0.0,
          "paired_len3_repair_avg_candidates": 39.0,
          "paired_len3_repair_avg_verified_candidates": 0.0,
          "paired_len3_pair_true_answer_consistency": 1.0,
          "paired_len3_pair_true_length_consistency": 1.0,
          "paired_len3_executor_pair_answer_consistency": 0.25,
          "paired_len3_executor_pair_both_correct": 0.0,
          "paired_len3_compiler_pair_program_consistency": 0.25,
          "paired_len3_compiler_pair_state_consistency": 0.25,
          "paired_len3_repair_pair_answer_consistency": 0.25,
          "paired_len3_repair_pair_both_correct": 0.0,
          "paired_len3_repair_pair_program_consistency": 0.25,
          "paired_len3_repair_pair_state_consistency": 0.25
        },
        {
          "variant": "copy_trace_state_repair",
          "stage": "trace_state_repair_short",
          "local_step": 2,
          "step": 2,
          "loss": 74.40383911132812,
          "trace_loss_active": true,
          "executor_loss_active": true,
          "state_loss_active": true,
          "state_loss_weight": 0.25,
          "lr_scale": 1.0,
          "executor_loss": 2.625028371810913,
          "state_loss": 2.217552900314331,
          "state_train_accuracy": 0.0,
          "init_loss": 0.5603394508361816,
          "op_loss": 0.0,
          "arg_loss": 5.5344929695129395,
          "init_attn_loss": 4.366597652435303,
          "op_attn_loss": 7.595768451690674,
          "arg_attn_loss": 8.47341251373291,
          "role_loss": 0.9890756011009216,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.0,
          "standard_len2_executor_target_mass": 0.008940440167862107,
          "standard_len2_init_accuracy": 0.5,
          "standard_len2_init_pos_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.625,
          "standard_len2_arg_accuracy": 0.0,
          "standard_len2_op_pos_accuracy": 0.0,
          "standard_len2_arg_pos_accuracy": 0.0,
          "standard_len2_program_exact": 0.0,
          "standard_len2_state_accuracy": 0.125,
          "standard_len2_state_all_exact": 0.0,
          "standard_len2_state_prefix_fraction": 0.125,
          "standard_len2_state_mean_correct_prefix": 0.25,
          "standard_len2_repair_executor_accuracy": 0.5,
          "standard_len2_repair_program_exact": 0.25,
          "standard_len2_repair_state_accuracy": 0.5,
          "standard_len2_repair_state_all_exact": 0.5,
          "standard_len2_repair_state_prefix_fraction": 0.5,
          "standard_len2_repair_state_mean_correct_prefix": 1.0,
          "standard_len2_repair_changed_fraction": 0.5,
          "standard_len2_repair_found_fraction": 0.5,
          "standard_len2_repair_avg_candidates": 23.0,
          "standard_len2_repair_avg_verified_candidates": 0.5,
          "standard_len3_direct_accuracy": NaN,
          "standard_len3_direct_target_mass": NaN,
          "standard_len3_executor_accuracy": 0.0,
          "standard_len3_executor_target_mass": 0.043237943667918444,
          "standard_len3_init_accuracy": 0.25,
          "standard_len3_init_pos_accuracy": 0.0,
          "standard_len3_op_accuracy": 0.6666666666666666,
          "standard_len3_arg_accuracy": 0.0,
          "standard_len3_op_pos_accuracy": 0.0,
          "standard_len3_arg_pos_accuracy": 0.0,
          "standard_len3_program_exact": 0.0,
          "standard_len3_state_accuracy": 0.16666666666666666,
          "standard_len3_state_all_exact": 0.0,
          "standard_len3_state_prefix_fraction": 0.16666666666666666,
          "standard_len3_state_mean_correct_prefix": 0.5,
          "standard_len3_repair_executor_accuracy": 0.0,
          "standard_len3_repair_program_exact": 0.0,
          "standard_len3_repair_state_accuracy": 0.16666666666666666,
          "standard_len3_repair_state_all_exact": 0.0,
          "standard_len3_repair_state_prefix_fraction": 0.16666666666666666,
          "standard_len3_repair_state_mean_correct_prefix": 0.5,
          "standard_len3_repair_changed_fraction": 0.0,
          "standard_len3_repair_found_fraction": 0.0,
          "standard_len3_repair_avg_candidates": 39.0,
          "standard_len3_repair_avg_verified_candidates": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.0,
          "paraphrase_len2_executor_target_mass": 0.0007054925445117988,
          "paraphrase_len2_init_accuracy": 0.5,
          "paraphrase_len2_init_pos_accuracy": 0.0,
          "paraphrase_len2_op_accuracy": 0.625,
          "paraphrase_len2_arg_accuracy": 0.0,
          "paraphrase_len2_op_pos_accuracy": 0.0,
          "paraphrase_len2_arg_pos_accuracy": 0.0,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len2_state_accuracy": 0.25,
          "paraphrase_len2_state_all_exact": 0.0,
