{
  "bench_version": 1,
  "track": "tabular",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "weight_decay": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0
        },
        "mean": 152.21145248413086
      },
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0001
        },
        "mean": 152.18722534179688
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 101.97886276245117
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0001
        },
        "mean": 101.97932815551758
      },
      {
        "cfg": {
          "lr": 0.01,
          "weight_decay": 0.0
        },
        "mean": 31.591657638549805
      },
      {
        "cfg": {
          "lr": 0.01,
          "weight_decay": 0.0001
        },
        "mean": 31.591760635375977
      }
    ],
    "full": {
      "mean": 33.03840112686157,
      "std": 2.3804098981535065,
      "per_seed": [
        34.395347595214844,
        31.198909759521484,
        32.69178009033203,
        28.08059310913086,
        36.69027328491211,
        33.26426696777344,
        34.30606460571289,
        33.67997360229492
      ],
      "n": 8
    }
  },
  "idea": {
    "per_seed": [
      10.268460273742676,
      10.212242126464844,
      9.813057899475098,
      10.562983512878418,
      13.34918212890625,
      11.693805694580078,
      10.852352142333984,
      13.36784839630127
    ],
    "mean": 11.264991521835327,
    "cfg": {
      "lr": 0.001,
      "weight_decay": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.01,
          "weight_decay": 0.0
        },
        "mean": 24.510720252990723
      },
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0
        },
        "mean": 10.214185953140259
      },
      {
        "cfg": {
          "lr": 0.01,
          "weight_decay": 0.0
        },
        "mean": 24.510720252990723
      }
    ]
  },
  "comparison": {
    "delta_mean": -21.773409605026245,
    "idea_wins": 8,
    "n_pairs": 8,
    "per_seed_diffs": [
      -24.126887321472168,
      -20.98666763305664,
      -22.878722190856934,
      -17.51760959625244,
      -23.34109115600586,
      -21.57046127319336,
      -23.453712463378906,
      -20.312125205993652
    ],
    "p_value": 0.0081,
    "mde": 1.8140862942829863,
    "mde_rel_pct": 5.4908416642718825,
    "verdict": "idea better (significant)",
    "system_worked": true
  },
  "mechanism_signature": {
    "quantity": "trained-NN posterior uncertainty; upper-vs-lower quartile test residual and interval coverage",
    "mean_high_to_low_sd_ratio": 3.6821917593479156,
    "mean_sd": 0.013744487892836332,
    "coverage_beta_2.5": 0.0084375,
    "predicted_high_uncertainty_ratio_gt_1": true,
    "confirmed": true,
    "track_match": "tabular regression with shared MLP representation and Bayesian Fourier readout"
  }
}