1{
  2  "bench_version": 1,
  3  "track": "vision",
  4  "model": "cnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.003
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.8612499833106995
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.8112499713897705
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.8918749839067459
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.828437477350235,
 33      "std": 0.03745701284007905,
 34      "per_seed": [
 35        0.79749995470047,
 36        0.8574999570846558,
 37        0.824999988079071,
 38        0.7649999856948853,
 39        0.8700000047683716,
 40        0.8574999570846558,
 41        0.8650000095367432,
 42        0.7899999618530273
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.843437485396862,
 49    "std": 0.03642066574738751,
 50    "per_seed": [
 51      0.7824999690055847,
 52      0.9024999737739563,
 53      0.8700000047683716,
 54      0.8075000047683716,
 55      0.8574999570846558,
 56      0.8349999785423279,
 57      0.8700000047683716,
 58      0.8224999904632568
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.015000008046627045,
 64    "idea_wins": 3,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -0.014999985694885254,
 68      0.04500001668930054,
 69      0.04500001668930054,
 70      0.04250001907348633,
 71      -0.01250004768371582,
 72      -0.02249997854232788,
 73      0.004999995231628418,
 74      0.03250002861022949
 75    ],
 76    "p_value": 0.19565,
 77    "mde": 0.024531031621491128,
 78    "mde_rel_pct": 2.9611204577506394,
 79    "verdict": "no significant win",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "prediction": "positive normalized filter is non-expansive in sup norm",
 84    "baseline_signature": {
 85      "test_metric": 0.79749995470047,
 86      "layer_ratios": [
 87        1.1681838035583496,
 88        1.1140750646591187,
 89        0.872684895992279
 90      ],
 91      "max_ratio": 1.1681838035583496,
 92      "prediction": "Markov filter ratio <= 1 relative to pre-filter activation"
 93    },
 94    "idea_signature": {
 95      "test_metric": 0.7824999690055847,
 96      "layer_ratios": [
 97        0.7950915694236755,
 98        0.97414231300354,
 99        0.9690303802490234
100      ],
101      "max_ratio": 0.97414231300354,
102      "prediction": "Markov filter ratio <= 1 relative to pre-filter activation"
103    },
104    "predicted_max_ratio": 1.0,
105    "observed_max_ratio": 0.97414231300354,
106    "confirmed": true
107  },
108  "idea_sweep": [
109    {
110      "cfg": {
111        "lr": 0.001
112      },
113      "mean": 0.8543749898672104
114    },
115    {
116      "cfg": {
117        "lr": 0.003
118      },
119      "mean": 0.840624988079071
120    },
121    {
122      "cfg": {
123        "lr": 0.01
124      },
125      "mean": 0.8699999898672104
126    }
127  ],
128  "protocol": {
129    "paired_seeds": [
130      0,
131      1,
132      2,
133      3,
134      4,
135      5,
136      6,
137      7
138    ],
139    "epochs": 5,
140    "n_train": 400,
141    "n_test": 400,
142    "filter_gamma": 0.7,
143    "same_lr_grid": true
144  }
145}