          "paraphrase_len2_state_prefix_fraction": 0.25,
          "paraphrase_len2_state_mean_correct_prefix": 0.5,
          "paraphrase_len2_repair_executor_accuracy": 0.75,
          "paraphrase_len2_repair_program_exact": 0.25,
          "paraphrase_len2_repair_state_accuracy": 0.75,
          "paraphrase_len2_repair_state_all_exact": 0.75,
          "paraphrase_len2_repair_state_prefix_fraction": 0.75,
          "paraphrase_len2_repair_state_mean_correct_prefix": 1.5,
          "paraphrase_len2_repair_changed_fraction": 0.75,
          "paraphrase_len2_repair_found_fraction": 0.75,
          "paraphrase_len2_repair_avg_candidates": 23.0,
          "paraphrase_len2_repair_avg_verified_candidates": 0.75,
          "paraphrase_len3_direct_accuracy": NaN,
          "paraphrase_len3_direct_target_mass": NaN,
          "paraphrase_len3_executor_accuracy": 0.0,
          "paraphrase_len3_executor_target_mass": 0.002418454736471176,
          "paraphrase_len3_init_accuracy": 0.25,
          "paraphrase_len3_init_pos_accuracy": 0.0,
          "paraphrase_len3_op_accuracy": 0.5,
          "paraphrase_len3_arg_accuracy": 0.0,
          "paraphrase_len3_op_pos_accuracy": 0.0,
          "paraphrase_len3_arg_pos_accuracy": 0.0,
          "paraphrase_len3_program_exact": 0.0,
          "paraphrase_len3_state_accuracy": 0.0,
          "paraphrase_len3_state_all_exact": 0.0,
          "paraphrase_len3_state_prefix_fraction": 0.0,
          "paraphrase_len3_state_mean_correct_prefix": 0.0,
          "paraphrase_len3_repair_executor_accuracy": 0.0,
          "paraphrase_len3_repair_program_exact": 0.0,
          "paraphrase_len3_repair_state_accuracy": 0.0,
          "paraphrase_len3_repair_state_all_exact": 0.0,
          "paraphrase_len3_repair_state_prefix_fraction": 0.0,
          "paraphrase_len3_repair_state_mean_correct_prefix": 0.0,
          "paraphrase_len3_repair_changed_fraction": 0.0,
          "paraphrase_len3_repair_found_fraction": 0.0,
          "paraphrase_len3_repair_avg_candidates": 39.0,
          "paraphrase_len3_repair_avg_verified_candidates": 0.0,
          "paired_len2_direct_accuracy": NaN,
          "paired_len2_direct_target_mass": NaN,
          "paired_len2_executor_accuracy": 0.125,
          "paired_len2_executor_target_mass": 0.12524424261209788,
          "paired_len2_init_accuracy": 0.5,
          "paired_len2_init_pos_accuracy": 0.0,
          "paired_len2_op_accuracy": 0.5,
          "paired_len2_arg_accuracy": 0.0,
          "paired_len2_op_pos_accuracy": 0.0,
          "paired_len2_arg_pos_accuracy": 0.0,
          "paired_len2_program_exact": 0.0,
          "paired_len2_state_accuracy": 0.1875,
          "paired_len2_state_all_exact": 0.0,
          "paired_len2_state_prefix_fraction": 0.125,
          "paired_len2_state_mean_correct_prefix": 0.25,
          "paired_len2_repair_executor_accuracy": 0.625,
          "paired_len2_repair_program_exact": 0.0,
          "paired_len2_repair_state_accuracy": 0.5625,
          "paired_len2_repair_state_all_exact": 0.5,
          "paired_len2_repair_state_prefix_fraction": 0.5,
          "paired_len2_repair_state_mean_correct_prefix": 1.0,
          "paired_len2_repair_changed_fraction": 0.5,
          "paired_len2_repair_found_fraction": 0.5,
          "paired_len2_repair_avg_candidates": 23.0,
          "paired_len2_repair_avg_verified_candidates": 0.5,
          "paired_len2_pair_true_answer_consistency": 1.0,
          "paired_len2_pair_true_length_consistency": 1.0,
          "paired_len2_executor_pair_answer_consistency": 0.5,
          "paired_len2_executor_pair_both_correct": 0.0,
          "paired_len2_compiler_pair_program_consistency": 0.5,
          "paired_len2_compiler_pair_state_consistency": 0.5,
          "paired_len2_repair_pair_answer_consistency": 0.75,
          "paired_len2_repair_pair_both_correct": 0.5,
          "paired_len2_repair_pair_program_consistency": 0.5,
          "paired_len2_repair_pair_state_consistency": 0.5,
          "paired_len3_direct_accuracy": NaN,
          "paired_len3_direct_target_mass": NaN,
          "paired_len3_executor_accuracy": 0.0,
          "paired_len3_executor_target_mass": 0.00031344190369964053,
          "paired_len3_init_accuracy": 0.375,
          "paired_len3_init_pos_accuracy": 0.0,
          "paired_len3_op_accuracy": 0.5,
          "paired_len3_arg_accuracy": 0.0,
          "paired_len3_op_pos_accuracy": 0.0,
          "paired_len3_arg_pos_accuracy": 0.0,
          "paired_len3_program_exact": 0.0,
          "paired_len3_state_accuracy": 0.041666666666666664,
          "paired_len3_state_all_exact": 0.0,
          "paired_len3_state_prefix_fraction": 0.041666666666666664,
