{
  "bench_version": 1,
  "track": "tabular",
  "model": "mlp_med",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.003
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 6.546770215034485
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 5.133943796157837
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 5.272171854972839
      }
    ],
    "full": {
      "mean": 5.413264155387878,
      "std": 0.7758719453955597,
      "per_seed": [
        4.240733623504639,
        5.491628170013428,
        5.039286136627197,
        5.764127254486084,
        6.573030948638916,
        6.121939659118652,
        5.759654521942139,
        4.315712928771973
      ],
      "n": 8
    }
  },
  "idea": {
    "best_cfg": {
      "lr": 0.01
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 6.933225512504578
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 5.442892074584961
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 5.218377470970154
      }
    ],
    "full": {
      "mean": 5.436576008796692,
      "std": 0.42245263195805716,
      "per_seed": [
        5.354506015777588,
        4.885490417480469,
        5.651634693145752,
        4.981878757476807,
        5.249680995941162,
        6.110184669494629,
        6.017618656158447,
        5.241613864898682
      ],
      "n": 8
    },
    "mean": 5.436576008796692,
    "std": 0.42245263195805716,
    "per_seed": [
      5.354506015777588,
      4.885490417480469,
      5.651634693145752,
      4.981878757476807,
      5.249680995941162,
      6.110184669494629,
      6.017618656158447,
      5.241613864898682
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.023311853408813477,
    "idea_wins": 4,
    "n_pairs": 8,
    "per_seed_diffs": [
      1.1137723922729492,
      -0.606137752532959,
      0.6123485565185547,
      -0.7822484970092773,
      -1.323349952697754,
      -0.011754989624023438,
      0.2579641342163086,
      0.925900936126709
    ],
    "p_value": 0.94485,
    "mde": 0.7252095318160074,
    "mde_rel_pct": 13.396899005828097,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "mechanism_signature": {
    "claim": "trained ICNN predictions satisfy midpoint Jensen convexity",
    "predicted_max_violation": 0.0,
    "observed_max_violation": 3.814697265625e-06,
    "observed_violation_rate": 0.0,
    "confirmed": true,
    "n_models": 8
  },
  "protocol_notes": {
    "structural_match": "tabular is the built-in track for architecture/regularization interventions",
    "paired_seeds": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      7
    ],
    "epochs": 25,
    "batch": 128,
    "baseline_and_idea_share_grid": true,
    "baseline_architecture": "2-layer width-32 ReLU MLP",
    "idea_architecture": "2-layer width-32 ICNN with softplus W>=0 and nonnegative output weights"
  }
}