Phase-Polytope Robust Neural Dynamics / bench_report.json
Failed on benchmark
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": 0.006,
10 "weight_decay": 0.0
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "weight_decay": 0.0
17 },
18 "mean": 0.0005779846687801182
19 },
20 {
21 "cfg": {
22 "lr": 0.003,
23 "weight_decay": 0.0
24 },
25 "mean": 7.249591180880088e-05
26 },
27 {
28 "cfg": {
29 "lr": 0.006,
30 "weight_decay": 0.0
31 },
32 "mean": 3.491968436719617e-05
33 }
34 ],
35 "full": {
36 "mean": 3.7997122717570164e-05,
37 "std": 8.625243996335504e-06,
38 "per_seed": [
39 3.5259286960354075e-05,
40 3.6509172787191346e-05,
41 2.411565583315678e-05,
42 4.3794621888082474e-05,
43 5.490284820552915e-05,
44 4.165745485806838e-05,
45 3.027328239113558e-05,
46 3.746465881704353e-05
47 ],
48 "n": 8,
49 "behavior_signature": {
50 "centroid_mse": 3.7997122717570164e-05,
51 "mean_vertex_mse": 0.012571478087920696,
52 "max_vertex_mse": 0.02287934930063784,
53 "vertex_spread_mse": 0.01890919718425721,
54 "prediction_spread": 0.012533481465652585,
55 "residual_centroid_variance": 3.736908956852858e-05,
56 "mean_head_residual_variance": 0.00790156953735277,
57 "independent_1_over_M_predicted_variance": 0.0019753923843381926,
58 "observed_to_independent_variance_ratio": 0.02923531417036429
59 }
60 }
61 },
62 "idea": {
63 "mean": 3.7003671650381875e-05,
64 "std": 6.8240779328978215e-06,
65 "per_seed": [
66 3.23551103065256e-05,
67 3.801030106842518e-05,
68 2.430881249892991e-05,
69 3.9502079744124785e-05,
70 4.615575380739756e-05,
71 4.588229421642609e-05,
72 3.254720650147647e-05,
73 3.726781505974941e-05
74 ],
75 "n": 8,
76 "behavior_signature": {
77 "centroid_mse": 3.7003671650381875e-05,
78 "mean_vertex_mse": 4.852679876421462e-05,
79 "max_vertex_mse": 5.404685998655623e-05,
80 "vertex_spread_mse": 1.0225702226307476e-05,
81 "prediction_spread": 1.1523133878199587e-05,
82 "residual_centroid_variance": 3.620897632572451e-05,
83 "mean_head_residual_variance": 4.694555445894366e-05,
84 "independent_1_over_M_predicted_variance": 1.1736388614735915e-05,
85 "observed_to_independent_variance_ratio": 3.1302984952926636
86 }
87 },
88 "comparison": {
89 "delta_mean": -9.93451067188289e-07,
90 "idea_wins": 4,
91 "n_pairs": 8,
92 "per_seed_diffs": [
93 -2.904176653828472e-06,
94 1.5011282812338322e-06,
95 1.9315666577313095e-07,
96 -4.292542143957689e-06,
97 -8.747094398131594e-06,
98 4.224839358357713e-06,
99 2.273924110340886e-06,
100 -1.968437572941184e-07
101 ],
102 "p_value": 0.5281,
103 "mde": 3.473895858258558e-06,
104 "mde_rel_pct": 9.142523459157085,
105 "verdict": "no significant win",
106 "system_worked": false
107 },
108 "idea_grid": [
109 {
110 "cfg": {
111 "lr": 0.001,
112 "weight_decay": 0.0,
113 "lambda_rob": 0.5
114 },
115 "result": {
116 "mean": 0.0006954554264666513,
117 "std": 0.00011255634752562823,
118 "per_seed": [
119 0.0007669797050766647,
120 0.0007185133290477097,
121 0.0006170952692627907,
122 0.0008686299552209675,
123 0.0006663062958978117,
124 0.000580831547267735,
125 0.0008218838484026492,
126 0.0005234034615568817
127 ],
128 "n": 8,
129 "behavior_signature": {
130 "centroid_mse": 0.0006954554264666513,
131 "mean_vertex_mse": 0.0007360787494690157,
132 "max_vertex_mse": 0.0007692473591305315,
133 "vertex_spread_mse": 7.699854904785752e-05,
134 "prediction_spread": 4.0623296627018135e-05,
135 "residual_centroid_variance": 0.0006928898219484836,
136 "mean_head_residual_variance": 0.0007312777961487882,
137 "independent_1_over_M_predicted_variance": 0.00018281944903719705,
138 "observed_to_independent_variance_ratio": 3.783972203731537
139 }
140 }
141 },
142 {
143 "cfg": {
144 "lr": 0.003,
145 "weight_decay": 0.0,
146 "lambda_rob": 0.5
147 },
148 "result": {
149 "mean": 7.912012461019913e-05,
150 "std": 1.5950926774965125e-05,
151 "per_seed": [
152 6.69353612465784e-05,
153 0.00011281413026154041,
