Phase-Aware Bias-Energy Trust Region / bench_report.json
Beats tuned baseline
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": 0.006,
10 "tau": 0.5
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "tau": 0.5
17 },
18 "mean": 50.39839839935303
19 },
20 {
21 "cfg": {
22 "lr": 0.001,
23 "tau": 1.0
24 },
25 "mean": 50.397457122802734
26 },
27 {
28 "cfg": {
29 "lr": 0.001,
30 "tau": 2.0
31 },
32 "mean": 50.39675712585449
33 },
34 {
35 "cfg": {
36 "lr": 0.002,
37 "tau": 0.5
38 },
39 "mean": 14.087330341339111
40 },
41 {
42 "cfg": {
43 "lr": 0.002,
44 "tau": 1.0
45 },
46 "mean": 14.09030532836914
47 },
48 {
49 "cfg": {
50 "lr": 0.002,
51 "tau": 2.0
52 },
53 "mean": 14.094738960266113
54 },
55 {
56 "cfg": {
57 "lr": 0.003,
58 "tau": 0.5
59 },
60 "mean": 10.800803184509277
61 },
62 {
63 "cfg": {
64 "lr": 0.003,
65 "tau": 1.0
66 },
67 "mean": 10.79708456993103
68 },
69 {
70 "cfg": {
71 "lr": 0.003,
72 "tau": 2.0
73 },
74 "mean": 10.818211793899536
75 },
76 {
77 "cfg": {
78 "lr": 0.004,
79 "tau": 0.5
80 },
81 "mean": 8.752828359603882
82 },
83 {
84 "cfg": {
85 "lr": 0.004,
86 "tau": 1.0
87 },
88 "mean": 8.755794763565063
89 },
90 {
91 "cfg": {
92 "lr": 0.004,
93 "tau": 2.0
94 },
95 "mean": 8.77496600151062
96 },
97 {
98 "cfg": {
99 "lr": 0.006,
100 "tau": 0.5
101 },
102 "mean": 7.511728048324585
103 },
104 {
105 "cfg": {
106 "lr": 0.006,
107 "tau": 1.0
108 },
109 "mean": 7.649782061576843
110 },
111 {
112 "cfg": {
113 "lr": 0.006,
114 "tau": 2.0
115 },
116 "mean": 7.626705288887024
117 }
118 ],
119 "full": {
120 "mean": 8.051172316074371,
121 "std": 1.1549823103901204,
122 "per_seed": [
123 7.279676914215088,
124 6.340634822845459,
125 7.740776062011719,
126 8.685824394226074,
127 9.63039493560791,
128 6.726284503936768,
129 8.459394454956055,
130 9.546392440795898
131 ],
132 "n": 8
133 }
134 },
135 "idea": {
136 "config": {
137 "lr": 0.006,
138 "tau0": 0.5,
139 "eta": 0.09
140 },
141 "mean": 7.376198410987854,
142 "std": 0.8078643252957813,
143 "per_seed": [
144 6.912933349609375,
145 5.868978977203369,
146 8.128691673278809,
147 8.1602783203125,
148 7.357755661010742,
149 6.566903114318848,
150 8.233548164367676,
151 7.780498027801514
152 ],
153 "n": 8,
154 "diagnostics": [
155 {
156 "metric": 6.912933349609375,
157 "update_norm_mean": 18.64707014958064,
158 "clip_fraction": 0.62500000724362,
159 "residual": 15.457989124257843,
160 "energy": 2.6955993475714886,
161 "final_tau": 10.0,
162 "phase": "energy"
163 },
164 {
165 "metric": 5.868978977203369,
166 "update_norm_mean": 15.81644619256258,
167 "clip_fraction": 0.5625000053809749,
168 "residual": 13.15220600542509,
169 "energy": 2.135818994000646,
170 "final_tau": 10.0,
171 "phase": "energy"
172 },
173 {
174 "metric": 8.128691673278809,
175 "update_norm_mean": 14.61398777945174,
176 "clip_fraction": 0.5555555599017276,
177 "residual": 12.13080305696672,
178 "energy": 1.9603608226717673,
179 "final_tau": 10.0,
180 "phase": "energy"
181 },
182 {
183 "metric": 8.1602783203125,
184 "update_norm_mean": 18.281507012744743,
185 "clip_fraction": 0.6157407452248864,
186 "residual": 15.370208648343882,
187 "energy": 2.409127195988459,
188 "final_tau": 10.0,
189 "phase": "energy"
190 },
191 {
192 "metric": 7.357755661010742,
193 "update_norm_mean": 16.17437604152494,
194 "clip_fraction": 0.5902777841935555,
195 "residual": 13.253949597995314,
196 "energy": 2.4005918734516047,
197 "final_tau": 10.0,
198 "phase": "energy"
199 },
200 {
201 "metric": 6.566903114318848,
202 "update_norm_mean": 19.33727982143561,
203 "clip_fraction": 0.650462968274951,
204 "residual": 16.040926492462557,
205 "energy": 2.8265877289300914,
206 "final_tau": 10.0,
207 "phase": "energy"
208 },
209 {
210 "metric": 8.233548164367676,
211 "update_norm_mean": 18.142026902900803,
212 "clip_fraction": 0.5925925980425544,
213 "residual": 15.166145478220036,
214 "energy": 2.4767839011183406,
215 "final_tau": 10.0,
216 "phase": "energy"
217 },
218 {
219 "metric": 7.780498027801514,
220 "update_norm_mean": 16.652106201483143,
221 "clip_fraction": 0.5717592653301027,
222 "residual": 13.809878197809061,
223 "energy": 2.3481363373500703,
224 "final_tau": 10.0,
225 "phase": "energy"
226 }
227 ]
228 },
229 "comparison": {
230 "delta_mean": -0.6749739050865173,
231 "idea_wins": 7,
232 "n_pairs": 8,
233 "per_seed_diffs": [
234 -0.3667435646057129,
235 -0.47165584564208984,
236 0.38791561126708984,
237 -0.5255460739135742,
238 -2.272639274597168,
239 -0.15938138961791992,
240 -0.2258462905883789,
241 -1.7658944129943848
242 ],
243 "p_value": 0.04815,
244 "mde": 0.7410767906092078,
245 "mde_rel_pct": 9.204582407578448,
246 "verdict": "idea better (significant)",
247 "system_worked": true
248 },
249 "mechanism_signature": {
250 "prediction": "alpha <= p*beta selects energy control; empirical energy approaches target",
251 "alpha": 1.0,
252 "beta": 1.0,
253 "p": 2.0,
254 "predicted_phase": "energy",
255 "observed_phase": "energy",
256 "observed_energy_mean": 2.4066257751353084,
257 "energy_target": 0.22,
258 "observed_residual_mean": 14.297763325185063,
259 "observed_final_tau_mean": 10.0,
260 "confirmed": false
261 }
262}