Hodge-dual electrostatic loss / bench_report.json
Beats tuned baseline
1{
2 "bench_version": 1,
3 "track": "poisson_hodge_periodic",
4 "model": "mlp_tiny",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.003
10 },
11 "sweep": [
12 {
13 "cfg": {
14 "lr": 0.001
15 },
16 "mean": 2.206587553024292
17 },
18 {
19 "cfg": {
20 "lr": 0.003
21 },
22 "mean": 1.1046047359704971
23 },
24 {
25 "cfg": {
26 "lr": 0.01
27 },
28 "mean": 1.2606632113456726
29 }
30 ],
31 "full": {
32 "mean": 1.0800504088401794,
33 "std": 0.10522868591444227,
34 "per_seed": [
35 1.1444686651229858,
36 0.8914604783058167,
37 1.1053142547607422,
38 1.2771755456924438,
39 1.091360092163086,
40 1.0986981391906738,
41 1.034430980682373,
42 0.9974951148033142
43 ],
44 "n": 8
45 }
46 },
47 "idea": {
48 "mean": 0.4011378400027752,
49 "std": 0.059436192075174005,
50 "per_seed": [
51 0.33402565121650696,
52 0.3543514013290405,
53 0.42588725686073303,
54 0.5025325417518616,
55 0.3978813886642456,
56 0.43913376331329346,
57 0.4413507878780365,
58 0.3139399290084839
59 ],
60 "n": 8,
61 "selected_cfg": {
62 "lr": 0.01
63 }
64 },
65 "comparison": {
66 "delta_mean": -0.6789125688374043,
67 "idea_wins": 8,
68 "n_pairs": 8,
69 "per_seed_diffs": [
70 -0.8104430139064789,
71 -0.5371090769767761,
72 -0.6794269979000092,
73 -0.7746430039405823,
74 -0.6934787034988403,
75 -0.6595643758773804,
76 -0.5930801928043365,
77 -0.6835551857948303
78 ],
79 "p_value": 0.0081,
80 "mde": 0.07381216084595137,
81 "mde_rel_pct": 6.834140355098345,
82 "verdict": "idea better (significant)",
83 "system_worked": true
84 },
85 "mechanism_signature": {
86 "mechanism_signature": {
87 "prediction": "div(curl A)=0 for trained dual model",
88 "baseline_observed_constraint_residual": 106.66156768798828,
89 "idea_observed_div_curl_max": 2.384185791015625e-07,
90 "confirmed": true
91 },
92 "custom_track": {
93 "name": "poisson_hodge_periodic",
94 "file": "bench_stage2.py",
95 "domain": "pde"
96 },
97 "runtime_sec": 21.82808494567871
98 }
99}