1{
  2  "bench_version": 1,
  3  "track": "custom:heavy_tailed_binary_radial",
  4  "model": "mlp_tiny",
  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": 0.003
 15        },
 16        "mean": 0.1743749901652336
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.01
 21        },
 22        "mean": 0.17874999344348907
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.03
 27        },
 28        "mean": 0.18687499687075615
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.17874999344348907,
 33      "std": 0.014469796810738268,
 34      "per_seed": [
 35        0.16749998927116394,
 36        0.17999999225139618,
 37        0.1824999898672104,
 38        0.16749998927116394,
 39        0.17249999940395355,
 40        0.19249999523162842,
 41        0.1599999964237213,
 42        0.20749999582767487
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.1849999949336052,
 49    "std": 0.015000001837810447,
 50    "per_seed": [
 51      0.16749998927116394,
 52      0.19499999284744263,
 53      0.1850000023841858,
 54      0.17749999463558197,
 55      0.17499999701976776,
 56      0.20250000059604645,
 57      0.16749998927116394,
 58      0.20999999344348907
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.006250001490116119,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.0,
 68      0.015000000596046448,
 69      0.002500012516975403,
 70      0.01000000536441803,
 71      0.002499997615814209,
 72      0.01000000536441803,
 73      0.007499992847442627,
 74      0.002499997615814209
 75    ],
 76    "p_value": 0.01485,
 77    "mde": 0.004327507057943815,
 78    "mde_rel_pct": 2.4209830582801866,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "custom_track": {
 83    "name": "heavy_tailed_binary_radial",
 84    "file": "bench_experiment.py",
 85    "domain": "binary_classification"
 86  },
 87  "math_check": {
 88    "max_abs_radius_error": 2.6645352591003757e-15,
 89    "finite": true
 90  },
 91  "idea_sweep": [
 92    {
 93      "cfg": {
 94        "lr": 0.003
 95      },
 96      "mean": 0.18124999478459358
 97    },
 98    {
 99      "cfg": {
100        "lr": 0.01
101      },
102      "mean": 0.18312499672174454
103    },
104    {
105      "cfg": {
106        "lr": 0.03
107      },
108      "mean": 0.1899999938905239
109    }
110  ],
111  "selection": {
112    "best_idea_cfg": {
113      "lr": 0.003
114    },
115    "shared_lr_grid": [
116      0.003,
117      0.01,
118      0.03
119    ]
120  },
121  "mechanism_signature": {
122    "prediction": "fractional radial readout produces measurable nonlinear curvature beyond affine radius difference on trained heavy-tailed systems",
123    "affine_logit_rmse_mean": 0.12869403761578724,
124    "fractional_basis_logit_rmse_mean": 3.302959132867045e-07,
125    "tail_abs_logit_mean": 3.0879115611314774,
126    "confirmed": true
127  }
128}