1{
  2  "bench_version": 1,
  3  "track": "tabular",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01,
 10      "weight_decay": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 17.79831099510193
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 13.153547525405884
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.01,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 7.754182696342468
 33      }
 34    ],
 35    "full": {
 36      "mean": 7.931941092014313,
 37      "std": 0.8101161551333251,
 38      "per_seed": [
 39        7.2100510597229,
 40        8.814193725585938,
 41        7.362153053283691,
 42        7.630332946777344,
 43        6.520580291748047,
 44        8.472823143005371,
 45        8.71557903289795,
 46        8.729815483093262
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 8.868400931358337,
 53    "std": 0.7219192817226784,
 54    "per_seed": [
 55      9.187010765075684,
 56      8.92486572265625,
 57      8.406270980834961,
 58      8.585136413574219,
 59      7.358145713806152,
 60      9.900330543518066,
 61      9.11246395111084,
 62      9.472983360290527
 63    ],
 64    "n": 8,
 65    "sweep": [
 66      {
 67        "cfg": {
 68          "lr": 0.001,
 69          "weight_decay": 0.0
 70        },
 71        "mean": 44.67299127578735
 72      },
 73      {
 74        "cfg": {
 75          "lr": 0.003,
 76          "weight_decay": 0.0
 77        },
 78        "mean": 12.883334636688232
 79      },
 80      {
 81        "cfg": {
 82          "lr": 0.01,
 83          "weight_decay": 0.0
 84        },
 85        "mean": 8.775820970535278
 86      }
 87    ]
 88  },
 89  "comparison": {
 90    "delta_mean": 0.9364598393440247,
 91    "idea_wins": 0,
 92    "n_pairs": 8,
 93    "per_seed_diffs": [
 94      1.9769597053527832,
 95      0.1106719970703125,
 96      1.0441179275512695,
 97      0.954803466796875,
 98      0.8375654220581055,
 99      1.4275074005126953,
100      0.3968849182128906,
101      0.7431678771972656
102    ],
103    "p_value": 0.0081,
104    "mde": 0.4850827510411753,
105    "mde_rel_pct": 6.115561694344213,
106    "verdict": "idea worse (significant)",
107    "system_worked": false
108  },
109  "mechanism_signature": {
110    "prediction": "idea starts in recovered spectral subspace, so projection energy is higher",
111    "predicted_vs_observed": {
112      "observed_energy_ratio": 1.7969543217617925,
113      "threshold": 1.2
114    },
115    "confirmed": true,
116    "trained_model_measurements": [
117      {
118        "seed": 0,
119        "spectral_gap": 2.3793046737761143,
120        "baseline_projection_energy": 0.153251051902771,
121        "idea_projection_energy": 0.28242650628089905,
122        "idea_metric": 9.187010765075684
123      },
124      {
125        "seed": 1,
126        "spectral_gap": 2.345487753974355,
127        "baseline_projection_energy": 0.14754755795001984,
128        "idea_projection_energy": 0.2536274790763855,
129        "idea_metric": 8.92486572265625
130      },
131      {
132        "seed": 2,
133        "spectral_gap": 2.5917119317759125,
134        "baseline_projection_energy": 0.1287820041179657,
135        "idea_projection_energy": 0.2661108076572418,
136        "idea_metric": 8.406270980834961
137      },
138      {
139        "seed": 3,
140        "spectral_gap": 2.0604149370605582,
141        "baseline_projection_energy": 0.15506194531917572,
142        "idea_projection_energy": 0.28266391158103943,
143        "idea_metric": 8.585136413574219
144      },
145      {
146        "seed": 4,
147        "spectral_gap": 2.544887882094244,
148        "baseline_projection_energy": 0.17158180475234985,
149        "idea_projection_energy": 0.24406076967716217,
150        "idea_metric": 7.358145713806152
151      },
152      {
153        "seed": 5,
154        "spectral_gap": 2.366815131166324,
155        "baseline_projection_energy": 0.1343323141336441,
156        "idea_projection_energy": 0.26110315322875977,
157        "idea_metric": 9.900330543518066
158      },
159      {
160        "seed": 6,
161        "spectral_gap": 2.2032378706059492,
162        "baseline_projection_energy": 0.16026736795902252,
163        "idea_projection_energy": 0.2490236610174179,
164        "idea_metric": 9.11246395111084
165      },
166      {
167        "seed": 7,
168        "spectral_gap": 2.286019756494266,
169        "baseline_projection_energy": 0.13112203776836395,
170        "idea_projection_energy": 0.2628439664840698,
171        "idea_metric": 9.472983360290527
172      }
173    ]
174  },
175  "protocol": {
176    "paired_seeds": [
177      0,
178      1,
179      2,
180      3,
181      4,
182      5,
183      6,
184      7
185    ],
186    "n_train": 400,
187    "n_test": 400,
188    "epochs": 20,
189    "batch": 128,
190    "common_grid": [
191      {
192        "lr": 0.001,
193        "weight_decay": 0.0
194      },
195      {
196        "lr": 0.003,
197        "weight_decay": 0.0
198      },
199      {
200        "lr": 0.01,
201        "weight_decay": 0.0
202      }
203    ],
204    "track_rationale": "Initialization intervention uses the designated tabular MLP track."
205  }
206}