{
  "bench_version": 1,
  "track": "tabular",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "epochs": 12,
      "alpha": 0.15
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015,
          "epochs": 12,
          "alpha": 0.05
        },
        "mean": 116.23149108886719
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 12,
          "alpha": 0.1
        },
        "mean": 29.072664737701416
      },
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 12,
          "alpha": 0.15
        },
        "mean": 20.75340461730957
      }
    ],
    "full": {
      "mean": 20.881689310073853,
      "std": 2.8097707761369097,
      "per_seed": [
        24.462068557739258,
        22.150100708007812,
        18.394702911376953,
        18.006746292114258,
        16.289684295654297,
        22.511140823364258,
        21.119823455810547,
        24.119247436523438
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 20.881689310073853,
    "std": 2.8097707761369097,
    "per_seed": [
      24.462068557739258,
      22.150100708007812,
      18.394702911376953,
      18.006746292114258,
      16.289684295654297,
      22.511140823364258,
      21.119823455810547,
      24.119247436523438
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.0,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.0,
      0.0,
      0.0,
      0.0,
      0.0,
      0.0,
      0.0,
      0.0
    ],
    "p_value": 1.0,
    "mde": 0.0,
    "mde_rel_pct": 0.0,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "selection": {
    "baseline_grid": [
      {
        "lr": 0.0015,
        "epochs": 12,
        "alpha": 0.05
      },
      {
        "lr": 0.003,
        "epochs": 12,
        "alpha": 0.1
      },
      {
        "lr": 0.006,
        "epochs": 12,
        "alpha": 0.15
      }
    ],
    "idea_grid": [
      {
        "lr": 0.0015,
        "epochs": 12,
        "alpha": 0.05
      },
      {
        "lr": 0.003,
        "epochs": 12,
        "alpha": 0.1
      },
      {
        "lr": 0.006,
        "epochs": 12,
        "alpha": 0.15
      }
    ],
    "baseline_best": {
      "lr": 0.006,
      "epochs": 12,
      "alpha": 0.15
    },
    "idea_best": {
      "lr": 0.006,
      "epochs": 12,
      "alpha": 0.15
    }
  },
  "structural_match": "Independent neural calibration residuals define a scalar acceptance set; explicit order-statistic boundary is the stated projective structure.",
  "mechanism_signature": {
    "prediction": "trained scalar residual order-statistic has K=1 and projective deletion; risk approximately Beta(1,n) mean",
    "observed_mean": {
      "boundary_size": 15.0,
      "empirical_test_risk": 0.175,
      "beta_predicted_mean_risk": 0.1485148514851485,
      "deletion_projectivity": 0.85,
      "deletion_equivalence": 0.99
    },
    "per_seed": [
      {
        "threshold": 6.851143836975098,
        "boundary_size": 15,
        "empirical_test_risk": 0.17,
        "beta_predicted_mean_risk": 0.1485148514851485,
        "deletion_projectivity": 0.85,
        "deletion_equivalence": 0.99
      },
      {
        "threshold": 5.987270355224609,
        "boundary_size": 15,
        "empirical_test_risk": 0.2,
        "beta_predicted_mean_risk": 0.1485148514851485,
        "deletion_projectivity": 0.85,
        "deletion_equivalence": 0.99
      },
      {
        "threshold": 5.549269676208496,
        "boundary_size": 15,
        "empirical_test_risk": 0.2175,
        "beta_predicted_mean_risk": 0.1485148514851485,
        "deletion_projectivity": 0.85,
        "deletion_equivalence": 0.99
      },
      {
        "threshold": 6.736217498779297,
        "boundary_size": 15,
        "empirical_test_risk": 0.115,
        "beta_predicted_mean_risk": 0.1485148514851485,
        "deletion_projectivity": 0.85,
        "deletion_equivalence": 0.99
      },
      {
        "threshold": 6.043928146362305,
        "boundary_size": 15,
        "empirical_test_risk": 0.14,
        "beta_predicted_mean_risk": 0.1485148514851485,
        "deletion_projectivity": 0.85,
        "deletion_equivalence": 0.99
      },
      {
        "threshold": 5.98091983795166,
        "boundary_size": 15,
        "empirical_test_risk": 0.235,
        "beta_predicted_mean_risk": 0.1485148514851485,
        "deletion_projectivity": 0.85,
        "deletion_equivalence": 0.99
      },
      {
        "threshold": 5.88011360168457,
        "boundary_size": 15,
        "empirical_test_risk": 0.225,
        "beta_predicted_mean_risk": 0.1485148514851485,
        "deletion_projectivity": 0.85,
        "deletion_equivalence": 0.99
      },
      {
        "threshold": 8.088789939880371,
        "boundary_size": 15,
        "empirical_test_risk": 0.0975,
        "beta_predicted_mean_risk": 0.1485148514851485,
        "deletion_projectivity": 0.85,
        "deletion_equivalence": 0.99
      }
    ],
    "confirmed": false
  }
}