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": 0.001
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.8866666704416275
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.8950000107288361
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.908333346247673
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.8875000029802322,
 33      "std": 0.02548147140610766,
 34      "per_seed": [
 35        0.9133333563804626,
 36        0.8999999761581421,
 37        0.8666666746139526,
 38        0.8666666746139526,
 39        0.8733333349227905,
 40        0.8933333158493042,
 41        0.9333333373069763,
 42        0.8533333539962769
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.890000008046627,
 49    "std": 0.027284505974877197,
 50    "per_seed": [
 51      0.9133333563804626,
 52      0.8999999761581421,
 53      0.8533333539962769,
 54      0.8600000143051147,
 55      0.9066666960716248,
 56      0.8933333158493042,
 57      0.9333333373069763,
 58      0.8600000143051147
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.002500005066394806,
 64    "idea_wins": 2,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.0,
 68      0.0,
 69      -0.013333320617675781,
 70      -0.006666660308837891,
 71      0.03333336114883423,
 72      0.0,
 73      0.0,
 74      0.006666660308837891
 75    ],
 76    "p_value": 0.8775,
 77    "mde": 0.011515956518988579,
 78    "mde_rel_pct": 1.2975725611625806,
 79    "verdict": "no measurable effect",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "track_choice": "vision: categorical class probability output matches exact DPD integral; label corruption tests robustness",
 84    "noise_rate": 0.2,
 85    "idea_sweep": [
 86      {
 87        "cfg": {
 88          "lr": 0.001,
 89          "alpha": 0.3
 90        },
 91        "result": {
 92          "mean": 0.8916666731238365,
 93          "std": 0.02920236995935498,
 94          "per_seed": [
 95            0.9133333563804626,
 96            0.9133333563804626,
 97            0.8666666746139526,
 98            0.8666666746139526,
 99            0.9066666960716248,
100            0.8933333158493042,
101            0.9333333373069763,
102            0.8399999737739563
103          ],
104          "n": 8
105        }
106      },
107      {
108        "cfg": {
109          "lr": 0.001,
110          "alpha": 0.5
111        },
112        "result": {
113          "mean": 0.890000008046627,
114          "std": 0.027284505974877197,
115          "per_seed": [
116            0.9133333563804626,
117            0.8999999761581421,
118            0.8533333539962769,
119            0.8600000143051147,
120            0.9066666960716248,
121            0.8933333158493042,
122            0.9333333373069763,
123            0.8600000143051147
124          ],
125          "n": 8
126        }
127      },
128      {
129        "cfg": {
130          "lr": 0.001,
131          "alpha": 0.7
132        },
133        "result": {
134          "mean": 0.8900000154972076,
135          "std": 0.029627313293764897,
136          "per_seed": [
137            0.9133333563804626,
138            0.9133333563804626,
139            0.846666693687439,
140            0.8533333539962769,
141            0.9066666960716248,
142            0.8866666555404663,
143            0.9333333373069763,
144            0.8666666746139526
145          ],
146          "n": 8
147        }
148      }
149    ],
150    "mechanism_signature": {
151      "prediction": "observed-label score gradient ratio equals (1+alpha)*q^alpha",
152      "alpha": 0.5,
153      "n_corrupted": 48,
154      "mean_predicted_ratio": 0.48781922459602356,
155      "mean_observed_ratio": 0.48781922459602356,
156      "max_relative_error": 3.0113764637462737e-07,
157      "corrupted_mean_q": 0.10800278186798096,
158      "confirmed": true
159    }
160  }
161}