Invariant-Sphere Recurrent State / 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": 0.0,
11 "epochs": 20
12 },
13 "sweep": [
14 {
15 "cfg": {
16 "lr": 0.001,
17 "weight_decay": 0.0,
18 "epochs": 20
19 },
20 "mean": 0.00010979082981066313
21 },
22 {
23 "cfg": {
24 "lr": 0.003,
25 "weight_decay": 0.0,
26 "epochs": 20
27 },
28 "mean": 1.23220872865204e-05
29 },
30 {
31 "cfg": {
32 "lr": 0.01,
33 "weight_decay": 0.0,
34 "epochs": 20
35 },
36 "mean": 5.326571056230023e-06
37 }
38 ],
39 "full": {
40 "mean": 6.485018303692414e-06,
41 "std": 3.0896387556205093e-06,
42 "per_seed": [
43 5.860124929313315e-06,
44 6.373980340868002e-06,
45 2.9345453640416963e-06,
46 6.137633590697078e-06,
47 7.82904498919379e-06,
48 4.8856131797947455e-06,
49 4.085992713953601e-06,
50 1.3773211321677081e-05
51 ],
52 "n": 8
53 }
54 },
55 "idea": {
56 "mean": 1.4861759154882748e-05,
57 "std": 7.4365681821899514e-06,
58 "per_seed": [
59 6.5364615693397354e-06,
60 1.7686747014522552e-05,
61 1.725437687127851e-05,
62 2.9500635719159618e-05,
63 1.0894022125285119e-05,
64 1.0834011845872737e-05,
65 5.783244432677748e-06,
66 2.0404573660925962e-05
67 ],
68 "n": 8
69 },
70 "comparison": {
71 "delta_mean": 8.376740851190334e-06,
72 "idea_wins": 0,
73 "n_pairs": 8,
74 "per_seed_diffs": [
75 6.763366400264204e-07,
76 1.131276667365455e-05,
77 1.4319831507236813e-05,
78 2.336300212846254e-05,
79 3.064977136091329e-06,
80 5.948398666077992e-06,
81 1.6972517187241465e-06,
82 6.631362339248881e-06
83 ],
84 "p_value": 0.0081,
85 "mde": 6.396115118271008e-06,
86 "mde_rel_pct": 98.62909892835944,
87 "verdict": "idea worse (significant)",
88 "system_worked": false
89 },
90 "mechanism_signature": {
91 "prediction": "trained final states have negative cubic radial coefficient and relax toward finite norm",
92 "trained_model_signature": {
93 "norm_mean": 0.7238094657659531,
94 "norm_std": 0.4153854250907898,
95 "radial_coeff_mean": -0.0551277962513268,
96 "radial_coeff_std": 0.006850266479887068,
97 "radial_velocity_mean": 0.6813768893480301,
98 "predicted_radius_scalar": 1.0,
99 "predicted_relaxation_rate": 2.0,
100 "radial_upper_bound": -0.015625,
101 "confirmed": true
102 },
103 "idea_trials": [
104 {
105 "cfg": {
106 "lr": 0.001,
107 "epochs": 20,
108 "lam": 1.0,
109 "dt": 0.05,
110 "cubic_scale": 1.0
111 },
112 "result": {
113 "mean": 3.4410899388603866e-05,
114 "std": 5.650175225500919e-06,
115 "per_seed": [
116 2.6207977498415858e-05,
117 2.940478952950798e-05,
118 3.0609477107645944e-05,
119 4.4194846850587055e-05,
120 4.1339764720760286e-05,
121 3.6304791137808934e-05,
122 3.362560528330505e-05,
123 3.3599942980799824e-05
124 ],
125 "n": 8
126 }
127 },
128 {
129 "cfg": {
130 "lr": 0.003,
131 "epochs": 20,
132 "lam": 1.0,
133 "dt": 0.05,
134 "cubic_scale": 1.0
135 },
136 "result": {
137 "mean": 1.609290268334007e-05,
138 "std": 4.353488098019092e-06,
139 "per_seed": [
140 1.1640918273769785e-05,
141 1.4350282071973197e-05,
142 1.364862600894412e-05,
143 2.4214104996644892e-05,
144 2.2511794668389484e-05,
145 1.5585039363941178e-05,
146 1.4041733265912626e-05,
147 1.2750722817145288e-05
148 ],
149 "n": 8
150 }
151 },
152 {
153 "cfg": {
154 "lr": 0.01,
155 "epochs": 20,
156 "lam": 1.0,
157 "dt": 0.05,
158 "cubic_scale": 1.0
159 },
160 "result": {
161 "mean": 1.4861759154882748e-05,
162 "std": 7.4365681821899514e-06,
163 "per_seed": [
164 6.5364615693397354e-06,
165 1.7686747014522552e-05,
166 1.725437687127851e-05,
167 2.9500635719159618e-05,
168 1.0894022125285119e-05,
169 1.0834011845872737e-05,
170 5.783244432677748e-06,
171 2.0404573660925962e-05
172 ],
173 "n": 8
174 }
175 }
176 ],
177 "custom_track": null
178 }
179}