1{
  2  "bench_version": 1,
  3  "track": "tabular",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 25.60164165496826
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 13.860755920410156
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 8.352817177772522
 29      }
 30    ],
 31    "full": {
 32      "mean": 8.567867696285248,
 33      "std": 0.9045233203943077,
 34      "per_seed": [
 35        7.95448637008667,
 36        9.628232955932617,
 37        7.726639747619629,
 38        8.101909637451172,
 39        7.126259803771973,
 40        8.949170112609863,
 41        9.36670970916748,
 42        9.689533233642578
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 92.21004962921143,
 49    "std": 10.383480498694608,
 50    "per_seed": [
 51      99.29408264160156,
 52      93.70222473144531,
 53      98.78123474121094,
 54      72.97736358642578,
 55      81.31901550292969,
 56      94.73509216308594,
 57      88.57316589355469,
 58      108.2982177734375
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 83.64218193292618,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      91.33959627151489,
 68      84.0739917755127,
 69      91.05459499359131,
 70      64.87545394897461,
 71      74.19275569915771,
 72      85.78592205047607,
 73      79.2064561843872,
 74      98.60868453979492
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 8.961902038617074,
 78    "mde_rel_pct": 104.598977905584,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "idea_sweep": [
 83    {
 84      "cfg": {
 85        "lr": 0.001,
 86        "r0": 0.02
 87      },
 88      "mean": 91.18872833251953
 89    },
 90    {
 91      "cfg": {
 92        "lr": 0.003,
 93        "r0": 0.02
 94      },
 95      "mean": 91.1887264251709
 96    },
 97    {
 98      "cfg": {
 99        "lr": 0.01,
100        "r0": 0.02
101      },
102      "mean": 91.18872833251953
103    }
104  ],
105  "mechanism_signature": {
106    "mean_q": 1.0,
107    "clip_fraction": 1.0,
108    "radius_final": 0.08441391633993081,
109    "radius_initial": 0.02,
110    "mean_scale": 0.15767846046228018,
111    "mean_proposal_norm": 0.2890951458546632,
112    "prediction": "sustained saturation expands radius and later unsaturation contracts it",
113    "observed_radius_increase": true,
114    "observed_clipping_fraction": 1.0,
115    "confirmed": false,
116    "observed_unsaturation": false,
117    "observed_radius_contraction": false,
118    "confirmation_reason": "Radius expanded, but q remained 1.0 and no post-saturation contraction was observed; the full bidirectional prediction was not confirmed."
119  }
120}