{
  "bench_version": 1,
  "track": "conditional_multitoken_diffusion",
  "model": "shared_encoder_mlp",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "epochs": 12
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "epochs": 12
        },
        "mean": 0.390290267765522
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 12
        },
        "mean": 0.3733091354370117
      },
      {
        "cfg": {
          "lr": 0.01,
          "epochs": 12
        },
        "mean": 0.36269813030958176
      }
    ],
    "full": {
      "mean": 0.3607390485703945,
      "std": 0.01654029145653189,
      "per_seed": [
        0.37062346935272217,
        0.32732465863227844,
        0.3899999260902405,
        0.36284446716308594,
        0.3566600978374481,
        0.3669852316379547,
        0.3537239730358124,
        0.3577505648136139
      ],
      "n": 8
    }
  },
  "idea": {
    "per_seed": [
      0.829505980014801,
      0.8032728433609009,
      0.6855372190475464,
      0.7373040318489075,
      0.7540731430053711,
      0.773855447769165,
      0.7264139652252197,
      0.7292614579200745
    ],
    "mean": 0.7549030110239983,
    "std": 0.046067620085947815
  },
  "comparison": {
    "delta_mean": 0.39416396245360374,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.45888251066207886,
      0.47594818472862244,
      0.2955372929573059,
      0.37445956468582153,
      0.397413045167923,
      0.4068702161312103,
      0.37268999218940735,
      0.37151089310646057
    ],
    "p_value": 0.0081,
    "mde": 0.047023554672292786,
    "mde_rel_pct": 13.03533810898673,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "mechanism_signature": {
      "observed_token_std": 0.7545899152755737,
      "predicted_token_std": 0.7607559561729431,
      "observed_sample_mean_std": 0.14106108248233795,
      "predicted_sample_mean_std": 0.1528526395559311,
      "confirmed": true
    },
    "idea_lambda": 8.0,
    "idea_runs": [
      {
        "lambda": 0.0,
        "result": {
          "per_seed": [
            0.924494743347168,
            1.0625691413879395,
            0.7344903349876404,
            0.912911593914032,
            0.8331963419914246,
            0.8818286657333374,
            0.8657779693603516,
            0.7827775478363037
          ],
          "mean": 0.8747557923197746,
          "std": 0.09942397904692929
        }
      },
      {
        "lambda": 2.0,
        "result": {
          "per_seed": [
            0.871182918548584,
            0.8115776777267456,
            0.6836869716644287,
            0.7553004026412964,
            0.7379289865493774,
            0.782780110836029,
            0.7276151776313782,
            0.7247848510742188
          ],
          "mean": 0.7618571370840073,
          "std": 0.058718477961378
        }
      },
      {
        "lambda": 8.0,
        "result": {
          "per_seed": [
            0.829505980014801,
            0.8032728433609009,
            0.6855372190475464,
            0.7373040318489075,
            0.7540731430053711,
            0.773855447769165,
            0.7264139652252197,
            0.7292614579200745
          ],
          "mean": 0.7549030110239983,
          "std": 0.046067620085947815
        }
      }
    ]
  }
}