1{
  2  "bench_version": 1,
  3  "track": "bloch_gauge_pde",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006,
 10      "epochs": 25,
 11      "weight_decay": 0.0
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.001,
 17          "epochs": 25,
 18          "weight_decay": 0.0
 19        },
 20        "mean": 0.4257675111293793
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.003,
 25          "epochs": 25,
 26          "weight_decay": 0.0
 27        },
 28        "mean": 0.40371501445770264
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.006,
 33          "epochs": 25,
 34          "weight_decay": 0.0
 35        },
 36        "mean": 0.3903501480817795
 37      }
 38    ],
 39    "full": {
 40      "mean": 0.39177029952406883,
 41      "std": 0.013526435694561944,
 42      "per_seed": [
 43        0.38321977853775024,
 44        0.3987618088722229,
 45        0.379114031791687,
 46        0.40030497312545776,
 47        0.37645071744918823,
 48        0.3769666254520416,
 49        0.4070354998111725,
 50        0.4123089611530304
 51      ],
 52      "n": 8
 53    }
 54  },
 55  "idea": {
 56    "mean": 0.02050885744392872,
 57    "std": 0.0010667659553666602,
 58    "per_seed": [
 59      0.02256021834909916,
 60      0.02063133381307125,
 61      0.018858235329389572,
 62      0.02092020958662033,
 63      0.02036798931658268,
 64      0.021304797381162643,
 65      0.01952664740383625,
 66      0.019901428371667862
 67    ],
 68    "n": 8,
 69    "best_cfg": {
 70      "lr": 0.006,
 71      "epochs": 25,
 72      "weight_decay": 0.0
 73    },
 74    "sweep": [
 75      {
 76        "cfg": {
 77          "lr": 0.001,
 78          "epochs": 25,
 79          "weight_decay": 0.0
 80        },
 81        "mean": 0.02632583351805806
 82      },
 83      {
 84        "cfg": {
 85          "lr": 0.003,
 86          "epochs": 25,
 87          "weight_decay": 0.0
 88        },
 89        "mean": 0.02197333751246333
 90      },
 91      {
 92        "cfg": {
 93          "lr": 0.006,
 94          "epochs": 25,
 95          "weight_decay": 0.0
 96        },
 97        "mean": 0.02074249926954508
 98      }
 99    ]
100  },
101  "comparison": {
102    "delta_mean": -0.3712614420801401,
103    "idea_wins": 8,
104    "n_pairs": 8,
105    "per_seed_diffs": [
106      -0.3606595601886511,
107      -0.37813047505915165,
108      -0.36025579646229744,
109      -0.37938476353883743,
110      -0.35608272813260555,
111      -0.355661828070879,
112      -0.38750885240733624,
113      -0.39240753278136253
114    ],
115    "p_value": 0.0081,
116    "mde": 0.012368184909651097,
117    "mde_rel_pct": 3.15699912032031,
118    "verdict": "idea better (significant)",
119    "system_worked": true
120  },
121  "mechanism_signature": {
122    "prediction": "canonicalized model has much smaller trained-model reciprocal-shift gauge discrepancy",
123    "rows": {
124      "baseline": {
125        "shift_m": 1,
126        "observed_gauge_rmse": 0.8507512807846069,
127        "relative_rmse": 0.9375168128658017
128      },
129      "idea": {
130        "shift_m": 1,
131        "observed_gauge_rmse": 5.5333178039518316e-08,
132        "relative_rmse": 5.384122013910342e-08
133      }
134    },
135    "confirmed": true
136  },
137  "custom_track": {
138    "name": "bloch_gauge_pde",
139    "file": "bloch_track.py",
140    "domain": "pde"
141  }
142}