          "paired_len3_state_mean_correct_prefix": 0.125,
          "paired_len3_repair_executor_accuracy": 0.0,
          "paired_len3_repair_program_exact": 0.0,
          "paired_len3_repair_state_accuracy": 0.041666666666666664,
          "paired_len3_repair_state_all_exact": 0.0,
          "paired_len3_repair_state_prefix_fraction": 0.041666666666666664,
          "paired_len3_repair_state_mean_correct_prefix": 0.125,
          "paired_len3_repair_changed_fraction": 0.0,
          "paired_len3_repair_found_fraction": 0.0,
          "paired_len3_repair_avg_candidates": 39.0,
          "paired_len3_repair_avg_verified_candidates": 0.0,
          "paired_len3_pair_true_answer_consistency": 1.0,
          "paired_len3_pair_true_length_consistency": 1.0,
          "paired_len3_executor_pair_answer_consistency": 0.5,
          "paired_len3_executor_pair_both_correct": 0.0,
          "paired_len3_compiler_pair_program_consistency": 0.5,
          "paired_len3_compiler_pair_state_consistency": 0.5,
          "paired_len3_repair_pair_answer_consistency": 0.5,
          "paired_len3_repair_pair_both_correct": 0.0,
          "paired_len3_repair_pair_program_consistency": 0.5,
          "paired_len3_repair_pair_state_consistency": 0.5
        },
        {
          "variant": "copy_trace_state_repair",
          "stage": "trace_state_repair_long",
          "local_step": 1,
          "step": 3,
          "loss": 73.86238098144531,
          "trace_loss_active": true,
          "executor_loss_active": true,
          "state_loss_active": true,
          "state_loss_weight": 0.25,
          "lr_scale": 1.0,
          "executor_loss": 11.085697174072266,
          "state_loss": 7.243434906005859,
          "state_train_accuracy": 0.0,
          "init_loss": 1.1411769390106201,
          "op_loss": 1.3483272790908813,
          "arg_loss": 3.696762800216675,
          "init_attn_loss": 4.459924697875977,
          "op_attn_loss": 6.548429012298584,
          "arg_attn_loss": 6.995225429534912,
          "role_loss": 1.2764781713485718,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.0,
          "standard_len2_executor_target_mass": 0.009148675995675148,
          "standard_len2_init_accuracy": 0.5,
          "standard_len2_init_pos_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.625,
          "standard_len2_arg_accuracy": 0.0,
          "standard_len2_op_pos_accuracy": 0.0,
          "standard_len2_arg_pos_accuracy": 0.0,
          "standard_len2_program_exact": 0.0,
          "standard_len2_state_accuracy": 0.125,
          "standard_len2_state_all_exact": 0.0,
          "standard_len2_state_prefix_fraction": 0.125,
          "standard_len2_state_mean_correct_prefix": 0.25,
          "standard_len2_repair_executor_accuracy": 0.5,
          "standard_len2_repair_program_exact": 0.25,
          "standard_len2_repair_state_accuracy": 0.5,
          "standard_len2_repair_state_all_exact": 0.5,
          "standard_len2_repair_state_prefix_fraction": 0.5,
          "standard_len2_repair_state_mean_correct_prefix": 1.0,
          "standard_len2_repair_changed_fraction": 0.5,
          "standard_len2_repair_found_fraction": 0.5,
          "standard_len2_repair_avg_candidates": 23.0,
          "standard_len2_repair_avg_verified_candidates": 0.5,
          "standard_len3_direct_accuracy": NaN,
          "standard_len3_direct_target_mass": NaN,
          "standard_len3_executor_accuracy": 0.0,
          "standard_len3_executor_target_mass": 0.04282327089458704,
          "standard_len3_init_accuracy": 0.25,
          "standard_len3_init_pos_accuracy": 0.0,
          "standard_len3_op_accuracy": 0.6666666666666666,
          "standard_len3_arg_accuracy": 0.0,
          "standard_len3_op_pos_accuracy": 0.0,
          "standard_len3_arg_pos_accuracy": 0.0,
          "standard_len3_program_exact": 0.0,
          "standard_len3_state_accuracy": 0.16666666666666666,
          "standard_len3_state_all_exact": 0.0,
          "standard_len3_state_prefix_fraction": 0.16666666666666666,
          "standard_len3_state_mean_correct_prefix": 0.5,
          "standard_len3_repair_executor_accuracy": 0.0,
          "standard_len3_repair_program_exact": 0.0,
          "standard_len3_repair_state_accuracy": 0.16666666666666666,
          "standard_len3_repair_state_all_exact": 0.0,
          "standard_len3_repair_state_prefix_fraction": 0.16666666666666666,
          "standard_len3_repair_state_mean_correct_prefix": 0.5,
          "standard_len3_repair_changed_fraction": 0.0,
          "standard_len3_repair_found_fraction": 0.0,
          "standard_len3_repair_avg_candidates": 39.0,
          "standard_len3_repair_avg_verified_candidates": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.0,
