1{
  2  "bench_version": 1,
  3  "track": "gaussian_score_matching_1d",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.001
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 1.1268193372525275
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 1.6297587641165592
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 1.749470497830771
 29      }
 30    ],
 31    "full": {
 32      "mean": 1.5025363577296957,
 33      "std": 1.8326529003467102,
 34      "per_seed": [
 35        4.069608688354492,
 36        0.41158851981163025,
 37        0.012981797568500042,
 38        0.013098343275487423,
 39        3.647211790084839,
 40        0.0021213474683463573,
 41        0.004117978271096945,
 42        3.859562397003174
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 1.5283892883453518,
 49    "std": 1.8366178277891556,
 50    "per_seed": [
 51      3.990293502807617,
 52      0.5237103700637817,
 53      0.006270362995564938,
 54      0.006991274654865265,
 55      3.7051634788513184,
 56      0.019198978319764137,
 57      0.007155240513384342,
 58      3.9683310985565186
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.025852930615656078,
 64    "idea_wins": 3,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -0.079315185546875,
 68      0.11212185025215149,
 69      -0.006711434572935104,
 70      -0.006107068620622158,
 71      0.05795168876647949,
 72      0.01707763085141778,
 73      0.0030372622422873974,
 74      0.10876870155334473
 75    ],
 76    "p_value": 0.3043,
 77    "mde": 0.05390015205194874,
 78    "mde_rel_pct": 3.5872777237411313,
 79    "verdict": "no significant win",
 80    "system_worked": false
 81  },
 82  "custom_track": {
 83    "name": "gaussian_score_matching_1d",
 84    "file": "ksd_track_1d.py",
 85    "domain": "loss"
 86  },
 87  "idea_sweep": [
 88    {
 89      "cfg": {
 90        "lr": 0.001
 91      },
 92      "mean": 1.1318163776304573
 93    },
 94    {
 95      "cfg": {
 96        "lr": 0.003
 97      },
 98      "mean": 1.2970416366588324
 99    },
100    {
101      "cfg": {
102        "lr": 0.01
103      },
104      "mean": 1.7874763637082651
105    }
106  ],
107  "training": {
108    "epochs": 25,
109    "batch": 64,
110    "gamma": 0.5
111  },
112  "mechanism_signature": {
113    "predicted_hs_over_trace_scale": 1.0,
114    "observed_bootstrap_sd_U_over_V": 1.4147629631608514,
115    "trained_output_v": 0.11319427937269211,
116    "trained_output_u": 0.08958904214593925,
117    "test_mse": 3.990293502807617,
118    "confirmed": true
119  }
120}