Work-trained neural Hamiltonian bridge / 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": 0.01,
10 "weight_decay": 0.0001
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "weight_decay": 0.0
17 },
18 "mean": 0.004298779065720737
19 },
20 {
21 "cfg": {
22 "lr": 0.001,
23 "weight_decay": 0.0001
24 },
25 "mean": 0.0042995543335564435
26 },
27 {
28 "cfg": {
29 "lr": 0.003,
30 "weight_decay": 0.0
31 },
32 "mean": 0.0034615940821822733
33 },
34 {
35 "cfg": {
36 "lr": 0.003,
37 "weight_decay": 0.0001
38 },
39 "mean": 0.003432174737099558
40 },
41 {
42 "cfg": {
43 "lr": 0.01,
44 "weight_decay": 0.0
45 },
46 "mean": 0.0014865802077110857
47 },
48 {
49 "cfg": {
50 "lr": 0.01,
51 "weight_decay": 0.0001
52 },
53 "mean": 0.0014641905727330595
54 }
55 ],
56 "full": {
57 "mean": 0.0014106978487689048,
58 "std": 0.00038184104551606325,
59 "per_seed": [
60 0.0009313806658610702,
61 0.0018688973505049944,
62 0.0010692982468754053,
63 0.0019871860276907682,
64 0.0009491611272096634,
65 0.0014028067234903574,
66 0.0016651722835376859,
67 0.0014116803649812937
68 ],
69 "n": 8
70 }
71 },
72 "idea": {
73 "mean": 0.0014106978487689048,
74 "std": 0.00038184104551606325,
75 "per_seed": [
76 0.0009313806658610702,
77 0.0018688973505049944,
78 0.0010692982468754053,
79 0.0019871860276907682,
80 0.0009491611272096634,
81 0.0014028067234903574,
82 0.0016651722835376859,
83 0.0014116803649812937
84 ],
85 "n": 8
86 },
87 "comparison": {
88 "delta_mean": 0.0,
89 "idea_wins": 0,
90 "n_pairs": 8,
91 "per_seed_diffs": [
92 0.0,
93 0.0,
94 0.0,
95 0.0,
96 0.0,
97 0.0,
98 0.0,
99 0.0
100 ],
101 "p_value": 1.0,
102 "mde": 0.0,
103 "mde_rel_pct": 0.0,
104 "verdict": "no measurable effect",
105 "system_worked": false
106 },
107 "mechanism_signature": {
108 "definition": "trained-model endpoint work proxy vs endpoint squared error",
109 "mean_work_proxy": -0.09426483511924744,
110 "std_work_proxy": 0.24112743139266968,
111 "mean_test_mse": 0.0009313806076534092,
112 "corr_work_error": -0.5097475039344403,
113 "predicted_direction": "work penalty should reduce endpoint energy change",
114 "observed_direction": "reduced",
115 "confirmed": true
116 },
117 "idea_sweep": [
118 {
119 "cfg": {
120 "lr": 0.01,
121 "weight_decay": 0.0001,
122 "lambda": 0.0
123 },
124 "result": {
125 "mean": 0.0014106978487689048,
126 "std": 0.00038184104551606325,
127 "per_seed": [
128 0.0009313806658610702,
129 0.0018688973505049944,
130 0.0010692982468754053,
131 0.0019871860276907682,
132 0.0009491611272096634,
133 0.0014028067234903574,
134 0.0016651722835376859,
135 0.0014116803649812937
136 ],
137 "n": 8
138 }
139 },
140 {
141 "cfg": {
142 "lr": 0.01,
143 "weight_decay": 0.0001,
144 "lambda": 0.01
145 },
146 "result": {
147 "mean": 0.0022155136612127535,
148 "std": 0.0009111505580800969,
149 "per_seed": [
150 0.0026169263292104006,
151 0.0008814591565169394,
152 0.002344637643545866,
153 0.001455318764783442,
154 0.0015512269455939531,
155 0.003643145551905036,
156 0.0034239226952195168,
157 0.0018074722029268742
158 ],
159 "n": 8
160 }
161 },
162 {
163 "cfg": {
164 "lr": 0.01,
165 "weight_decay": 0.0001,
166 "lambda": 0.05
167 },
168 "result": {
169 "mean": 0.020211070426739752,
170 "std": 0.0041872810874073125,
171 "per_seed": [
172 0.0175981055945158,
173 0.014253576286137104,
174 0.024302082136273384,
175 0.020239518955349922,
176 0.014925491064786911,
177 0.025781074538826942,
178 0.02489893138408661,
179 0.019689783453941345
180 ],
181 "n": 8
182 }
183 },
184 {
185 "cfg": {
186 "lr": 0.001,
187 "weight_decay": 0.0001,
188 "lambda": 0.05
189 },
190 "result": {
191 "mean": 0.03044215077534318,
192 "std": 0.008291443683791245,
193 "per_seed": [
194 0.045904748141765594,
195 0.022312993183732033,
196 0.031132273375988007,
197 0.03332684561610222,
198 0.03957861661911011,
199 0.025154752656817436,
200 0.025864429771900177,
201 0.020262546837329865
202 ],
203 "n": 8
204 }
205 },
206 {
207 "cfg": {
208 "lr": 0.003,
209 "weight_decay": 0.0001,
210 "lambda": 0.05
211 },
212 "result": {
213 "mean": 0.031030800193548203,
214 "std": 0.008264055684726617,
215 "per_seed": [
216 0.030995534732937813,
217 0.030705397948622704,
218 0.023237835615873337,
219 0.049849774688482285,
220 0.0321778804063797,
221 0.034446511417627335,
222 0.022313375025987625,
223 0.024520091712474823
224 ],
225 "n": 8
226 }
227 }
228 ],
229 "protocol": {
230 "epochs": 12,
231 "n_train": 400,
232 "n_test": 200,
233 "batch": 128,
234 "paired_seeds": [
235 0,
236 1,
237 2,
238 3,
239 4,
240 5,
241 6,
242 7
243 ],
244 "structural_match": "dynamics/control"
245 }
246}