Action-calibrated cycle-hopping RNN / 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": 0.01,
10 "weight_decay": 0.0001
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "weight_decay": 0.0
17 },
18 "mean": 0.004103701037820429
19 },
20 {
21 "cfg": {
22 "lr": 0.001,
23 "weight_decay": 0.0001
24 },
25 "mean": 0.0041049906285479665
26 },
27 {
28 "cfg": {
29 "lr": 0.003,
30 "weight_decay": 0.0
31 },
32 "mean": 0.0033861721749417484
33 },
34 {
35 "cfg": {
36 "lr": 0.003,
37 "weight_decay": 0.0001
38 },
39 "mean": 0.0033592699037399143
40 },
41 {
42 "cfg": {
43 "lr": 0.01,
44 "weight_decay": 0.0
45 },
46 "mean": 0.0014241317112464458
47 },
48 {
49 "cfg": {
50 "lr": 0.01,
51 "weight_decay": 0.0001
52 },
53 "mean": 0.0014055168576305732
54 }
55 ],
56 "full": {
57 "mean": 0.0013874754731659777,
58 "std": 0.0003779629342121644,
59 "per_seed": [
60 0.0008414968033321202,
61 0.0018421959830448031,
62 0.0010265976889058948,
63 0.0019117769552394748,
64 0.0010145766427740455,
65 0.0013230722397565842,
66 0.0016772722592577338,
67 0.0014628152130171657
68 ],
69 "n": 8
70 }
71 },
72 "idea": {
73 "mean": 0.0014499883909593336,
74 "std": 0.0006258243181704281,
75 "per_seed": [
76 0.00110855873208493,
77 0.0011564278975129128,
78 0.0009631213033571839,
79 0.002061811974272132,
80 0.0009511980460956693,
81 0.0017834099708124995,
82 0.0008573447703383863,
83 0.0027180344332009554
84 ],
85 "n": 8
86 },
87 "comparison": {
88 "delta_mean": 6.251291779335588e-05,
89 "idea_wins": 4,
90 "n_pairs": 8,
91 "per_seed_diffs": [
92 0.00026706192875280976,
93 -0.0006857680855318904,
94 -6.347638554871082e-05,
95 0.00015003501903265715,
96 -6.33785966783762e-05,
97 0.00046033773105591536,
98 -0.0008199274889193475,
99 0.0012552192201837897
100 ],
101 "p_value": 0.82325,
102 "mde": 0.000547451419528248,
103 "mde_rel_pct": 39.45665563940089,
104 "verdict": "no significant win",
105 "system_worked": false
106 },
107 "mechanism_signature": {
108 "prediction": "cycle-stabilizing intervention lowers local recurrent hidden-state gain",
109 "baseline_hidden_gain": 0.5601232647895813,
110 "idea_hidden_gain": 0.5631288886070251,
111 "relative_gain_change": 0.005366004150734417,
112 "confirmed": false,
113 "measurement": "finite 1e-3 hidden perturbation through trained GRU final-step transition"
114 },
115 "idea_sweep": [
116 {
117 "cfg": {
118 "lr": 0.001,
119 "weight_decay": 0.0001,
120 "lambda": 0.02
121 },
122 "result": {
123 "mean": 0.0048043747956398875,
124 "std": 0.0015046526831977705,
125 "per_seed": [
126 0.007243611849844456,
127 0.0027444041334092617,
128 0.004059470724314451,
129 0.003256669035181403,
130 0.004964617546647787,
131 0.005492699332535267,
132 0.003940005321055651,
133 0.006733520422130823
134 ],
135 "n": 8
136 }
137 },
138 {
139 "cfg": {
140 "lr": 0.003,
141 "weight_decay": 0.0001,
142 "lambda": 0.02
143 },
144 "result": {
145 "mean": 0.0029061136301606894,
146 "std": 0.0006765360005780435,
147 "per_seed": [
148 0.0028327745385468006,
149 0.0025881670881062746,
150 0.0032603004947304726,
151 0.004262231756001711,
152 0.002710104687139392,
153 0.003395120846107602,
154 0.0021831300109624863,
155 0.002017079619690776
156 ],
157 "n": 8
158 }
159 },
160 {
161 "cfg": {
162 "lr": 0.01,
163 "weight_decay": 0.0001,
164 "lambda": 0.02
165 },
166 "result": {
167 "mean": 0.0014499883909593336,
168 "std": 0.0006258243181704281,
169 "per_seed": [
170 0.00110855873208493,
171 0.0011564278975129128,
172 0.0009631213033571839,
173 0.002061811974272132,
174 0.0009511980460956693,
175 0.0017834099708124995,
176 0.0008573447703383863,
177 0.0027180344332009554
178 ],
179 "n": 8
180 }
181 }
182 ],
183 "protocol_notes": {
184 "epochs": 12,
185 "n_train": 400,
186 "n_test": 400,
187 "baseline_grid": [
188 {
189 "lr": 0.001,
190 "weight_decay": 0.0
191 },
192 {
193 "lr": 0.001,
194 "weight_decay": 0.0001
195 },
196 {
197 "lr": 0.003,
198 "weight_decay": 0.0
199 },
200 {
201 "lr": 0.003,
202 "weight_decay": 0.0001
203 },
204 {
205 "lr": 0.01,
206 "weight_decay": 0.0
207 },
208 {
209 "lr": 0.01,
210 "weight_decay": 0.0001
211 }
212 ],
213 "idea_grid": [
214 {
215 "lr": 0.001,
216 "weight_decay": 0.0001,
217 "lambda": 0.02
218 },
219 {
220 "lr": 0.003,
221 "weight_decay": 0.0001,
222 "lambda": 0.02
223 },
224 {
225 "lr": 0.01,
226 "weight_decay": 0.0001,
227 "lambda": 0.02
228 }
229 ]
230 }
231}