{
  "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": 24.645838737487793
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 13.62548279762268
      },
      {
        "cfg": {
          "lr": 0.01,
          "weight_decay": 0.0
        },
        "mean": 8.338012218475342
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0001
        },
        "mean": 13.600164413452148
      }
    ],
    "full": {
      "mean": 8.617289543151855,
      "std": 0.739608681399212,
      "per_seed": [
        8.399882316589355,
        7.785224437713623,
        7.834766864776611,
        9.332175254821777,
        7.886314392089844,
        8.663776397705078,
        9.938685417175293,
        9.097491264343262
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 8.601594865322113,
    "std": 0.7732046694513341,
    "per_seed": [
      8.518109321594238,
      7.764858722686768,
      7.801291465759277,
      9.256509780883789,
      7.880615234375,
      8.263826370239258,
      9.904521942138672,
      9.423026084899902
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": -0.01569467782974243,
    "idea_wins": 6,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.11822700500488281,
      -0.02036571502685547,
      -0.033475399017333984,
      -0.07566547393798828,
      -0.00569915771484375,
      -0.3999500274658203,
      -0.034163475036621094,
      0.3255348205566406
    ],
    "p_value": 0.75635,
    "mde": 0.1691312343107815,
    "mde_rel_pct": 1.9626964309814772,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001,
        "lambda_k": 0.0001
      },
      "mean": 23.893144130706787
    },
    {
      "cfg": {
        "lr": 0.003,
        "lambda_k": 0.0001
      },
      "mean": 13.534956455230713
    },
    {
      "cfg": {
        "lr": 0.01,
        "lambda_k": 0.0001
      },
      "mean": 8.335192322731018
    }
  ],
  "matched_structure": "MLP/tabular optimizer-regularizer track",
  "mechanism_signature": {
    "predicted_duplicate_fraction_mean": 0.0,
    "observed_duplicate_fraction_mean": 0.0,
    "probe_prediction_rms_mean": 15.10029935836792,
    "confirmed": false
  }
}