1{
  2  "bench_version": 1,
  3  "track": "expert_balanced_regression",
  4  "model": "top1_moe_shared",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01,
 10      "epochs": 12,
 11      "rho": 0.5
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.001,
 17          "epochs": 12,
 18          "rho": 0.5
 19        },
 20        "mean": 0.5207640379667282
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.001,
 25          "epochs": 12,
 26          "rho": 1.0
 27        },
 28        "mean": 0.5207640379667282
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.003,
 33          "epochs": 12,
 34          "rho": 0.5
 35        },
 36        "mean": 0.3618839532136917
 37      },
 38      {
 39        "cfg": {
 40          "lr": 0.003,
 41          "epochs": 12,
 42          "rho": 1.0
 43        },
 44        "mean": 0.3618839532136917
 45      },
 46      {
 47        "cfg": {
 48          "lr": 0.01,
 49          "epochs": 12,
 50          "rho": 0.5
 51        },
 52        "mean": 0.21788224577903748
 53      },
 54      {
 55        "cfg": {
 56          "lr": 0.01,
 57          "epochs": 12,
 58          "rho": 1.0
 59        },
 60        "mean": 0.21788224577903748
 61      }
 62    ],
 63    "full": {
 64      "mean": 0.21290172263979912,
 65      "std": 0.02201949504273137,
 66      "per_seed": [
 67        0.23030464351177216,
 68        0.19820281863212585,
 69        0.18614302575588226,
 70        0.25687849521636963,
 71        0.21424928307533264,
 72        0.18679030239582062,
 73        0.21802611649036407,
 74        0.2126190960407257
 75      ],
 76      "n": 8
 77    }
 78  },
 79  "idea": {
 80    "mean": 0.2056338880211115,
 81    "std": 0.020574444566337256,
 82    "per_seed": [
 83      0.2081810086965561,
 84      0.20322394371032715,
 85      0.18413016200065613,
 86      0.24816179275512695,
 87      0.19936268031597137,
 88      0.17569208145141602,
 89      0.20796272158622742,
 90      0.21835671365261078
 91    ],
 92    "n": 8,
 93    "config": {
 94      "lr": 0.01,
 95      "epochs": 12,
 96      "rho": 1.0
 97    }
 98  },
 99  "comparison": {
100    "delta_mean": -0.00726783461868763,
101    "idea_wins": 6,
102    "n_pairs": 8,
103    "per_seed_diffs": [
104      -0.022123634815216064,
105      0.005021125078201294,
106      -0.0020128637552261353,
107      -0.008716702461242676,
108      -0.014886602759361267,
109      -0.011098220944404602,
110      -0.010063394904136658,
111      0.005737617611885071
112    ],
113    "p_value": 0.08055,
114    "mde": 0.008064204749019211,
115    "mde_rel_pct": 3.7877592764539316,
116    "verdict": "no significant win",
117    "system_worked": false
118  },
119  "mechanism_signature": {
120    "prediction": "trained VDC assignments maintain max leaf prefix discrepancy <=1",
121    "predicted_max_discrepancy": 1.0,
122    "observed_mean_max_discrepancy": 0.875,
123    "observed_final_load_variance": 0.0,
124    "confirmed": true
125  },
126  "custom_track": {
127    "name": "expert_balanced_regression",
128    "file": "expert_balanced_regression.py",
129    "domain": "moe-routing"
130  }
131}