{
  "bench_version": 1,
  "track": "tabular",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "epochs": 8
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "epochs": 8
        },
        "mean": 217.77801513671875
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 8
        },
        "mean": 149.99735641479492
      },
      {
        "cfg": {
          "lr": 0.01,
          "epochs": 8
        },
        "mean": 26.118374347686768
      }
    ],
    "full": {
      "mean": 35.374396085739136,
      "std": 12.759078946030915,
      "per_seed": [
        29.38343620300293,
        32.629642486572266,
        22.46150779724121,
        19.998910903930664,
        34.86793899536133,
        63.3486328125,
        36.457733154296875,
        43.84736633300781
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 13.662639737129211,
    "std": 1.2255122413284667,
    "per_seed": [
      13.658586502075195,
      10.993560791015625,
      14.850048065185547,
      14.97375202178955,
      12.930068016052246,
      13.51315689086914,
      13.623165130615234,
      14.758780479431152
    ],
    "n": 8,
    "sweep": [
      {
        "mean": 16.703367471694946,
        "std": 1.902843853945096,
        "per_seed": [
          16.672517776489258,
          12.807530403137207,
          17.32662010192871,
          16.859230041503906,
          15.626265525817871,
          18.918928146362305,
          16.08186149597168,
          19.333986282348633
        ],
        "n": 8,
        "cfg": {
          "lr": 0.01
        }
      },
      {
        "mean": 31.656866788864136,
        "std": 7.626313276119716,
        "per_seed": [
          25.898263931274414,
          24.708200454711914,
          19.921588897705078,
          45.958133697509766,
          34.698883056640625,
          37.070926666259766,
          33.396942138671875,
          31.60199546813965
        ],
        "n": 8,
        "cfg": {
          "lr": 0.005
        }
      },
      {
        "mean": 13.662639737129211,
        "std": 1.2255122413284667,
        "per_seed": [
          13.658586502075195,
          10.993560791015625,
          14.850048065185547,
          14.97375202178955,
          12.930068016052246,
          13.51315689086914,
          13.623165130615234,
          14.758780479431152
        ],
        "n": 8,
        "cfg": {
          "lr": 0.02
        }
      }
    ],
    "best_cfg": {
      "lr": 0.02
    }
  },
  "comparison": {
    "delta_mean": -21.711756348609924,
    "idea_wins": 8,
    "n_pairs": 8,
    "per_seed_diffs": [
      -15.724849700927734,
      -21.63608169555664,
      -7.611459732055664,
      -5.025158882141113,
      -21.937870979309082,
      -49.83547592163086,
      -22.83456802368164,
      -29.08858585357666
    ],
    "p_value": 0.0081,
    "mde": 11.653916222940111,
    "mde_rel_pct": 32.94449520691119,
    "verdict": "idea better (significant)",
    "system_worked": true
  },
  "mechanism_signature": {
    "prediction": "spectral floor and rotation safeguard bound inverse-Hessian spectrum",
    "predicted_floor": 0.0001,
    "observed_min_eig_min": 0.00015548363137000883,
    "observed_max_rotation": 0.9999999997777941,
    "safeguards_total": 374,
    "confirmed": true
  },
  "math_check": {
    "secant_error": 3.1249111476034713e-15,
    "floor_min_eig": 0.001,
    "secant_pass": true,
    "floor_pass": true
  }
}