{
  "bench_version": 1,
  "track": "vision",
  "model": "cnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.003
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.8612499833106995
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.8112499713897705
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.8918749839067459
      }
    ],
    "full": {
      "mean": 0.828437477350235,
      "std": 0.03745701284007905,
      "per_seed": [
        0.79749995470047,
        0.8574999570846558,
        0.824999988079071,
        0.7649999856948853,
        0.8700000047683716,
        0.8574999570846558,
        0.8650000095367432,
        0.7899999618530273
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.843437485396862,
    "std": 0.03642066574738751,
    "per_seed": [
      0.7824999690055847,
      0.9024999737739563,
      0.8700000047683716,
      0.8075000047683716,
      0.8574999570846558,
      0.8349999785423279,
      0.8700000047683716,
      0.8224999904632568
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.015000008046627045,
    "idea_wins": 3,
    "n_pairs": 8,
    "per_seed_diffs": [
      -0.014999985694885254,
      0.04500001668930054,
      0.04500001668930054,
      0.04250001907348633,
      -0.01250004768371582,
      -0.02249997854232788,
      0.004999995231628418,
      0.03250002861022949
    ],
    "p_value": 0.19565,
    "mde": 0.024531031621491128,
    "mde_rel_pct": 2.9611204577506394,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "prediction": "positive normalized filter is non-expansive in sup norm",
    "baseline_signature": {
      "test_metric": 0.79749995470047,
      "layer_ratios": [
        1.1681838035583496,
        1.1140750646591187,
        0.872684895992279
      ],
      "max_ratio": 1.1681838035583496,
      "prediction": "Markov filter ratio <= 1 relative to pre-filter activation"
    },
    "idea_signature": {
      "test_metric": 0.7824999690055847,
      "layer_ratios": [
        0.7950915694236755,
        0.97414231300354,
        0.9690303802490234
      ],
      "max_ratio": 0.97414231300354,
      "prediction": "Markov filter ratio <= 1 relative to pre-filter activation"
    },
    "predicted_max_ratio": 1.0,
    "observed_max_ratio": 0.97414231300354,
    "confirmed": true
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001
      },
      "mean": 0.8543749898672104
    },
    {
      "cfg": {
        "lr": 0.003
      },
      "mean": 0.840624988079071
    },
    {
      "cfg": {
        "lr": 0.01
      },
      "mean": 0.8699999898672104
    }
  ],
  "protocol": {
    "paired_seeds": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      7
    ],
    "epochs": 5,
    "n_train": 400,
    "n_test": 400,
    "filter_gamma": 0.7,
    "same_lr_grid": true
  }
}