Conjugate Bayesian Latent Dynamics Head / bench_report.json
Beats tuned baseline
1{
2 "bench_version": 1,
3 "track": "dynamics",
4 "model": "shared_gru_latent_linear_head",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.01
10 },
11 "sweep": [
12 {
13 "cfg": {
14 "lr": 0.001
15 },
16 "mean": 0.8660691231489182
17 },
18 {
19 "cfg": {
20 "lr": 0.003
21 },
22 "mean": 0.16844721138477325
23 },
24 {
25 "cfg": {
26 "lr": 0.01
27 },
28 "mean": 0.030930949840694666
29 }
30 ],
31 "full": {
32 "mean": 0.03257687599398196,
33 "std": 0.00921168354607454,
34 "per_seed": [
35 0.027142083272337914,
36 0.02880755066871643,
37 0.039878763258457184,
38 0.027895402163267136,
39 0.050688546150922775,
40 0.021972693502902985,
41 0.024402838200330734,
42 0.0398271307349205
43 ],
44 "n": 8
45 }
46 },
47 "idea": {
48 "mean": 0.0003928114747395739,
49 "std": 0.00021742121198405602,
50 "per_seed": [
51 8.355393947567791e-05,
52 0.0004895747406408191,
53 0.0001629032049095258,
54 0.0004371287068352103,
55 0.0004481615615077317,
56 0.0004002868663519621,
57 0.0008424268453381956,
58 0.0002784559328574687
59 ],
60 "n": 8
61 },
62 "comparison": {
63 "delta_mean": -0.032184064519242384,
64 "idea_wins": 8,
65 "n_pairs": 8,
66 "per_seed_diffs": [
67 -0.027058529332862236,
68 -0.02831797592807561,
69 -0.03971586005354766,
70 -0.027458273456431925,
71 -0.050240384589415044,
72 -0.021572406636551023,
73 -0.02356041135499254,
74 -0.03954867480206303
75 ],
76 "p_value": 0.0081,
77 "mde": 0.008286675296304802,
78 "mde_rel_pct": 25.437292691403652,
79 "verdict": "idea better (significant)",
80 "system_worked": true
81 },
82 "mechanism_signature": {
83 "idea_sweep": [
84 {
85 "cfg": {
86 "lr": 0.001
87 },
88 "mean": 0.00040330794945475645,
89 "full": {
90 "mean": 0.00040330794945475645,
91 "std": 0.0001870869897132185,
92 "per_seed": [
93 0.00025814544642344117,
94 0.00038167217280715704,
95 0.0006836550310254097,
96 0.0006370171322487295,
97 0.00021974655101075768,
98 0.00024331139866262674,
99 0.0005830578738823533,
100 0.00021985798957757652
101 ],
102 "n": 8
103 }
104 },
105 {
106 "cfg": {
107 "lr": 0.003
108 },
109 "mean": 0.0004725444596260786,
110 "full": {
111 "mean": 0.0004725444596260786,
112 "std": 0.00012726428121603695,
113 "per_seed": [
114 0.0003564116486813873,
115 0.000323356653098017,
116 0.0004201834963168949,
117 0.0007025530212558806,
118 0.00039990260847844183,
119 0.0004050004936289042,
120 0.0006209795828908682,
121 0.0005519681726582348
122 ],
123 "n": 8
124 }
125 },
126 {
127 "cfg": {
128 "lr": 0.01
129 },
130 "mean": 0.0003928114747395739,
131 "full": {
132 "mean": 0.0003928114747395739,
133 "std": 0.00021742121198405602,
134 "per_seed": [
135 8.355393947567791e-05,
136 0.0004895747406408191,
137 0.0001629032049095258,
138 0.0004371287068352103,
139 0.0004481615615077317,
140 0.0004002868663519621,
141 0.0008424268453381956,
142 0.0002784559328574687
143 ],
144 "n": 8
145 }
146 }
147 ],
148 "paired_delta": {
149 "per_seed": [
150 -0.027058529332862236,
151 -0.02831797592807561,
152 -0.03971586005354766,
153 -0.027458273456431925,
154 -0.050240384589415044,
155 -0.021572406636551023,
156 -0.02356041135499254,
157 -0.03954867480206303
158 ],
159 "mean": -0.032184064519242384,
160 "permutation_p": 0.0081
161 },
162 "mechanism_signature": {
163 "near_predictive_variance": 0.0006225919933058321,
164 "far_predictive_variance": 0.0006237539928406477,
165 "observed_far_near_ratio": 1.0018663901034859,
166 "predicted_far_near_ratio": 1.0018666982650757,
167 "interval_90pct_coverage": 1.0,
168 "confirmed": true
169 },
170 "structural_match": "controlled damped pendulum multi-step dynamics"
171 }
172}