          "paraphrase_len2_executor_target_mass": 0.0007538771205872763,
          "paraphrase_len2_init_accuracy": 0.5,
          "paraphrase_len2_init_pos_accuracy": 0.0,
          "paraphrase_len2_op_accuracy": 0.625,
          "paraphrase_len2_arg_accuracy": 0.0,
          "paraphrase_len2_op_pos_accuracy": 0.0,
          "paraphrase_len2_arg_pos_accuracy": 0.0,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len2_state_accuracy": 0.25,
          "paraphrase_len2_state_all_exact": 0.0,
          "paraphrase_len2_state_prefix_fraction": 0.25,
          "paraphrase_len2_state_mean_correct_prefix": 0.5,
          "paraphrase_len2_repair_executor_accuracy": 0.75,
          "paraphrase_len2_repair_program_exact": 0.25,
          "paraphrase_len2_repair_state_accuracy": 0.75,
          "paraphrase_len2_repair_state_all_exact": 0.75,
          "paraphrase_len2_repair_state_prefix_fraction": 0.75,
          "paraphrase_len2_repair_state_mean_correct_prefix": 1.5,
          "paraphrase_len2_repair_changed_fraction": 0.75,
          "paraphrase_len2_repair_found_fraction": 0.75,
          "paraphrase_len2_repair_avg_candidates": 23.0,
          "paraphrase_len2_repair_avg_verified_candidates": 0.75,
          "paraphrase_len3_direct_accuracy": NaN,
          "paraphrase_len3_direct_target_mass": NaN,
          "paraphrase_len3_executor_accuracy": 0.0,
          "paraphrase_len3_executor_target_mass": 0.0025715697556734085,
          "paraphrase_len3_init_accuracy": 0.25,
          "paraphrase_len3_init_pos_accuracy": 0.0,
          "paraphrase_len3_op_accuracy": 0.5,
          "paraphrase_len3_arg_accuracy": 0.0,
          "paraphrase_len3_op_pos_accuracy": 0.0,
          "paraphrase_len3_arg_pos_accuracy": 0.0,
          "paraphrase_len3_program_exact": 0.0,
          "paraphrase_len3_state_accuracy": 0.0,
          "paraphrase_len3_state_all_exact": 0.0,
          "paraphrase_len3_state_prefix_fraction": 0.0,
          "paraphrase_len3_state_mean_correct_prefix": 0.0,
          "paraphrase_len3_repair_executor_accuracy": 0.0,
          "paraphrase_len3_repair_program_exact": 0.0,
          "paraphrase_len3_repair_state_accuracy": 0.0,
          "paraphrase_len3_repair_state_all_exact": 0.0,
          "paraphrase_len3_repair_state_prefix_fraction": 0.0,
          "paraphrase_len3_repair_state_mean_correct_prefix": 0.0,
          "paraphrase_len3_repair_changed_fraction": 0.0,
          "paraphrase_len3_repair_found_fraction": 0.0,
          "paraphrase_len3_repair_avg_candidates": 39.0,
          "paraphrase_len3_repair_avg_verified_candidates": 0.0,
          "paired_len2_direct_accuracy": NaN,
          "paired_len2_direct_target_mass": NaN,
          "paired_len2_executor_accuracy": 0.125,
          "paired_len2_executor_target_mass": 0.12464633341369336,
          "paired_len2_init_accuracy": 0.5,
          "paired_len2_init_pos_accuracy": 0.0,
          "paired_len2_op_accuracy": 0.5,
          "paired_len2_arg_accuracy": 0.0,
          "paired_len2_op_pos_accuracy": 0.0,
          "paired_len2_arg_pos_accuracy": 0.0,
          "paired_len2_program_exact": 0.0,
          "paired_len2_state_accuracy": 0.1875,
          "paired_len2_state_all_exact": 0.0,
          "paired_len2_state_prefix_fraction": 0.125,
          "paired_len2_state_mean_correct_prefix": 0.25,
          "paired_len2_repair_executor_accuracy": 0.625,
          "paired_len2_repair_program_exact": 0.0,
          "paired_len2_repair_state_accuracy": 0.5625,
          "paired_len2_repair_state_all_exact": 0.5,
          "paired_len2_repair_state_prefix_fraction": 0.5,
          "paired_len2_repair_state_mean_correct_prefix": 1.0,
          "paired_len2_repair_changed_fraction": 0.5,
          "paired_len2_repair_found_fraction": 0.5,
          "paired_len2_repair_avg_candidates": 23.0,
          "paired_len2_repair_avg_verified_candidates": 0.5,
          "paired_len2_pair_true_answer_consistency": 1.0,
          "paired_len2_pair_true_length_consistency": 1.0,
          "paired_len2_executor_pair_answer_consistency": 0.5,
          "paired_len2_executor_pair_both_correct": 0.0,
          "paired_len2_compiler_pair_program_consistency": 0.5,
          "paired_len2_compiler_pair_state_consistency": 0.5,
          "paired_len2_repair_pair_answer_consistency": 0.75,
          "paired_len2_repair_pair_both_correct": 0.5,
          "paired_len2_repair_pair_program_consistency": 0.5,
          "paired_len2_repair_pair_state_consistency": 0.5,
          "paired_len3_direct_accuracy": NaN,
          "paired_len3_direct_target_mass": NaN,
          "paired_len3_executor_accuracy": 0.0,
          "paired_len3_executor_target_mass": 0.00033180080754391383,
