{
  "bench_version": 1,
  "track": "tabular",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.002,
      "weight_decay": 0.0001
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0
        },
        "mean": 140.17193984985352
      },
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0001
        },
        "mean": 140.17272758483887
      },
      {
        "cfg": {
          "lr": 0.002,
          "weight_decay": 0.0
        },
        "mean": 46.896180152893066
      },
      {
        "cfg": {
          "lr": 0.002,
          "weight_decay": 0.0001
        },
        "mean": 46.89336395263672
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 425.859375
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0001
        },
        "mean": 425.85955810546875
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0
        },
        "mean": 150.0003662109375
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0001
        },
        "mean": 150.00535202026367
      }
    ],
    "full": {
      "mean": 53.96195363998413,
      "std": 40.99246488816214,
      "per_seed": [
        30.952533721923828,
        22.93033218383789,
        83.2318344116211,
        50.45875549316406,
        149.6821746826172,
        21.348867416381836,
        47.70417022705078,
        25.386960983276367
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 45.54406785964966,
    "std": 8.338173492397852,
    "per_seed": [
      43.00838851928711,
      44.718257904052734,
      28.018310546875,
      49.26393508911133,
      56.6592903137207,
      41.586570739746094,
      54.953060150146484,
      46.14472961425781
    ],
    "n": 8,
    "selected_cfg": {
      "lr": 0.002,
      "weight_decay": 0.0001
    }
  },
  "comparison": {
    "delta_mean": -8.417885780334473,
    "idea_wins": 3,
    "n_pairs": 8,
    "per_seed_diffs": [
      12.055854797363281,
      21.787925720214844,
      -55.213523864746094,
      -1.1948204040527344,
      -93.02288436889648,
      20.237703323364258,
      7.248889923095703,
      20.757768630981445
    ],
    "p_value": 0.60105,
    "mde": 35.54445282513561,
    "mde_rel_pct": 65.86946992741618,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "math_check": {
      "bound_max_ratio": 0.827890485216382,
      "bound": 1.0,
      "contraction_predicted": [
        0.9619047619047619,
        0.8666666666666667,
        0.6,
        0.36
      ],
      "contraction_measured": [
        0.9619047619047619,
        0.8666666666666666,
        0.5999999999999999,
        0.35999999999999976
      ],
      "max_contraction_error": 2.220446049250313e-16
    },
    "trained_model_signature": {
      "mean_observed_diag_gramian": 0.2610868251649663,
      "mean_update_factor_predicted": 0.8549541880687078,
      "mean_update_factor_observed": 0.8544371028741201,
      "relative_factor_error": 0.0006048104118371221,
      "confirmed": true
    },
    "idea_configs": [
      {
        "cfg": {
          "lr": 0.002,
          "weight_decay": 0.0001
        },
        "result": {
          "mean": 45.54406785964966,
          "std": 8.338173492397852,
          "per_seed": [
            43.00838851928711,
            44.718257904052734,
            28.018310546875,
            49.26393508911133,
            56.6592903137207,
            41.586570739746094,
            54.953060150146484,
            46.14472961425781
          ],
          "n": 8,
          "selected_cfg": {
            "lr": 0.002,
            "weight_decay": 0.0001
          }
        }
      },
      {
        "cfg": {
          "lr": 0.002,
          "weight_decay": 0.0001
        },
        "result": {
          "mean": 45.54406785964966,
          "std": 8.338173492397852,
          "per_seed": [
            43.00838851928711,
            44.718257904052734,
            28.018310546875,
            49.26393508911133,
            56.6592903137207,
            41.586570739746094,
            54.953060150146484,
            46.14472961425781
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0001
        },
        "result": {
          "mean": 53.81347846984863,
          "std": 9.649444093972434,
          "per_seed": [
            54.86537170410156,
            67.9201431274414,
            36.08623123168945,
            58.39324188232422,
            65.55195617675781,
            47.311336517333984,
            51.460205078125,
            48.919342041015625
          ],
          "n": 8
        }
      }
    ]
  },
  "protocol_notes": "Optimizer modification on structurally matched Friedman tabular regression; paired seeds 0-7, batch 128, 12 epochs, same architecture and lr/weight-decay union."
}