{
  "bench_version": 1,
  "track": "vision",
  "model": "cnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.001
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.8866666704416275
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.8950000107288361
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.908333346247673
      }
    ],
    "full": {
      "mean": 0.8875000029802322,
      "std": 0.02548147140610766,
      "per_seed": [
        0.9133333563804626,
        0.8999999761581421,
        0.8666666746139526,
        0.8666666746139526,
        0.8733333349227905,
        0.8933333158493042,
        0.9333333373069763,
        0.8533333539962769
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.890000008046627,
    "std": 0.027284505974877197,
    "per_seed": [
      0.9133333563804626,
      0.8999999761581421,
      0.8533333539962769,
      0.8600000143051147,
      0.9066666960716248,
      0.8933333158493042,
      0.9333333373069763,
      0.8600000143051147
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.002500005066394806,
    "idea_wins": 2,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.0,
      0.0,
      -0.013333320617675781,
      -0.006666660308837891,
      0.03333336114883423,
      0.0,
      0.0,
      0.006666660308837891
    ],
    "p_value": 0.8775,
    "mde": 0.011515956518988579,
    "mde_rel_pct": 1.2975725611625806,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "mechanism_signature": {
    "track_choice": "vision: categorical class probability output matches exact DPD integral; label corruption tests robustness",
    "noise_rate": 0.2,
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "alpha": 0.3
        },
        "result": {
          "mean": 0.8916666731238365,
          "std": 0.02920236995935498,
          "per_seed": [
            0.9133333563804626,
            0.9133333563804626,
            0.8666666746139526,
            0.8666666746139526,
            0.9066666960716248,
            0.8933333158493042,
            0.9333333373069763,
            0.8399999737739563
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.001,
          "alpha": 0.5
        },
        "result": {
          "mean": 0.890000008046627,
          "std": 0.027284505974877197,
          "per_seed": [
            0.9133333563804626,
            0.8999999761581421,
            0.8533333539962769,
            0.8600000143051147,
            0.9066666960716248,
            0.8933333158493042,
            0.9333333373069763,
            0.8600000143051147
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.001,
          "alpha": 0.7
        },
        "result": {
          "mean": 0.8900000154972076,
          "std": 0.029627313293764897,
          "per_seed": [
            0.9133333563804626,
            0.9133333563804626,
            0.846666693687439,
            0.8533333539962769,
            0.9066666960716248,
            0.8866666555404663,
            0.9333333373069763,
            0.8666666746139526
          ],
          "n": 8
        }
      }
    ],
    "mechanism_signature": {
      "prediction": "observed-label score gradient ratio equals (1+alpha)*q^alpha",
      "alpha": 0.5,
      "n_corrupted": 48,
      "mean_predicted_ratio": 0.48781922459602356,
      "mean_observed_ratio": 0.48781922459602356,
      "max_relative_error": 3.0113764637462737e-07,
      "corrupted_mean_q": 0.10800278186798096,
      "confirmed": true
    }
  }
}