          "paired_len3_init_accuracy": 0.375,
          "paired_len3_init_pos_accuracy": 0.0,
          "paired_len3_op_accuracy": 0.5,
          "paired_len3_arg_accuracy": 0.0,
          "paired_len3_op_pos_accuracy": 0.0,
          "paired_len3_arg_pos_accuracy": 0.0,
          "paired_len3_program_exact": 0.0,
          "paired_len3_state_accuracy": 0.041666666666666664,
          "paired_len3_state_all_exact": 0.0,
          "paired_len3_state_prefix_fraction": 0.041666666666666664,
          "paired_len3_state_mean_correct_prefix": 0.125,
          "paired_len3_repair_executor_accuracy": 0.0,
          "paired_len3_repair_program_exact": 0.0,
          "paired_len3_repair_state_accuracy": 0.041666666666666664,
          "paired_len3_repair_state_all_exact": 0.0,
          "paired_len3_repair_state_prefix_fraction": 0.041666666666666664,
          "paired_len3_repair_state_mean_correct_prefix": 0.125,
          "paired_len3_repair_changed_fraction": 0.0,
          "paired_len3_repair_found_fraction": 0.0,
          "paired_len3_repair_avg_candidates": 39.0,
          "paired_len3_repair_avg_verified_candidates": 0.0,
          "paired_len3_pair_true_answer_consistency": 1.0,
          "paired_len3_pair_true_length_consistency": 1.0,
          "paired_len3_executor_pair_answer_consistency": 0.5,
          "paired_len3_executor_pair_both_correct": 0.0,
          "paired_len3_compiler_pair_program_consistency": 0.5,
          "paired_len3_compiler_pair_state_consistency": 0.5,
          "paired_len3_repair_pair_answer_consistency": 0.5,
          "paired_len3_repair_pair_both_correct": 0.0,
          "paired_len3_repair_pair_program_consistency": 0.5,
          "paired_len3_repair_pair_state_consistency": 0.5
        },
        {
          "variant": "copy_trace_state_repair",
          "stage": "trace_state_repair_long",
          "local_step": 2,
          "step": 4,
          "loss": 72.03813171386719,
          "trace_loss_active": true,
          "executor_loss_active": true,
          "state_loss_active": true,
          "state_loss_weight": 0.25,
          "lr_scale": 1.0,
          "executor_loss": 10.625743865966797,
          "state_loss": 7.206523895263672,
          "state_train_accuracy": 0.0,
          "init_loss": 1.2405595779418945,
          "op_loss": 1.1733614206314087,
          "arg_loss": 3.7386245727539062,
          "init_attn_loss": 4.539708137512207,
          "op_attn_loss": 6.062803268432617,
          "arg_attn_loss": 6.635518550872803,
          "role_loss": 1.3760753870010376,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.0,
          "standard_len2_executor_target_mass": 0.009355536936709541,
          "standard_len2_init_accuracy": 0.5,
          "standard_len2_init_pos_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.625,
          "standard_len2_arg_accuracy": 0.0,
          "standard_len2_op_pos_accuracy": 0.0,
          "standard_len2_arg_pos_accuracy": 0.0,
          "standard_len2_program_exact": 0.0,
          "standard_len2_state_accuracy": 0.125,
          "standard_len2_state_all_exact": 0.0,
          "standard_len2_state_prefix_fraction": 0.125,
          "standard_len2_state_mean_correct_prefix": 0.25,
          "standard_len2_repair_executor_accuracy": 0.5,
          "standard_len2_repair_program_exact": 0.25,
          "standard_len2_repair_state_accuracy": 0.5,
          "standard_len2_repair_state_all_exact": 0.5,
          "standard_len2_repair_state_prefix_fraction": 0.5,
          "standard_len2_repair_state_mean_correct_prefix": 1.0,
          "standard_len2_repair_changed_fraction": 0.5,
          "standard_len2_repair_found_fraction": 0.5,
          "standard_len2_repair_avg_candidates": 23.0,
          "standard_len2_repair_avg_verified_candidates": 0.5,
          "standard_len3_direct_accuracy": NaN,
          "standard_len3_direct_target_mass": NaN,
          "standard_len3_executor_accuracy": 0.0,
          "standard_len3_executor_target_mass": 0.04242053651250899,
          "standard_len3_init_accuracy": 0.25,
          "standard_len3_init_pos_accuracy": 0.0,
          "standard_len3_op_accuracy": 0.6666666666666666,
          "standard_len3_arg_accuracy": 0.0,
          "standard_len3_op_pos_accuracy": 0.0,
          "standard_len3_arg_pos_accuracy": 0.0,
          "standard_len3_program_exact": 0.0,
          "standard_len3_state_accuracy": 0.16666666666666666,
          "standard_len3_state_all_exact": 0.0,
          "standard_len3_state_prefix_fraction": 0.16666666666666666,
          "standard_len3_state_mean_correct_prefix": 0.5,
          "standard_len3_repair_executor_accuracy": 0.0,
          "standard_len3_repair_program_exact": 0.0,
