Intrinsic Schrödinger Bridge Diffusion / bench_report.json
Mechanism confirmed, baseline not beaten
1{
2 "bench_version": 1,
3 "track": "manifold_pendulum",
4 "model": "rnn_small",
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": 0.30809812992811203
17 },
18 {
19 "cfg": {
20 "lr": 0.003
21 },
22 "mean": 0.04099804675206542
23 },
24 {
25 "cfg": {
26 "lr": 0.01
27 },
28 "mean": 0.007313748239539564
29 }
30 ],
31 "full": {
32 "mean": 0.0062498191255144775,
33 "std": 0.001496702937330929,
34 "per_seed": [
35 0.00792852696031332,
36 0.00787876546382904,
37 0.0054243444465100765,
38 0.008023356087505817,
39 0.00436387537047267,
40 0.006577811203896999,
41 0.004088760819286108,
42 0.005713112652301788
43 ],
44 "n": 8
45 }
46 },
47 "idea": {
48 "mean": 0.017734745517373085,
49 "std": 0.0075728509012216545,
50 "per_seed": [
51 0.024637481197714806,
52 0.015422593802213669,
53 0.03352177515625954,
54 0.011271449737250805,
55 0.01013706810772419,
56 0.020766805857419968,
57 0.011616603471338749,
58 0.014504186809062958
59 ],
60 "n": 8
61 },
62 "comparison": {
63 "delta_mean": 0.011484926391858608,
64 "idea_wins": 0,
65 "n_pairs": 8,
66 "per_seed_diffs": [
67 0.016708954237401485,
68 0.007543828338384628,
69 0.02809743070974946,
70 0.0032480936497449875,
71 0.00577319273725152,
72 0.014188994653522968,
73 0.007527842652052641,
74 0.00879107415676117
75 ],
76 "p_value": 0.0081,
77 "mde": 0.006700379570878075,
78 "mde_rel_pct": 107.20917575877058,
79 "verdict": "idea worse (significant)",
80 "system_worked": false
81 },
82 "mechanism_signature": {
83 "custom_track": {
84 "name": "manifold_pendulum",
85 "file": "manifold_dynamics_track.py",
86 "domain": "dynamics_and_embedded_manifolds"
87 },
88 "idea_config": {
89 "lr": 0.01
90 },
91 "idea_sweep": [
92 {
93 "cfg": {
94 "lr": 0.001
95 },
96 "result": {
97 "mean": 0.26221400313079357,
98 "std": 0.05336933195312093,
99 "per_seed": [
100 0.352046400308609,
101 0.3472537696361542,
102 0.249238058924675,
103 0.23107391595840454,
104 0.24011655151844025,
105 0.25662562251091003,
106 0.19462718069553375,
107 0.22673052549362183
108 ],
109 "n": 8
110 }
111 },
112 {
113 "cfg": {
114 "lr": 0.003
115 },
116 "result": {
117 "mean": 0.08073520846664906,
118 "std": 0.02050061232661446,
119 "per_seed": [
120 0.11787223815917969,
121 0.08001995086669922,
122 0.1040424108505249,
123 0.06887325644493103,
124 0.0729207918047905,
125 0.08841918408870697,
126 0.06128842756152153,
127 0.05244540795683861
128 ],
129 "n": 8
130 }
131 },
132 {
133 "cfg": {
134 "lr": 0.01
135 },
136 "result": {
137 "mean": 0.017734745517373085,
138 "std": 0.0075728509012216545,
139 "per_seed": [
140 0.024637481197714806,
141 0.015422593802213669,
142 0.03352177515625954,
143 0.011271449737250805,
144 0.01013706810772419,
145 0.020766805857419968,
146 0.011616603471338749,
147 0.014504186809062958
148 ],
149 "n": 8
150 }
151 }
152 ],
153 "mechanism_signature": {
154 "prediction": "intrinsic retraction keeps every predicted embedded angular state on S1; Euclidean output has nonzero norm error",
155 "predicted_baseline_violation_order": "nonzero",
156 "predicted_idea_violation": 0.0,
157 "observed_baseline_max_violation": 0.16893672943115234,
158 "observed_idea_max_violation": 1.1920928955078125e-07,
159 "observed_idea_metric_seed0": 0.024637481197714806,
160 "confirmed": true
161 }
162 }
163}