{
  "bench_version": 1,
  "track": "vision",
  "model": "cnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.001
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.8887499868869781
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.8899999856948853
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.8912499845027924
      }
    ],
    "full": {
      "mean": 0.8762499839067459,
      "std": 0.024843258193087,
      "per_seed": [
        0.8949999809265137,
        0.8849999904632568,
        0.8999999761581421,
        0.875,
        0.8299999833106995,
        0.9049999713897705,
        0.8449999690055847,
        0.875
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.8762499839067459,
    "std": 0.024843258193087,
    "per_seed": [
      0.8949999809265137,
      0.8849999904632568,
      0.8999999761581421,
      0.875,
      0.8299999833106995,
      0.9049999713897705,
      0.8449999690055847,
      0.875
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.0,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.0,
      0.0,
      0.0,
      0.0,
      0.0,
      0.0,
      0.0,
      0.0
    ],
    "p_value": 1.0,
    "mde": 0.0,
    "mde_rel_pct": 0.0,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001
      },
      "result": {
        "mean": 0.8762499839067459,
        "std": 0.024843258193087,
        "per_seed": [
          0.8949999809265137,
          0.8849999904632568,
          0.8999999761581421,
          0.875,
          0.8299999833106995,
          0.9049999713897705,
          0.8449999690055847,
          0.875
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.003
      },
      "result": {
        "mean": 0.877499982714653,
        "std": 0.029895645776799453,
        "per_seed": [
          0.9049999713897705,
          0.8799999952316284,
          0.9049999713897705,
          0.8700000047683716,
          0.8499999642372131,
          0.9049999713897705,
          0.8899999856948853,
          0.8149999976158142
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.01
      },
      "result": {
        "mean": 0.877499982714653,
        "std": 0.02958039211486051,
        "per_seed": [
          0.8949999809265137,
          0.8799999952316284,
          0.9149999618530273,
          0.875,
          0.824999988079071,
          0.9149999618530273,
          0.8449999690055847,
          0.8700000047683716
        ],
        "n": 8
      }
    }
  ],
  "protocol_note": "Verifier pruning is inference-time only; training and standard test metric are unchanged systems.",
  "mechanism_signature": {
    "fixed_sign_fraction": 0.5423773306387443,
    "max_fixed_substitution_error": 0.0,
    "predicted_reduction_fraction": 0.5423773306387443,
    "fixed_sign_mean": 959721.5,
    "unstable_mean": 809750.5,
    "total_relu_instances_mean": 1769472.0,
    "confirmed": true
  },
  "experiment_settings": {
    "epochs": 3,
    "n_train": 400,
    "n_test": 200,
    "batch": 128,
    "lr_union": [
      0.001,
      0.003,
      0.01
    ],
    "seeds": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      7
    ]
  }
}