{
  "bench_version": 1,
  "track": "vision",
  "model": "cnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.0015
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015
        },
        "mean": 0.7512499988079071
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.7537499964237213
      },
      {
        "cfg": {
          "lr": 0.006
        },
        "mean": 0.878125011920929
      }
    ],
    "full": {
      "mean": 0.7653124928474426,
      "std": 0.035143397058649814,
      "per_seed": [
        0.7350000143051147,
        0.7825000286102295,
        0.7799999713897705,
        0.7074999809265137,
        0.8324999809265137,
        0.7699999809265137,
        0.7425000071525574,
        0.7724999785423279
      ],
      "n": 8
    },
    "idea_union_sweep": [
      {
        "cfg": {
          "lr": 0.0015
        },
        "mean": 0.7781250029802322
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.7737499922513962
      },
      {
        "cfg": {
          "lr": 0.006
        },
        "mean": 0.7937500029802322
      }
    ]
  },
  "idea": {
    "mean": 0.7856250032782555,
    "std": 0.029993484958144172,
    "per_seed": [
      0.7699999809265137,
      0.8374999761581421,
      0.737500011920929,
      0.75,
      0.7975000143051147,
      0.7975000143051147,
      0.8050000071525574,
      0.7900000214576721
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.020312510430812836,
    "idea_wins": 2,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.034999966621398926,
      0.0549999475479126,
      -0.04249995946884155,
      0.04250001907348633,
      -0.034999966621398926,
      0.027500033378601074,
      0.0625,
      0.01750004291534424
    ],
    "p_value": 0.17155,
    "mde": 0.03277566818472898,
    "mde_rel_pct": 4.282651660733112,
    "verdict": "no significant win",
    "system_worked": false
  },
  "parameterization": {
    "alpha": 0.5,
    "r": 0.0,
    "epochs": 8,
    "batch": 128,
    "lr_union": [
      0.0015,
      0.003,
      0.006
    ],
    "task": "CIFAR-10 subset classification"
  },
  "mechanism_signature": {
    "quantity": "mean probability assigned to observed target event on trained test models",
    "baseline_target_event_prob": 0.14888361282646656,
    "idea_target_event_prob": 0.16992626152932644,
    "prediction": "r=0, alpha=0.5 changes power-law gradient weighting and should reduce gradient variability",
    "observed_baseline_grad_var": 3.603046359352762,
    "observed_idea_grad_var": 0.48674841174291406,
    "confirmed": true
  }
}