          "standard_len3_repair_state_accuracy": 0.16666666666666666,
          "standard_len3_repair_state_all_exact": 0.0,
          "standard_len3_repair_state_prefix_fraction": 0.16666666666666666,
          "standard_len3_repair_state_mean_correct_prefix": 0.5,
          "standard_len3_repair_changed_fraction": 0.0,
          "standard_len3_repair_found_fraction": 0.0,
          "standard_len3_repair_avg_candidates": 39.0,
          "standard_len3_repair_avg_verified_candidates": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.0,
          "paraphrase_len2_executor_target_mass": 0.0008053266501519829,
          "paraphrase_len2_init_accuracy": 0.5,
          "paraphrase_len2_init_pos_accuracy": 0.0,
          "paraphrase_len2_op_accuracy": 0.625,
          "paraphrase_len2_arg_accuracy": 0.0,
          "paraphrase_len2_op_pos_accuracy": 0.0,
          "paraphrase_len2_arg_pos_accuracy": 0.0,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len2_state_accuracy": 0.25,
          "paraphrase_len2_state_all_exact": 0.0,
          "paraphrase_len2_state_prefix_fraction": 0.25,
          "paraphrase_len2_state_mean_correct_prefix": 0.5,
          "paraphrase_len2_repair_executor_accuracy": 0.75,
          "paraphrase_len2_repair_program_exact": 0.25,
          "paraphrase_len2_repair_state_accuracy": 0.75,
          "paraphrase_len2_repair_state_all_exact": 0.75,
          "paraphrase_len2_repair_state_prefix_fraction": 0.75,
          "paraphrase_len2_repair_state_mean_correct_prefix": 1.5,
          "paraphrase_len2_repair_changed_fraction": 0.75,
          "paraphrase_len2_repair_found_fraction": 0.75,
          "paraphrase_len2_repair_avg_candidates": 23.0,
          "paraphrase_len2_repair_avg_verified_candidates": 0.75,
          "paraphrase_len3_direct_accuracy": NaN,
          "paraphrase_len3_direct_target_mass": NaN,
          "paraphrase_len3_executor_accuracy": 0.0,
          "paraphrase_len3_executor_target_mass": 0.002731189364567399,
          "paraphrase_len3_init_accuracy": 0.25,
          "paraphrase_len3_init_pos_accuracy": 0.0,
          "paraphrase_len3_op_accuracy": 0.5,
          "paraphrase_len3_arg_accuracy": 0.0,
          "paraphrase_len3_op_pos_accuracy": 0.0,
          "paraphrase_len3_arg_pos_accuracy": 0.0,
          "paraphrase_len3_program_exact": 0.0,
          "paraphrase_len3_state_accuracy": 0.0,
          "paraphrase_len3_state_all_exact": 0.0,
          "paraphrase_len3_state_prefix_fraction": 0.0,
          "paraphrase_len3_state_mean_correct_prefix": 0.0,
          "paraphrase_len3_repair_executor_accuracy": 0.0,
          "paraphrase_len3_repair_program_exact": 0.0,
          "paraphrase_len3_repair_state_accuracy": 0.0,
          "paraphrase_len3_repair_state_all_exact": 0.0,
          "paraphrase_len3_repair_state_prefix_fraction": 0.0,
          "paraphrase_len3_repair_state_mean_correct_prefix": 0.0,
          "paraphrase_len3_repair_changed_fraction": 0.0,
          "paraphrase_len3_repair_found_fraction": 0.0,
          "paraphrase_len3_repair_avg_candidates": 39.0,
          "paraphrase_len3_repair_avg_verified_candidates": 0.0,
          "paired_len2_direct_accuracy": NaN,
          "paired_len2_direct_target_mass": NaN,
          "paired_len2_executor_accuracy": 0.125,
          "paired_len2_executor_target_mass": 0.12405277679135907,
          "paired_len2_init_accuracy": 0.5,
          "paired_len2_init_pos_accuracy": 0.0,
          "paired_len2_op_accuracy": 0.5,
          "paired_len2_arg_accuracy": 0.0,
          "paired_len2_op_pos_accuracy": 0.0,
          "paired_len2_arg_pos_accuracy": 0.0,
          "paired_len2_program_exact": 0.0,
          "paired_len2_state_accuracy": 0.1875,
          "paired_len2_state_all_exact": 0.0,
          "paired_len2_state_prefix_fraction": 0.125,
          "paired_len2_state_mean_correct_prefix": 0.25,
          "paired_len2_repair_executor_accuracy": 0.625,
          "paired_len2_repair_program_exact": 0.0,
          "paired_len2_repair_state_accuracy": 0.5625,
          "paired_len2_repair_state_all_exact": 0.5,
          "paired_len2_repair_state_prefix_fraction": 0.5,
          "paired_len2_repair_state_mean_correct_prefix": 1.0,
          "paired_len2_repair_changed_fraction": 0.5,
          "paired_len2_repair_found_fraction": 0.5,
          "paired_len2_repair_avg_candidates": 23.0,
          "paired_len2_repair_avg_verified_candidates": 0.5,
          "paired_len2_pair_true_answer_consistency": 1.0,
          "paired_len2_pair_true_length_consistency": 1.0,
          "paired_len2_executor_pair_answer_consistency": 0.5,
          "paired_len2_executor_pair_both_correct": 0.0,
          "paired_len2_compiler_pair_program_consistency": 0.5,
