Covariance-Conditioned Neural Rollouts / 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": 0.006
10 },
11 "sweep": [
12 {
13 "cfg": {
14 "lr": 0.001
15 },
16 "mean": 0.0007079141651047394
17 },
18 {
19 "cfg": {
20 "lr": 0.003
21 },
22 "mean": 0.00019564990725484677
23 },
24 {
25 "cfg": {
26 "lr": 0.006
27 },
28 "mean": 4.179111101620947e-05
29 }
30 ],
31 "full": {
32 "mean": 3.971884348175081e-05,
33 "std": 7.221693826148878e-06,
34 "per_seed": [
35 4.624741632142104e-05,
36 5.4139207350090146e-05,
37 2.856133687600959e-05,
38 3.82164835173171e-05,
39 4.160348544246517e-05,
40 3.6566088965628296e-05,
41 3.5268141800770536e-05,
42 3.714858758030459e-05
43 ],
44 "n": 8
45 }
46 },
47 "idea": {
48 "mean": 8.592086487624329e-05,
49 "std": 7.508073923840228e-05,
50 "per_seed": [
51 0.00011502060806378722,
52 6.843992014182732e-05,
53 0.00027265894459560513,
54 4.1655894165160134e-05,
55 7.501575601054356e-05,
56 4.787765283253975e-05,
57 3.653249950730242e-05,
58 3.0165643693180755e-05
59 ],
60 "n": 8
61 },
62 "comparison": {
63 "delta_mean": 4.6202021394492476e-05,
64 "idea_wins": 1,
65 "n_pairs": 8,
66 "per_seed_diffs": [
67 6.877319174236618e-05,
68 1.4300712791737169e-05,
69 0.00024409760771959554,
70 3.439410647843033e-06,
71 3.3412270568078384e-05,
72 1.1311563866911456e-05,
73 1.2643577065318823e-06,
74 -6.982943887123838e-06
75 ],
76 "p_value": 0.03865,
77 "mde": 6.977709898395673e-05,
78 "mde_rel_pct": 175.67756980642312,
79 "verdict": "idea worse (significant)",
80 "system_worked": false
81 },
82 "math_check": {
83 "joint_min_eig": 0.542706194000172,
84 "schur_min_eig": 0.6700584265307041,
85 "psd": true
86 },
87 "idea_sweep": [
88 {
89 "cfg": {
90 "lr": 0.001
91 },
92 "result": {
93 "mean": 8.592086487624329e-05,
94 "std": 7.508073923840228e-05,
95 "per_seed": [
96 0.00011502060806378722,
97 6.843992014182732e-05,
98 0.00027265894459560513,
99 4.1655894165160134e-05,
100 7.501575601054356e-05,
101 4.787765283253975e-05,
102 3.653249950730242e-05,
103 3.0165643693180755e-05
104 ],
105 "n": 8
106 }
107 },
108 {
109 "cfg": {
110 "lr": 0.003
111 },
112 "result": {
113 "mean": 0.00017516915022497415,
114 "std": 0.00014182129868597662,
115 "per_seed": [
116 0.00031262406264431775,
117 4.148928201175295e-05,
118 0.0004854888829868287,
119 8.179861470125616e-05,
120 0.00012095236161258072,
121 5.18705346621573e-05,
122 0.00014855564222671092,
123 0.0001585738209541887
124 ],
125 "n": 8
126 }
127 },
128 {
129 "cfg": {
130 "lr": 0.006
131 },
132 "result": {
133 "mean": 0.0001611893669632991,
134 "std": 0.00019513377033064536,
135 "per_seed": [
136 1.1867330613313243e-05,
137 0.000506327603943646,
138 1.7718439266900532e-05,
139 0.00010111019219039008,
140 0.0004840953042730689,
141 2.954796582343988e-05,
142 8.626402268419042e-05,
143 5.2584076911443844e-05
144 ],
145 "n": 8
146 }
147 }
148 ],
149 "budget": {
150 "epochs": 18,
151 "batch": 128,
152 "n_train": 1600,
153 "n_test": 500
154 },
155 "mechanism_signature": {
156 "prediction": "joint Gaussian head yields calibrated predictive variance; 95% interval coverage near nominal 0.95",
157 "observed_nll": -2.5754659175872803,
158 "observed_coverage95": 1.0,
159 "observed_mean_variance": 0.0008512148051522672,
160 "confirmed": true
161 }
162}