1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "rnn_small",
  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": 0.0014785313687752932
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "weight_decay": 0.0001
 24        },
 25        "mean": 0.0014645301271229982
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.0008221298194257542
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "weight_decay": 0.0001
 38        },
 39        "mean": 0.0008076394879026338
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.01,
 44          "weight_decay": 0.0
 45        },
 46        "mean": 0.00032767555967438966
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.01,
 51          "weight_decay": 0.0001
 52        },
 53        "mean": 0.00036696675670100376
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.0002959608482342446,
 58      "std": 0.0001316782688333712,
 59      "per_seed": [
 60        0.0004682304570451379,
 61        0.0003980887704528868,
 62        0.00017163399024866521,
 63        0.0002727490209508687,
 64        0.00015086510393302888,
 65        0.00030392719781957567,
 66        0.00012582281487993896,
 67        0.00047636943054385483
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 0.001754267912474461,
 74    "std": 0.00043015553066540066,
 75    "per_seed": [
 76      0.0012427795445546508,
 77      0.001973426202312112,
 78      0.0013205275172367692,
 79      0.0024030047934502363,
 80      0.002149099251255393,
 81      0.0012187783140689135,
 82      0.0020814703311771154,
 83      0.0016450573457404971
 84    ],
 85    "n": 8,
 86    "sweep": [
 87      {
 88        "cfg": {
 89          "lr": 0.001,
 90          "weight_decay": 0.0
 91        },
 92        "mean": 0.024796552257612348,
 93        "std": 0.004088271419227938,
 94        "per_seed": [
 95          0.022675156593322754,
 96          0.02744518220424652,
 97          0.023942390456795692,
 98          0.03346814215183258,
 99          0.025026222690939903,
100          0.019205331802368164,
101          0.021013041958212852,
102          0.025596950203180313
103        ]
104      },
105      {
106        "cfg": {
107          "lr": 0.001,
108          "weight_decay": 0.0001
109        },
110        "mean": 0.024904589634388685,
111        "std": 0.004089925380114752,
112        "per_seed": [
113          0.022724661976099014,
114          0.02740549109876156,
115          0.024043483659625053,
116          0.03363170102238655,
117          0.025220675393939018,
118          0.01937870867550373,
119          0.021087665110826492,
120          0.025744330137968063
121        ]
122      },
123      {
124        "cfg": {
125          "lr": 0.003,
126          "weight_decay": 0.0
127        },
128        "mean": 0.006038541585439816,
129        "std": 0.002237967748799504,
130        "per_seed": [
131          0.00579321151599288,
132          0.006784805096685886,
133          0.005182903725653887,
134          0.011018984019756317,
135          0.005261050537228584,
136          0.0027950687799602747,
137          0.00687775295227766,
138          0.004594556055963039
139        ]
140      },
141      {
142        "cfg": {
143          "lr": 0.003,
144          "weight_decay": 0.0001
145        },
146        "mean": 0.006389912683516741,
147        "std": 0.0023475474959639103,
148        "per_seed": [
149          0.006135317496955395,
150          0.007102395873516798,
151          0.005393301136791706,
152          0.011719641275703907,
153          0.005564150866121054,
154          0.0029893447645008564,
155          0.007032498717308044,
156          0.005182651337236166
157        ]
158      },
159      {
160        "cfg": {
161          "lr": 0.01,
162          "weight_decay": 0.0
163        },
164        "mean": 0.001754267912474461,
165        "std": 0.00043015553066540066,
166        "per_seed": [
167          0.0012427795445546508,
168          0.001973426202312112,
169          0.0013205275172367692,
170          0.0024030047934502363,
171          0.002149099251255393,
172          0.0012187783140689135,
