1{
  2  "bench_version": 1,
  3  "track": "phase_synchronization_regression",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.5544544011354446
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.5093307048082352
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.4870074987411499
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.7184864804148674,
 33      "std": 0.6005355689639164,
 34      "per_seed": [
 35        0.5706357359886169,
 36        0.43496403098106384,
 37        0.42483416199684143,
 38        0.5175960659980774,
 39        0.4055282473564148,
 40        0.73640376329422,
 41        0.3769463896751404,
 42        2.2809834480285645
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.21742316987365484,
 49    "std": 0.0751846884485342,
 50    "per_seed": [
 51      0.21176113188266754,
 52      0.1522442102432251,
 53      0.22154822945594788,
 54      0.09710203856229782,
 55      0.3232591152191162,
 56      0.24092552065849304,
 57      0.1655595749616623,
 58      0.32698553800582886
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -0.5010633105412126,
 64    "idea_wins": 8,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -0.3588746041059494,
 68      -0.28271982073783875,
 69      -0.20328593254089355,
 70      -0.42049402743577957,
 71      -0.08226913213729858,
 72      -0.49547824263572693,
 73      -0.2113868147134781,
 74      -1.9539979100227356
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.5030248995072959,
 78    "mde_rel_pct": 70.01174179600986,
 79    "verdict": "idea better (significant)",
 80    "system_worked": true
 81  },
 82  "custom_track": {
 83    "name": "phase_synchronization_regression",
 84    "file": "phase_track.py",
 85    "domain": "geometry/graph synchronization"
 86  },
 87  "idea_sweep": [
 88    {
 89      "cfg": {
 90        "lr": 0.001
 91      },
 92      "result": {
 93        "mean": 0.2453548777848482,
 94        "std": 0.07094670006301455,
 95        "per_seed": [
 96          0.27867943048477173,
 97          0.1840101033449173,
 98          0.20311710238456726,
 99          0.15075239539146423,
100          0.3714803159236908,
101          0.24977633357048035,
102          0.19869858026504517,
103          0.3263247609138489
104        ],
105        "n": 8
106      }
107    },
108    {
109      "cfg": {
110        "lr": 0.003
111      },
112      "result": {
113        "mean": 0.21742316987365484,
114        "std": 0.0751846884485342,
115        "per_seed": [
116          0.21176113188266754,
117          0.1522442102432251,
118          0.22154822945594788,
119          0.09710203856229782,
120          0.3232591152191162,
121          0.24092552065849304,
122          0.1655595749616623,
123          0.32698553800582886
124        ],
125        "n": 8
126      }
127    },
128    {
129      "cfg": {
130        "lr": 0.01
131      },
132      "result": {
133        "mean": 0.2902024118229747,
134        "std": 0.1461298164928991,
135        "per_seed": [
136          0.3044014573097229,
137          0.08412133902311325,
138          0.27784958481788635,
139          0.13880692422389984,
140          0.478528767824173,
141          0.47114890813827515,
142          0.14776097238063812,
143          0.41900134086608887
144        ],
145        "n": 8
146      }
147    }
148  ],
149  "runtime_sec": 49.546178102493286,
150  "mechanism_signature": {
151    "prediction": "spectral alignment reduces downstream test MSE and residual spread",
152    "predicted_baseline_mse": 0.4704003967344761,
153    "observed_aligned_mse": 0.3038702141493559,
154    "observed_aligned_residual_std": 0.4300309270620346,
155    "confirmed": true
156  }
157}