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": 0.0001
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 15.90550446510315
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "weight_decay": 0.0001
 24        },
 25        "mean": 15.877723932266235
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 9.71069884300232
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "weight_decay": 0.0001
 38        },
 39        "mean": 9.578425407409668
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.01,
 44          "weight_decay": 0.0
 45        },
 46        "mean": 7.003860950469971
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.01,
 51          "weight_decay": 0.0001
 52        },
 53        "mean": 6.940991640090942
 54      }
 55    ],
 56    "full": {
 57      "mean": 7.183188855648041,
 58      "std": 0.6070482381931488,
 59      "per_seed": [
 60        6.755654335021973,
 61        7.163381099700928,
 62        6.924261093139648,
 63        6.920670032501221,
 64        6.168270111083984,
 65        7.485415458679199,
 66        7.807327747344971,
 67        8.240530967712402
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 220.65338706970215,
 74    "std": 6.711567657993523,
 75    "per_seed": [
 76      227.8157958984375,
 77      220.93875122070312,
 78      227.989990234375,
 79      210.31065368652344,
 80      223.4892578125,
 81      218.9625701904297,
 82      210.06019592285156,
 83      225.65988159179688
 84    ],
 85    "n": 8
 86  },
 87  "comparison": {
 88    "delta_mean": 213.4701982140541,
 89    "idea_wins": 0,
 90    "n_pairs": 8,
 91    "per_seed_diffs": [
 92      221.06014156341553,
 93      213.7753701210022,
 94      221.06572914123535,
 95      203.38998365402222,
 96      217.32098770141602,
 97      211.4771547317505,
 98      202.2528681755066,
 99      217.41935062408447
100    ],
101    "p_value": 0.0081,
102    "mde": 6.138174864466896,
103    "mde_rel_pct": 85.45194881853251,
104    "verdict": "idea worse (significant)",
105    "system_worked": false
106  },
107  "mechanism_signature": {
108    "prediction": "noise enters acceleration only; at temperature=0 injected-noise std is zero",
109    "predicted_injected_noise_std": 0.0,
110    "observed_trained_acceleration_norm_mean": 15.645100752559667,
111    "observed_trained_gradient_norm_mean": 98.2094500956436,
112    "observed_acceleration_to_gradient_ratio_mean": 0.1593389100379601,
113    "confirmed": false,
114    "note": "State acceleration is nonzero from deterministic gradients, so this does not confirm a zero total-state prediction; it directly measures trained-model behavior."
115  },
116  "idea_sweep": [
117    {
118      "cfg": {
119        "lr": 0.001,
120        "gamma": 0.5,
121        "temperature": 0.0
122      },
123      "result": {
124        "mean": 235.77590942382812,
125        "std": 7.662576847576715,
126        "per_seed": [
127          243.6631317138672,
128          235.733154296875,
129          241.19017028808594,
130          221.46832275390625,
131          245.36273193359375,
132          231.23699951171875,
133          228.62933349609375,
134          238.92343139648438
135        ],
136        "n": 8
137      }
138    },
139    {
140      "cfg": {
141        "lr": 0.003,
142        "gamma": 1.0,
143        "temperature": 0.0
144      },
145      "result": {
146        "mean": 220.65338706970215,
147        "std": 6.711567657993523,
148        "per_seed": [
149          227.8157958984375,
150          220.93875122070312,
151          227.989990234375,
152          210.31065368652344,
153          223.4892578125,
154          218.9625701904297,
155          210.06019592285156,
156          225.65988159179688
157        ],
158        "n": 8
159      }
160    },
161    {
162      "cfg": {
163        "lr": 0.01,
164        "gamma": 2.0,
165        "temperature": 0.0
166      },
167      "result": {
168        "mean": 471.48626041412354,
169        "std": 384.5950909075575,
170        "per_seed": [
171          105.4653091430664,
172          324.9513854980469,
173          402.4331970214844,
174          1098.881591796875,
175          90.00657653808594,
176          657.0219116210938,
177          78.45921325683594,
178          1014.6708984375
179        ],
180        "n": 8
181      }
182    }
183  ],
184  "protocol_notes": {
185    "paired_seeds": [
186      0,
187      1,
188      2,
189      3,
190      4,
191      5,
192      6,
193      7
194    ],
195    "baseline_sweep_seeds": [
196      0,
197      1,
198      2,
199      3
200    ],
201    "dataset_sizes": [
202      400,
203      400
204    ],
205    "epochs": 30,
206    "batch": 128,
207    "baseline": "Adam via bench.train_model",
208    "idea": "third-order Langevin; only optimizer loop differs",
209    "search_space_union_learning_rates": [
210      0.001,
211      0.003,
212      0.01
213    ]
214  }
215}