173          0.0020814703311771154,
174          0.0016450573457404971
175        ]
176      },
177      {
178        "cfg": {
179          "lr": 0.01,
180          "weight_decay": 0.0001
181        },
182        "mean": 0.0017732195847202092,
183        "std": 0.0004718425245521136,
184        "per_seed": [
185          0.001313930144533515,
186          0.002031040843576193,
187          0.0011662384495139122,
188          0.002488275757059455,
189          0.0022803107276558876,
190          0.0012248011771589518,
191          0.002009331015869975,
192          0.0016718285623937845
193        ]
194      }
195    ]
196  },
197  "comparison": {
198    "delta_mean": 0.0014583070642402163,
199    "idea_wins": 0,
200    "n_pairs": 8,
201    "per_seed_diffs": [
202      0.0007745490875095129,
203      0.001575337431859225,
204      0.001148893526988104,
205      0.0021302557724993676,
206      0.001998234147322364,
207      0.0009148511162493378,
208      0.0019556475162971765,
209      0.0011686879151966423
210    ],
211    "p_value": 0.0081,
212    "mde": 0.0004409319997531043,
213    "mde_rel_pct": 148.98321936289327,
214    "verdict": "idea worse (significant)",
215    "system_worked": false
216  },
217  "mechanism_signature": {
218    "basis_size": 25,
219    "trained_model_observations": [
220      {
221        "seed": 0,
222        "metric": 0.0012427795445546508,
223        "mean_coeff_norm": 0.28018829226493835,
224        "max_coeff_norm": 0.6576406955718994,
225        "mass_error": 2.384185791015625e-07
226      },
227      {
228        "seed": 1,
229        "metric": 0.001973426202312112,
230        "mean_coeff_norm": 0.30172738432884216,
231        "max_coeff_norm": 0.6713038086891174,
232        "mass_error": 2.384185791015625e-07
233      },
234      {
235        "seed": 2,
236        "metric": 0.0013205275172367692,
237        "mean_coeff_norm": 0.31470319628715515,
238        "max_coeff_norm": 0.7678529024124146,
239        "mass_error": 1.7881393432617188e-07
240      },
241      {
242        "seed": 3,
243        "metric": 0.0024030047934502363,
244        "mean_coeff_norm": 0.3005197048187256,
245        "max_coeff_norm": 0.6498276591300964,
246        "mass_error": 2.384185791015625e-07
247      },
248      {
249        "seed": 4,
250        "metric": 0.002149099251255393,
251        "mean_coeff_norm": 0.304610013961792,
252        "max_coeff_norm": 0.6628603935241699,
253        "mass_error": 2.384185791015625e-07
254      },
255      {
256        "seed": 5,
257        "metric": 0.0012187783140689135,
258        "mean_coeff_norm": 0.28839680552482605,
259        "max_coeff_norm": 0.5926362872123718,
260        "mass_error": 1.7881393432617188e-07
261      },
262      {
263        "seed": 6,
264        "metric": 0.0020814703311771154,
265        "mean_coeff_norm": 0.30479809641838074,
266        "max_coeff_norm": 0.6872127056121826,
267        "mass_error": 2.384185791015625e-07
268      },
269      {
270        "seed": 7,
271        "metric": 0.0016450573457404971,
272        "mean_coeff_norm": 0.288260281085968,
273        "max_coeff_norm": 0.6056498885154724,
274        "mass_error": 1.7881393432617188e-07
275      }
276    ],
277    "predicted": "nonnegative unit-mass coefficients remain bounded",
278    "observed_mean_mass_error": 2.1606683731079102e-07,
279    "observed_max_coeff_norm": 0.7678529024124146,
280    "confirmed": true,
281    "math_sanity": {
282      "mass_error": 2.220446049250313e-16,
283      "rho": 1.0000000000000004,
284      "observed_log_norm_slope": -3.223008136576186e-16,
285      "predicted_log_rho": 4.440892098500625e-16,
286      "slope_abs_error": 7.663900235076811e-16
287    }
288  },
289  "protocol_notes": {
290    "n_train": 800,
291    "n_test": 400,
292    "epochs": 12,
293    "baseline_and_idea_lr_union": [
294      0.001,
295      0.003,
296      0.01
297    ],
298    "paired_seeds": [
299      0,
300      1,
301      2,
302      3,
303      4,
304      5,
305      6,
306      7
307    ],
308    "architecture": "same 64-unit GRU encoder; scalar linear head vs PF RBF expectation head"
309  }
310}