{
  "bench_version": 1,
  "track": "tabular",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "weight_decay": 0.0001
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015,
          "weight_decay": 0.0001
        },
        "mean": 16.405593872070312
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0001
        },
        "mean": 13.136067628860474
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0001
        },
        "mean": 9.729963302612305
      }
    ],
    "full": {
      "mean": 9.7748162150383,
      "std": 1.146524142314583,
      "per_seed": [
        10.155755043029785,
        11.133987426757812,
        8.668990135192871,
        8.96112060546875,
        7.614502429962158,
        10.594099044799805,
        10.266080856323242,
        10.803994178771973
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 8.963683664798737,
    "std": 1.2184976932247173,
    "per_seed": [
      10.102669715881348,
      11.417046546936035,
      9.143009185791016,
      8.528031349182129,
      7.16295862197876,
      8.287325859069824,
      8.166311264038086,
      8.902116775512695
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": -0.811132550239563,
    "idea_wins": 6,
    "n_pairs": 8,
    "per_seed_diffs": [
      -0.0530853271484375,
      0.28305912017822266,
      0.47401905059814453,
      -0.4330892562866211,
      -0.45154380798339844,
      -2.3067731857299805,
      -2.0997695922851562,
      -1.9018774032592773
    ],
    "p_value": 0.10015,
    "mde": 0.9366083105537301,
    "mde_rel_pct": 9.581850849663882,
    "verdict": "no significant win",
    "system_worked": false
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.0015,
        "weight_decay": 0.0001
      },
      "mean": 17.79743456840515
    },
    {
      "cfg": {
        "lr": 0.003,
        "weight_decay": 0.0001
      },
      "mean": 13.299631118774414
    },
    {
      "cfg": {
        "lr": 0.006,
        "weight_decay": 0.0001
      },
      "mean": 9.797689199447632
    }
  ],
  "selection_note": "Both methods used the same lr/weight-decay union; idea selected on seeds 0-3.",
  "mechanism_signature": {
    "type": "trained_model_jacobian_low_rank",
    "rank": 16,
    "samples": [
      {
        "seed": 0,
        "predicted_captured_fraction": 0.17476074945303136,
        "observed_total_energy": 26951.634765625,
        "projected_energy": 4710.087890625
      },
      {
        "seed": 1,
        "predicted_captured_fraction": 0.2573280523615646,
        "observed_total_energy": 32180.689453125,
        "projected_energy": 8280.994140625
      },
      {
        "seed": 2,
        "predicted_captured_fraction": 0.24594845290024767,
        "observed_total_energy": 26192.974609375,
        "projected_energy": 6442.12158203125
      },
      {
        "seed": 3,
        "predicted_captured_fraction": 0.17532369020915245,
        "observed_total_energy": 24280.84765625,
        "projected_energy": 4257.0078125
      }
    ],
    "predicted_vs_observed_mean_ratio": 0.21334023623099901,
    "confirmed": false
  },
  "runtime_sec": 21.001119817025028
}