154 5.5623815569560975e-05,
155 8.786085527390242e-05,
156 6.82869358570315e-05,
157 8.024623821256682e-05,
158 8.025208080653101e-05,
159 8.094157965388149e-05
160 ],
161 "n": 8,
162 "behavior_signature": {
163 "centroid_mse": 7.912012461019913e-05,
164 "mean_vertex_mse": 9.344643240183359e-05,
165 "max_vertex_mse": 0.00010036833737103734,
166 "vertex_spread_mse": 1.4143542102829088e-05,
167 "prediction_spread": 1.4326290056487778e-05,
168 "residual_centroid_variance": 7.598952061016462e-05,
169 "mean_head_residual_variance": 8.980537677416578e-05,
170 "independent_1_over_M_predicted_variance": 2.2451344193541445e-05,
171 "observed_to_independent_variance_ratio": 3.3957497477531433
172 }
173 }
174 },
175 {
176 "cfg": {
177 "lr": 0.006,
178 "weight_decay": 0.0,
179 "lambda_rob": 0.5
180 },
181 "result": {
182 "mean": 3.7003671650381875e-05,
183 "std": 6.8240779328978215e-06,
184 "per_seed": [
185 3.23551103065256e-05,
186 3.801030106842518e-05,
187 2.430881249892991e-05,
188 3.9502079744124785e-05,
189 4.615575380739756e-05,
190 4.588229421642609e-05,
191 3.254720650147647e-05,
192 3.726781505974941e-05
193 ],
194 "n": 8,
195 "behavior_signature": {
196 "centroid_mse": 3.7003671650381875e-05,
197 "mean_vertex_mse": 4.852679876421462e-05,
198 "max_vertex_mse": 5.404685998655623e-05,
199 "vertex_spread_mse": 1.0225702226307476e-05,
200 "prediction_spread": 1.1523133878199587e-05,
201 "residual_centroid_variance": 3.620897632572451e-05,
202 "mean_head_residual_variance": 4.694555445894366e-05,
203 "independent_1_over_M_predicted_variance": 1.1736388614735915e-05,
204 "observed_to_independent_variance_ratio": 3.1302984952926636
205 }
206 }
207 }
208 ],
209 "protocol_notes": "Matched dynamics track; same GRU and four heads; only robust loss differs.",
210 "mechanism_signature": {
211 "prediction": "independent phase errors imply centroid variance sigma^2/M",
212 "M": 4,
213 "variance_check": [
214 {
215 "M": 1,
216 "observed": 2.8750873157106502,
217 "predicted": 2.8899999999999997,
218 "ratio": 0.9948399016299829
219 },
220 {
221 "M": 2,
222 "observed": 1.442905932122292,
223 "predicted": 1.4449999999999998,
224 "ratio": 0.9985508180777107
225 },
226 {
227 "M": 4,
228 "observed": 0.7237846720794168,
229 "predicted": 0.7224999999999999,
230 "ratio": 1.0017780928434836
231 },
232 {
233 "M": 8,
234 "observed": 0.3604275026112933,
235 "predicted": 0.36124999999999996,
236 "ratio": 0.9977231906194971
237 },
238 {
239 "M": 16,
240 "observed": 0.18043369597547337,
241 "predicted": 0.18062499999999998,
242 "ratio": 0.9989408773728631
243 }
244 ],
245 "observed_to_independent_variance_ratio": 3.1302984952926636,
246 "confirmed": false,
247 "trained_model_behavior": {
248 "centroid_mse": 3.7003671650381875e-05,
249 "mean_vertex_mse": 4.852679876421462e-05,
250 "max_vertex_mse": 5.404685998655623e-05,
251 "vertex_spread_mse": 1.0225702226307476e-05,
252 "prediction_spread": 1.1523133878199587e-05,
253 "residual_centroid_variance": 3.620897632572451e-05,
254 "mean_head_residual_variance": 4.694555445894366e-05,
255 "independent_1_over_M_predicted_variance": 1.1736388614735915e-05,
256 "observed_to_independent_variance_ratio": 3.1302984952926636
257 }
258 },
259 "stage1_math_check": [
260 {
261 "M": 1,
262 "observed": 2.8750873157106502,
263 "predicted": 2.8899999999999997,
264 "ratio": 0.9948399016299829
265 },
266 {
267 "M": 2,
268 "observed": 1.442905932122292,
269 "predicted": 1.4449999999999998,
270 "ratio": 0.9985508180777107
271 },
272 {
273 "M": 4,
274 "observed": 0.7237846720794168,
275 "predicted": 0.7224999999999999,
276 "ratio": 1.0017780928434836
277 },
278 {
279 "M": 8,
280 "observed": 0.3604275026112933,
281 "predicted": 0.36124999999999996,
282 "ratio": 0.9977231906194971
283 },
284 {
285 "M": 16,
286 "observed": 0.18043369597547337,
287 "predicted": 0.18062499999999998,
288 "ratio": 0.9989408773728631
289 }
290 ]
291}