{
  "bench_version": 1,
  "track": "conditional_multitoken_diffusion",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.003
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.08654761128127575
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.07295109704136848
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.07489615492522717
      }
    ],
    "full": {
      "mean": 0.07604722864925861,
      "std": 0.007765201697800962,
      "per_seed": [
        0.06751493364572525,
        0.06798635423183441,
        0.08183492720127106,
        0.07446817308664322,
        0.08383433520793915,
        0.0655355304479599,
        0.08018559962511063,
        0.0870179757475853
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.07605153322219849,
    "std": 0.007456295188304263,
    "per_seed": [
      0.06727362424135208,
      0.06813433766365051,
      0.08250146359205246,
      0.07412734627723694,
      0.0823955237865448,
      0.06638552248477936,
      0.08205603063106537,
      0.08553841710090637
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 4.304572939872742e-06,
    "idea_wins": 4,
    "n_pairs": 8,
    "per_seed_diffs": [
      -0.00024130940437316895,
      0.00014798343181610107,
      0.0006665363907814026,
      -0.0003408268094062805,
      -0.0014388114213943481,
      0.000849992036819458,
      0.0018704310059547424,
      -0.0014795586466789246
    ],
    "p_value": 1.0,
    "mde": 0.0009544449208548851,
    "mde_rel_pct": 1.2550686432728932,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "mechanism_signature": {
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.08863587398082018,
        "std": 0.005865804558271554
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.07605153322219849,
        "std": 0.007456295188304263
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.0774571718648076,
        "std": 0.0066361820117857954
      }
    ],
    "selected_idea_cfg": {
      "lr": 0.003
    },
    "mechanism_signature": {
      "prediction": "scrambled Sobol batch-mean variance lower and slope closer to -1 than IID",
      "trained_model_batch_mean_sd": {
        "iid": 0.0008616268068651571,
        "sobol": 0.0003355427185340906
      },
      "observed_log2N_slopes": {
        "iid": -0.6601498584773617,
        "sobol": -0.4193229684723195
      },
      "variance_ratio_sobol_over_iid": 0.3894292933560067,
      "confirmed": false
    },
    "custom_track": {
      "name": "conditional_multitoken_diffusion",
      "file": "bench/custom_tracks/conditional_multitoken_diffusion.py",
      "domain": "diffusion-sampling"
    }
  }
}