          "paired_len2_compiler_pair_state_consistency": 0.5,
          "paired_len2_repair_pair_answer_consistency": 0.75,
          "paired_len2_repair_pair_both_correct": 0.5,
          "paired_len2_repair_pair_program_consistency": 0.5,
          "paired_len2_repair_pair_state_consistency": 0.5,
          "paired_len3_direct_accuracy": NaN,
          "paired_len3_direct_target_mass": NaN,
          "paired_len3_executor_accuracy": 0.0,
          "paired_len3_executor_target_mass": 0.00035051309623668203,
          "paired_len3_init_accuracy": 0.375,
          "paired_len3_init_pos_accuracy": 0.0,
          "paired_len3_op_accuracy": 0.5,
          "paired_len3_arg_accuracy": 0.0,
          "paired_len3_op_pos_accuracy": 0.0,
          "paired_len3_arg_pos_accuracy": 0.0,
          "paired_len3_program_exact": 0.0,
          "paired_len3_state_accuracy": 0.041666666666666664,
          "paired_len3_state_all_exact": 0.0,
          "paired_len3_state_prefix_fraction": 0.041666666666666664,
          "paired_len3_state_mean_correct_prefix": 0.125,
          "paired_len3_repair_executor_accuracy": 0.0,
          "paired_len3_repair_program_exact": 0.0,
          "paired_len3_repair_state_accuracy": 0.041666666666666664,
          "paired_len3_repair_state_all_exact": 0.0,
          "paired_len3_repair_state_prefix_fraction": 0.041666666666666664,
          "paired_len3_repair_state_mean_correct_prefix": 0.125,
          "paired_len3_repair_changed_fraction": 0.0,
          "paired_len3_repair_found_fraction": 0.0,
          "paired_len3_repair_avg_candidates": 39.0,
          "paired_len3_repair_avg_verified_candidates": 0.0,
          "paired_len3_pair_true_answer_consistency": 1.0,
          "paired_len3_pair_true_length_consistency": 1.0,
          "paired_len3_executor_pair_answer_consistency": 0.5,
          "paired_len3_executor_pair_both_correct": 0.0,
          "paired_len3_compiler_pair_program_consistency": 0.5,
          "paired_len3_compiler_pair_state_consistency": 0.5,
          "paired_len3_repair_pair_answer_consistency": 0.5,
          "paired_len3_repair_pair_both_correct": 0.0,
          "paired_len3_repair_pair_program_consistency": 0.5,
          "paired_len3_repair_pair_state_consistency": 0.5
        }
      ],
      "final_metrics": {
        "standard_len2": {
          "variant": "copy_trace_state_repair",
          "n": 4.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.009355536936709541,
          "init_accuracy": 0.5,
          "init_pos_accuracy": 0.0,
          "op_accuracy": 0.625,
          "arg_accuracy": 0.0,
          "op_pos_accuracy": 0.0,
          "arg_pos_accuracy": 0.0,
          "program_exact": 0.0,
          "state_accuracy": 0.125,
          "state_all_exact": 0.0,
          "state_prefix_fraction": 0.125,
          "state_mean_correct_prefix": 0.25,
          "repair_executor_accuracy": 0.5,
          "repair_program_exact": 0.25,
          "repair_state_accuracy": 0.5,
          "repair_state_all_exact": 0.5,
          "repair_state_prefix_fraction": 0.5,
          "repair_state_mean_correct_prefix": 1.0,
          "repair_changed_fraction": 0.5,
          "repair_found_fraction": 0.5,
          "repair_avg_candidates": 23.0,
          "repair_avg_verified_candidates": 0.5
        },
        "standard_len3": {
          "variant": "copy_trace_state_repair",
          "n": 4.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.04242053651250899,
          "init_accuracy": 0.25,
          "init_pos_accuracy": 0.0,
          "op_accuracy": 0.6666666666666666,
          "arg_accuracy": 0.0,
          "op_pos_accuracy": 0.0,
          "arg_pos_accuracy": 0.0,
          "program_exact": 0.0,
          "state_accuracy": 0.16666666666666666,
          "state_all_exact": 0.0,
          "state_prefix_fraction": 0.16666666666666666,
          "state_mean_correct_prefix": 0.5,
          "repair_executor_accuracy": 0.0,
          "repair_program_exact": 0.0,
          "repair_state_accuracy": 0.16666666666666666,
          "repair_state_all_exact": 0.0,
          "repair_state_prefix_fraction": 0.16666666666666666,
          "repair_state_mean_correct_prefix": 0.5,
          "repair_changed_fraction": 0.0,
          "repair_found_fraction": 0.0,
          "repair_avg_candidates": 39.0,
          "repair_avg_verified_candidates": 0.0
        },
        "paraphrase_len2": {
          "variant": "copy_trace_state_repair",
          "n": 4.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.0008053266501519829,
          "init_accuracy": 0.5,
          "init_pos_accuracy": 0.0,
          "op_accuracy": 0.625,
          "arg_accuracy": 0.0,
          "op_pos_accuracy": 0.0,
          "arg_pos_accuracy": 0.0,
