Local Characteristic Residual Gating / bench_report.json
Beats tuned baseline
1{
2 "bench_version": 1,
3 "track": "local_shock_characteristic",
4 "model": "mlp_tiny",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.01,
10 "alpha": 0.0
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "alpha": 0.0
17 },
18 "mean": 0.00868597999215126
19 },
20 {
21 "cfg": {
22 "lr": 0.003,
23 "alpha": 0.0
24 },
25 "mean": 0.00661642977502197
26 },
27 {
28 "cfg": {
29 "lr": 0.01,
30 "alpha": 0.0
31 },
32 "mean": 0.004176105372607708
33 }
34 ],
35 "full": {
36 "mean": 0.003825201274594292,
37 "std": 0.0005842251285976379,
38 "per_seed": [
39 0.0037032742984592915,
40 0.004018085543066263,
41 0.00398260960355401,
42 0.005000452045351267,
43 0.003473947523161769,
44 0.003305204212665558,
45 0.002951973583549261,
46 0.004166063386946917
47 ],
48 "n": 8
49 }
50 },
51 "idea": {
52 "mean": 0.0017611028160899878,
53 "std": 0.0002040852129581439,
54 "per_seed": [
55 0.0017537641106173396,
56 0.0017673540860414505,
57 0.0015747904544696212,
58 0.001804820611141622,
59 0.0014342450303956866,
60 0.0017936605727300048,
61 0.0017648270586505532,
62 0.002195360604673624
63 ],
64 "n": 8
65 },
66 "comparison": {
67 "delta_mean": -0.0020640984585043043,
68 "idea_wins": 8,
69 "n_pairs": 8,
70 "per_seed_diffs": [
71 -0.0019495101878419518,
72 -0.0022507314570248127,
73 -0.002407819149084389,
74 -0.003195631434209645,
75 -0.0020397024927660823,
76 -0.001511543639935553,
77 -0.0011871465248987079,
78 -0.0019707027822732925
79 ],
80 "p_value": 0.0081,
81 "mde": 0.0005027018540061575,
82 "mde_rel_pct": 13.141840596596449,
83 "verdict": "idea better (significant)",
84 "system_worked": true
85 },
86 "custom_track": {
87 "name": "local_shock_characteristic",
88 "file": "pde_local_track.py",
89 "domain": "pde"
90 },
91 "baseline_grid": [
92 {
93 "lr": 0.001,
94 "alpha": 0.0
95 },
96 {
97 "lr": 0.003,
98 "alpha": 0.0
99 },
100 {
101 "lr": 0.01,
102 "alpha": 0.0
103 }
104 ],
105 "idea_grid": [
106 {
107 "lr": 0.001,
108 "alpha": 1.0
109 },
110 {
111 "lr": 0.001,
112 "alpha": 2.0
113 },
114 {
115 "lr": 0.001,
116 "alpha": 4.0
117 },
118 {
119 "lr": 0.003,
120 "alpha": 1.0
121 },
122 {
123 "lr": 0.003,
124 "alpha": 2.0
125 },
126 {
127 "lr": 0.003,
128 "alpha": 4.0
129 },
130 {
131 "lr": 0.01,
132 "alpha": 1.0
133 },
134 {
135 "lr": 0.01,
136 "alpha": 2.0
137 },
138 {
139 "lr": 0.01,
140 "alpha": 4.0
141 }
142 ],
143 "idea_sweep": [
144 {
145 "cfg": {
146 "lr": 0.001,
147 "alpha": 1.0
148 },
149 "mean": 0.007479952648282051
150 },
151 {
152 "cfg": {
153 "lr": 0.001,
154 "alpha": 2.0
155 },
156 "mean": 0.007875820272602141
157 },
158 {
159 "cfg": {
160 "lr": 0.001,
161 "alpha": 4.0
162 },
163 "mean": 0.008559924317523837
164 },
165 {
166 "cfg": {
167 "lr": 0.003,
168 "alpha": 1.0
169 },
170 "mean": 0.003235555544961244
171 },
172 {
173 "cfg": {
174 "lr": 0.003,
175 "alpha": 2.0
176 },
177 "mean": 0.0036418562522158027
178 },
179 {
180 "cfg": {
181 "lr": 0.003,
182 "alpha": 4.0
183 },
184 "mean": 0.004329836694523692
185 },
186 {
187 "cfg": {
188 "lr": 0.01,
189 "alpha": 1.0
190 },
191 "mean": 0.0017251823155675083
192 },
193 {
194 "cfg": {
195 "lr": 0.01,
196 "alpha": 2.0
197 },
198 "mean": 0.0022461695480160415
199 },
200 {
201 "cfg": {
202 "lr": 0.01,
203 "alpha": 4.0
204 },
205 "mean": 0.002906196634285152
206 }
207 ],
208 "selected_idea_cfg": {
209 "lr": 0.01,
210 "alpha": 1.0
211 },
212 "mechanism_signature": {
213 "quantity": "trained-model input residual energy",
214 "predicted": "characteristic gating suppresses oscillatory residual energy",
215 "observed_raw_rms": 0.6787973418831825,
216 "observed_gated_rms": 0.03473339881747961,
217 "observed_suppression_fraction": 0.9488309740266987,
218 "confirmed": true
219 }
220}