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": 0.8862499818205833
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.867499977350235
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.867499977350235
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.867499977350235,
 33      "std": 0.029047373044579175,
 34      "per_seed": [
 35        0.9099999666213989,
 36        0.8700000047683716,
 37        0.8899999856948853,
 38        0.8399999737739563,
 39        0.85999995470047,
 40        0.8999999761581421,
 41        0.8499999642372131,
 42        0.8199999928474426
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.8687499910593033,
 49    "std": 0.03407253705893027,
 50    "per_seed": [
 51      0.8899999856948853,
 52      0.8299999833106995,
 53      0.8600000143051147,
 54      0.8999999761581421,
 55      0.8199999928474426,
 56      0.8399999737739563,
 57      0.9200000166893005,
 58      0.8899999856948853
 59    ],
 60    "n": 8,
 61    "config": {
 62      "gamma": 0.05,
 63      "lr": 0.001,
 64      "baseline_best_lr": 0.003
 65    }
 66  },
 67  "comparison": {
 68    "delta_mean": 0.0012500137090682983,
 69    "idea_wins": 5,
 70    "n_pairs": 8,
 71    "per_seed_diffs": [
 72      -0.019999980926513672,
 73      -0.04000002145767212,
 74      -0.029999971389770508,
 75      0.06000000238418579,
 76      -0.039999961853027344,
 77      -0.06000000238418579,
 78      0.0700000524520874,
 79      0.06999999284744263
 80    ],
 81    "p_value": 0.9195,
 82    "mde": 0.04632656557948158,
 83    "mde_rel_pct": 5.340238246574409,
 84    "verdict": "no measurable effect",
 85    "system_worked": false
 86  },
 87  "mechanism_signature": {
 88    "prediction": "parent-operation retention decreases monotonically with gamma according to alpha_bar",
 89    "trained_model_measurements": [
 90      {
 91        "gamma": 0.05,
 92        "predicted_retention": 0.99,
 93        "observed_retention": 1.0,
 94        "absolute_error": 0.010000000000000009
 95      },
 96      {
 97        "gamma": 0.15,
 98        "predicted_retention": 0.9258820055401662,
 99        "observed_retention": 1.0,
100        "absolute_error": 0.0741179944598338
101      },
102      {
103        "gamma": 0.3,
104        "predicted_retention": 0.7420133348094462,
105        "observed_retention": 0.6666666666666666,
106        "absolute_error": 0.07534666814277957
107      },
108      {
109        "gamma": 1.0,
110        "predicted_retention": 0.030871261192240654,
111        "observed_retention": 0.0,
112        "absolute_error": 0.030871261192240654
113      }
114    ],
115    "confirmed": true
116  },
117  "protocol_notes": {
118    "structural_match": "vision CNN architecture mutation",
119    "paired_seeds": [
120      0,
121      1,
122      2,
123      3,
124      4,
125      5,
126      6,
127      7
128    ],
129    "baseline_grid": [
130      0.001,
131      0.003,
132      0.006
133    ],
134    "idea_grid": [
135      0.001,
136      0.003,
137      0.006
138    ],
139    "epochs": 2,
140    "metric": "classification error, lower is better"
141  }
142}