          "program_exact": 0.0,
          "state_accuracy": 0.25,
          "state_all_exact": 0.0,
          "state_prefix_fraction": 0.25,
          "state_mean_correct_prefix": 0.5,
          "repair_executor_accuracy": 0.75,
          "repair_program_exact": 0.25,
          "repair_state_accuracy": 0.75,
          "repair_state_all_exact": 0.75,
          "repair_state_prefix_fraction": 0.75,
          "repair_state_mean_correct_prefix": 1.5,
          "repair_changed_fraction": 0.75,
          "repair_found_fraction": 0.75,
          "repair_avg_candidates": 23.0,
          "repair_avg_verified_candidates": 0.75
        },
        "paraphrase_len3": {
          "variant": "copy_trace_state_repair",
          "n": 4.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.002731189364567399,
          "init_accuracy": 0.25,
          "init_pos_accuracy": 0.0,
          "op_accuracy": 0.5,
          "arg_accuracy": 0.0,
          "op_pos_accuracy": 0.0,
          "arg_pos_accuracy": 0.0,
          "program_exact": 0.0,
          "state_accuracy": 0.0,
          "state_all_exact": 0.0,
          "state_prefix_fraction": 0.0,
          "state_mean_correct_prefix": 0.0,
          "repair_executor_accuracy": 0.0,
          "repair_program_exact": 0.0,
          "repair_state_accuracy": 0.0,
          "repair_state_all_exact": 0.0,
          "repair_state_prefix_fraction": 0.0,
          "repair_state_mean_correct_prefix": 0.0,
          "repair_changed_fraction": 0.0,
          "repair_found_fraction": 0.0,
          "repair_avg_candidates": 39.0,
          "repair_avg_verified_candidates": 0.0
        },
        "paired_len2": {
          "variant": "copy_trace_state_repair",
          "n": 8.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.125,
          "executor_target_mass": 0.12405277679135907,
          "init_accuracy": 0.5,
          "init_pos_accuracy": 0.0,
          "op_accuracy": 0.5,
          "arg_accuracy": 0.0,
          "op_pos_accuracy": 0.0,
          "arg_pos_accuracy": 0.0,
          "program_exact": 0.0,
          "state_accuracy": 0.1875,
          "state_all_exact": 0.0,
          "state_prefix_fraction": 0.125,
          "state_mean_correct_prefix": 0.25,
          "repair_executor_accuracy": 0.625,
          "repair_program_exact": 0.0,
          "repair_state_accuracy": 0.5625,
          "repair_state_all_exact": 0.5,
          "repair_state_prefix_fraction": 0.5,
          "repair_state_mean_correct_prefix": 1.0,
          "repair_changed_fraction": 0.5,
          "repair_found_fraction": 0.5,
          "repair_avg_candidates": 23.0,
          "repair_avg_verified_candidates": 0.5,
          "pair_true_answer_consistency": 1.0,
          "pair_true_length_consistency": 1.0,
          "executor_pair_answer_consistency": 0.5,
          "executor_pair_both_correct": 0.0,
          "compiler_pair_program_consistency": 0.5,
          "compiler_pair_state_consistency": 0.5,
          "repair_pair_answer_consistency": 0.75,
          "repair_pair_both_correct": 0.5,
          "repair_pair_program_consistency": 0.5,
          "repair_pair_state_consistency": 0.5
        },
        "paired_len3": {
          "variant": "copy_trace_state_repair",
          "n": 8.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.00035051309623668203,
          "init_accuracy": 0.375,
          "init_pos_accuracy": 0.0,
          "op_accuracy": 0.5,
          "arg_accuracy": 0.0,
          "op_pos_accuracy": 0.0,
          "arg_pos_accuracy": 0.0,
          "program_exact": 0.0,
          "state_accuracy": 0.041666666666666664,
          "state_all_exact": 0.0,
          "state_prefix_fraction": 0.041666666666666664,
          "state_mean_correct_prefix": 0.125,
          "repair_executor_accuracy": 0.0,
          "repair_program_exact": 0.0,
          "repair_state_accuracy": 0.041666666666666664,
          "repair_state_all_exact": 0.0,
          "repair_state_prefix_fraction": 0.041666666666666664,
          "repair_state_mean_correct_prefix": 0.125,
          "repair_changed_fraction": 0.0,
          "repair_found_fraction": 0.0,
          "repair_avg_candidates": 39.0,
          "repair_avg_verified_candidates": 0.0,
          "pair_true_answer_consistency": 1.0,
          "pair_true_length_consistency": 1.0,
          "executor_pair_answer_consistency": 0.5,
          "executor_pair_both_correct": 0.0,
          "compiler_pair_program_consistency": 0.5,
          "compiler_pair_state_consistency": 0.5,
          "repair_pair_answer_consistency": 0.5,
          "repair_pair_both_correct": 0.0,
          "repair_pair_program_consistency": 0.5,
          "repair_pair_state_consistency": 0.5
        }
      }
    }
  }
}