Conservative Parallel-Edge Decomposition / bench_report.json
Mechanism confirmed, baseline not beaten
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 "epochs": 24
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.0015,
16 "epochs": 24
17 },
18 "mean": 0.0027078510902356356
19 },
20 {
21 "cfg": {
22 "lr": 0.003,
23 "epochs": 24
24 },
25 "mean": 0.002131043642293662
26 },
27 {
28 "cfg": {
29 "lr": 0.006,
30 "epochs": 24
31 },
32 "mean": 0.0017603125161258504
33 }
34 ],
35 "full": {
36 "mean": 0.0017603125161258504,
37 "std": 0.0003135227734284653,
38 "per_seed": [
39 0.001380653353407979,
40 0.0020586533937603235,
41 0.00220737443305552,
42 0.0021458740811794996,
43 0.0016572552267462015,
44 0.00137868314050138,
45 0.0015466997865587473,
46 0.001707306713797152
47 ],
48 "n": 8
49 }
50 },
51 "idea": {
52 "mean": 0.0019414099806454033,
53 "std": 0.000537525437857345,
54 "per_seed": [
55 0.001175302779302001,
56 0.0021009868942201138,
57 0.0018062262097373605,
58 0.0030378331430256367,
59 0.0023114001378417015,
60 0.001854351139627397,
61 0.0013667570892721415,
62 0.0018784224521368742
63 ],
64 "n": 8,
65 "selected_cfg": {
66 "lr": 0.006,
67 "epochs": 24
68 }
69 },
70 "comparison": {
71 "delta_mean": 0.0001810974645195529,
72 "idea_wins": 3,
73 "n_pairs": 8,
74 "per_seed_diffs": [
75 -0.000205350574105978,
76 4.233350045979023e-05,
77 -0.0004011482233181596,
78 0.000891959061846137,
79 0.0006541449110955,
80 0.0004756679991260171,
81 -0.00017994269728660583,
82 0.00017111573833972216
83 ],
84 "p_value": 0.2882,
85 "mde": 0.00038107323489683164,
86 "mde_rel_pct": 21.648044390180743,
87 "verdict": "no significant win",
88 "system_worked": false
89 },
90 "mechanism_signature": {
91 "quantity": "trained-model mean absolute 1^T B P",
92 "predicted": "zero internal net exchange",
93 "observed_max": 0.0,
94 "observed_mean": 0.0,
95 "per_seed": [
96 {
97 "seed": 0,
98 "mean_abs_internal_residual": 0.0,
99 "mean_abs_predicted_output": 0.7605881690979004
100 },
101 {
102 "seed": 1,
103 "mean_abs_internal_residual": 0.0,
104 "mean_abs_predicted_output": 0.7429919838905334
105 },
106 {
107 "seed": 2,
108 "mean_abs_internal_residual": 0.0,
109 "mean_abs_predicted_output": 0.7579967379570007
110 },
111 {
112 "seed": 3,
113 "mean_abs_internal_residual": 0.0,
114 "mean_abs_predicted_output": 0.8042439818382263
115 },
116 {
117 "seed": 4,
118 "mean_abs_internal_residual": 0.0,
119 "mean_abs_predicted_output": 0.7317724227905273
120 },
121 {
122 "seed": 5,
123 "mean_abs_internal_residual": 0.0,
124 "mean_abs_predicted_output": 0.7670498490333557
125 },
126 {
127 "seed": 6,
128 "mean_abs_internal_residual": 0.0,
129 "mean_abs_predicted_output": 0.756341278553009
130 },
131 {
132 "seed": 7,
133 "mean_abs_internal_residual": 0.0,
134 "mean_abs_predicted_output": 0.6808968782424927
135 }
136 ],
137 "confirmed": true
138 },
139 "protocol": {
140 "paired_seeds": [
141 0,
142 1,
143 2,
144 3,
145 4,
146 5,
147 6,
148 7
149 ],
150 "grid": [
151 {
152 "lr": 0.0015,
153 "epochs": 24
154 },
155 {
156 "lr": 0.003,
157 "epochs": 24
158 },
159 {
160 "lr": 0.006,
161 "epochs": 24
162 }
163 ],
164 "baseline_best_cfg": {
165 "lr": 0.006,
166 "epochs": 24
167 },
168 "idea_grid_results": [
169 {
170 "cfg": {
171 "lr": 0.0015,
172 "epochs": 24
173 },
174 "result": {
175 "mean": 0.003114120219834149,
176 "std": 0.0007247547558258706,
177 "per_seed": [
178 0.002160337520763278,
179 0.003606663551181555,
180 0.003154655685648322,
181 0.003743839683011174,
182 0.0032033741008490324,
183 0.0022424152120947838,
184 0.002473633736371994,
185 0.004328042268753052
186 ],
187 "n": 8
188 }
189 },
190 {
191 "cfg": {
192 "lr": 0.003,
193 "epochs": 24
194 },
195 "result": {
196 "mean": 0.002424486941890791,
197 "std": 0.0005536617718879092,
198 "per_seed": [
199 0.0017574743833392859,
200 0.0028110698331147432,
201 0.00233089760877192,
202 0.0031846617348492146,
203 0.002762784017249942,
204 0.0018244534730911255,
205 0.0017341335769742727,
206 0.0029904209077358246
207 ],
208 "n": 8
209 }
210 },
211 {
212 "cfg": {
213 "lr": 0.006,
214 "epochs": 24
215 },
216 "result": {
217 "mean": 0.0019414099806454033,
218 "std": 0.000537525437857345,
219 "per_seed": [
220 0.001175302779302001,
221 0.0021009868942201138,
222 0.0018062262097373605,
223 0.0030378331430256367,
224 0.0023114001378417015,
225 0.001854351139627397,
226 0.0013667570892721415,
227 0.0018784224521368742
228 ],
229 "n": 8,
230 "selected_cfg": {
231 "lr": 0.006,
232 "epochs": 24
233 }
234 }
235 }
236 ]
237 }
238}