{
  "bench_version": 1,
  "track": "tabular",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 187.7414321899414
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 19.22788381576538
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 15.669923067092896
      }
    ],
    "full": {
      "mean": 16.668858289718628,
      "std": 2.7275460099983677,
      "per_seed": [
        15.673293113708496,
        11.123649597167969,
        17.505971908569336,
        18.37677764892578,
        14.736111640930176,
        21.006744384765625,
        17.87649917602539,
        17.05181884765625
      ],
      "n": 8
    },
    "idea_parity_sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 8.131636142730713
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 11.476103067398071
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 9.458438754081726
      }
    ]
  },
  "idea": {
    "mean": 9.025225281715393,
    "std": 1.7380340756046946,
    "per_seed": [
      8.675573348999023,
      6.953583717346191,
      8.074329376220703,
      8.823058128356934,
      13.168746948242188,
      8.126770973205566,
      8.566627502441406,
      9.813112258911133
    ],
    "n": 8,
    "selected_cfg": {
      "lr": 0.001
    }
  },
  "comparison": {
    "delta_mean": -7.643633008003235,
    "idea_wins": 8,
    "n_pairs": 8,
    "per_seed_diffs": [
      -6.997719764709473,
      -4.170065879821777,
      -9.431642532348633,
      -9.553719520568848,
      -1.5673646926879883,
      -12.879973411560059,
      -9.309871673583984,
      -7.238706588745117
    ],
    "p_value": 0.0081,
    "mde": 2.941158171817024,
    "mde_rel_pct": 17.644628808387758,
    "verdict": "idea better (significant)",
    "system_worked": true
  },
  "mechanism_signature": {
    "alpha": 1.0,
    "rho": 0.9,
    "prediction": "accepted_eta ~= rho / observed_directional_Lhat",
    "per_seed": [
      {
        "predicted_eta": 0.004098492150815332,
        "observed_eta": 0.021143166023050074,
        "ratio_observed_to_predicted": 5.1587669916225085
      },
      {
        "predicted_eta": 0.004114928975852553,
        "observed_eta": 0.015304616531639226,
        "ratio_observed_to_predicted": 3.719290568913971
      },
      {
        "predicted_eta": 0.003983274857888152,
        "observed_eta": 0.0124441517321635,
        "ratio_observed_to_predicted": 3.1241006900440524
      },
      {
        "predicted_eta": 0.004497933758441668,
        "observed_eta": 0.031266616003165305,
        "ratio_observed_to_predicted": 6.951328694977888
      },
      {
        "predicted_eta": 0.0055342093536037895,
        "observed_eta": 0.015162049792218038,
        "ratio_observed_to_predicted": 2.7396957403399913
      },
      {
        "predicted_eta": 0.004462357038057677,
        "observed_eta": 0.02690663006815206,
        "ratio_observed_to_predicted": 6.029690102937991
      },
      {
        "predicted_eta": 0.003190808863205975,
        "observed_eta": 0.021336048993906148,
        "ratio_observed_to_predicted": 6.686721113237942
      },
      {
        "predicted_eta": 0.0022129334132261974,
        "observed_eta": 0.012751969561600147,
        "ratio_observed_to_predicted": 5.762473233665567
      }
    ],
    "mean_ratio": 5.0215083919674885,
    "confirmed": false
  },
  "runtime_seconds": 7.080888032913208,
  "protocol_notes": {
    "epochs": 8,
    "batch": 128,
    "dataset_sizes": [
      400,
      200
    ],
    "structural_match": "optimizer -> tabular"
  }
}