{
  "bench_version": 1,
  "track": "tabular",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "epochs": 12
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "epochs": 12
        },
        "mean": 201.0537109375
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 12
        },
        "mean": 58.3254508972168
      },
      {
        "cfg": {
          "lr": 0.01,
          "epochs": 12
        },
        "mean": 15.770377159118652
      }
    ],
    "full": {
      "mean": 16.54259741306305,
      "std": 2.506983808304334,
      "per_seed": [
        16.479936599731445,
        13.55350399017334,
        18.048686981201172,
        14.999381065368652,
        14.820082664489746,
        16.00037956237793,
        16.12912368774414,
        22.30968475341797
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 16.655847430229187,
    "std": 2.25515162281877,
    "per_seed": [
      17.07933807373047,
      14.434268951416016,
      16.626239776611328,
      15.863788604736328,
      14.660067558288574,
      16.29985809326172,
      16.1023006439209,
      22.180917739868164
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.1132500171661377,
    "idea_wins": 4,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.5994014739990234,
      0.8807649612426758,
      -1.4224472045898438,
      0.8644075393676758,
      -0.16001510620117188,
      0.29947853088378906,
      -0.026823043823242188,
      -0.1287670135498047
    ],
    "p_value": 0.6628,
    "mde": 0.6269619015296144,
    "mde_rel_pct": 3.7899846431282116,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "idea_config": {
    "lr": 0.01,
    "epochs": 12
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001,
        "epochs": 12
      },
      "result": {
        "mean": 207.19963264465332,
        "std": 20.029543267954704,
        "per_seed": [
          213.74232482910156,
          188.0177001953125,
          228.82228088378906,
          174.27357482910156,
          202.07249450683594,
          209.18301391601562,
          200.1274871826172,
          241.35818481445312
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.003,
        "epochs": 12
      },
      "result": {
        "mean": 61.74348449707031,
        "std": 23.30208403807316,
        "per_seed": [
          52.126747131347656,
          57.39317321777344,
          69.72927856445312,
          56.225345611572266,
          38.751182556152344,
          66.92098999023438,
          36.6944465637207,
          116.1067123413086
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.01,
        "epochs": 12
      },
      "result": {
        "mean": 16.655847430229187,
        "std": 2.25515162281877,
        "per_seed": [
          17.07933807373047,
          14.434268951416016,
          16.626239776611328,
          15.863788604736328,
          14.660067558288574,
          16.29985809326172,
          16.1023006439209,
          22.180917739868164
        ],
        "n": 8
      }
    }
  ],
  "mechanism_signature": {
    "prediction": "training lowers symmetric nearest cross-color embedding distance",
    "predicted_direction": "lower",
    "observed_idea_cross_color_nearest_mean": 0.07249955972656608,
    "observed_idea_cross_color_nearest_per_seed": [
      0.07713642716407776,
      0.06891444325447083,
      0.08666811883449554,
      0.060866888612508774,
      0.06561814993619919,
      0.08363944292068481,
      0.06347937881946564,
      0.07367362827062607
    ],
    "confirmed": true
  }
}