{
  "bench_version": 1,
  "track": "conditional_multitoken_diffusion",
  "model": "transformer_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.001
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.09234148263931274
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.10227461904287338
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.13437412679195404
      }
    ],
    "full": {
      "mean": 0.09546016063541174,
      "std": 0.0068981652474538425,
      "per_seed": [
        0.08567251265048981,
        0.08461520820856094,
        0.09914777427911758,
        0.09993043541908264,
        0.10520027577877045,
        0.09470903873443604,
        0.09319207072257996,
        0.1012139692902565
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.09004521276801825,
    "std": 0.0046136095927630846,
    "per_seed": [
      0.08257821947336197,
      0.08880516141653061,
      0.09218183159828186,
      0.09180948883295059,
      0.0943688154220581,
      0.08264964818954468,
      0.09331666678190231,
      0.09465187042951584
    ],
    "n": 8,
    "cfg": {
      "lr": 0.001,
      "components": 3
    }
  },
  "comparison": {
    "delta_mean": -0.005414947867393494,
    "idea_wins": 6,
    "n_pairs": 8,
    "per_seed_diffs": [
      -0.003094293177127838,
      0.0041899532079696655,
      -0.006965942680835724,
      -0.00812094658613205,
      -0.010831460356712341,
      -0.012059390544891357,
      0.00012459605932235718,
      -0.006562098860740662
    ],
    "p_value": 0.0468,
    "mde": 0.004610854431296922,
    "mde_rel_pct": 4.830134791944282,
    "verdict": "idea better (significant)",
    "system_worked": true
  },
  "mechanism_signature": {
    "mean_component_spread": 1.1394079849123955,
    "residual_excess_kurtosis": 1.5870434984444928,
    "prediction": "multimodal conditional transport has separated learned components",
    "confirmed": true
  },
  "idea_sweep": [
    {
      "mean": 0.09004521276801825,
      "std": 0.0046136095927630846,
      "per_seed": [
        0.08257821947336197,
        0.08880516141653061,
        0.09218183159828186,
        0.09180948883295059,
        0.0943688154220581,
        0.08264964818954468,
        0.09331666678190231,
        0.09465187042951584
      ],
      "n": 8,
      "cfg": {
        "lr": 0.001,
        "components": 3
      }
    },
    {
      "mean": 0.11219676397740841,
      "std": 0.012698939549116916,
      "per_seed": [
        0.10995763540267944,
        0.10517711192369461,
        0.10637346655130386,
        0.10468488186597824,
        0.1260760873556137,
        0.09370528906583786,
        0.11482667922973633,
        0.13677296042442322
      ],
      "n": 8,
      "cfg": {
        "lr": 0.003,
        "components": 3
      }
    },
    {
      "mean": 0.2344836015254259,
      "std": 0.07579298087152198,
      "per_seed": [
        0.21607452630996704,
        0.25600194931030273,
        0.2064877152442932,
        0.16402755677700043,
        0.15939156711101532,
        0.41217851638793945,
        0.263763964176178,
        0.19794301688671112
      ],
      "n": 8,
      "cfg": {
        "lr": 0.01,
        "components": 3
      }
    }
  ],
  "math_check": {
    "normal_inverse_cdf_pit_sup_deviation": 0.006606725739433772,
    "expected_below": 0.02
  },
  "custom_track": {
    "name": "conditional_multitoken_diffusion",
    "file": "/home/maxwelhelp/all/math2nn/bench/custom_tracks/conditional_multitoken_diffusion.py",
    "domain": "diffusion-sampling"
  },
  "protocol_notes": "Matched eight-token diffusion task; independently trained shared transformer backbone; identical test MSE readout; baseline and idea share lr union."
}