Spectral-Edge Criticality Controller / 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 "epochs": 12
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "epochs": 12
17 },
18 "mean": 0.010470356093719602
19 },
20 {
21 "cfg": {
22 "lr": 0.003,
23 "epochs": 12
24 },
25 "mean": 0.005202302592806518
26 },
27 {
28 "cfg": {
29 "lr": 0.006,
30 "epochs": 12
31 },
32 "mean": 0.0028439466550480574
33 }
34 ],
35 "full": {
36 "per_seed": [
37 0.0017406052211299539,
38 0.0037223200779408216,
39 0.002092725830152631,
40 0.0038201354909688234,
41 0.0038962536491453648,
42 0.0017676886636763811,
43 0.0043351962231099606,
44 0.0036920576822012663
45 ],
46 "mean": 0.0031333728547906503,
47 "signatures": [
48 {
49 "edge_mean": 0.6060569062829018,
50 "gain_final": 1.0
51 },
52 {
53 "edge_mean": 0.6419745236635208,
54 "gain_final": 1.0
55 },
56 {
57 "edge_mean": 0.6132491379976273,
58 "gain_final": 1.0
59 },
60 {
61 "edge_mean": 0.6298487186431885,
62 "gain_final": 1.0
63 },
64 {
65 "edge_mean": 0.6433217823505402,
66 "gain_final": 1.0
67 },
68 {
69 "edge_mean": 0.6161705479025841,
70 "gain_final": 1.0
71 },
72 {
73 "edge_mean": 0.6346533372998238,
74 "gain_final": 1.0
75 },
76 {
77 "edge_mean": 0.6225631684064865,
78 "gain_final": 1.0
79 }
80 ]
81 }
82 },
83 "idea": {
84 "best_cfg": {
85 "lr": 0.006,
86 "epochs": 12
87 },
88 "sweep": [
89 {
90 "cfg": {
91 "lr": 0.001,
92 "epochs": 12
93 },
94 "mean": 0.013453479390591383
95 },
96 {
97 "cfg": {
98 "lr": 0.003,
99 "epochs": 12
100 },
101 "mean": 0.004181064839940518
102 },
103 {
104 "cfg": {
105 "lr": 0.006,
106 "epochs": 12
107 },
108 "mean": 0.0015980473981471732
109 }
110 ],
111 "per_seed": [
112 0.000768338271882385,
113 0.0010514536406844854,
114 0.0033124624751508236,
115 0.0012599352048709989,
116 0.0007571351598016918,
117 0.0012769970344379544,
118 0.002315002726390958,
119 0.001677743741311133
120 ],
121 "mean": 0.0015523835318163037,
122 "signatures": [
123 {
124 "edge_mean": 0.8978222989738791,
125 "gain_final": 1.507260824388018
126 },
127 {
128 "edge_mean": 0.8991352558280741,
129 "gain_final": 1.4971718810377141
130 },
131 {
132 "edge_mean": 0.9008403422532384,
133 "gain_final": 1.4441157022221471
134 },
135 {
136 "edge_mean": 0.9019607306183688,
137 "gain_final": 1.4800354971216234
138 },
139 {
140 "edge_mean": 0.9020658948006778,
141 "gain_final": 1.4013476220899848
142 },
143 {
144 "edge_mean": 0.9014030846181611,
145 "gain_final": 1.4688920100836202
146 },
147 {
148 "edge_mean": 0.8958116604630089,
149 "gain_final": 1.365893652127335
150 },
151 {
152 "edge_mean": 0.8992522999588555,
153 "gain_final": 1.5118891100008007
154 }
155 ]
156 },
157 "comparison": {
158 "delta_mean": -0.0015809893229743466,
159 "per_seed_diffs": [
160 -0.0009722669492475688,
161 -0.002670866437256336,
162 0.0012197366449981928,
163 -0.0025602002860978246,
164 -0.003139118489343673,
165 -0.0004906916292384267,
166 -0.0020201934967190027,
167 -0.0020143139408901334
168 ],
169 "idea_wins": 7,
170 "n_pairs": 8,
171 "p_value": 0.0324,
172 "verdict": "idea better (significant)"
173 },
174 "mechanism_signature": {
175 "prediction": "controlled effective edge near target below 1",
176 "predicted_target": 0.9,
177 "observed_edge_mean": 0.899786445939283,
178 "relative_error": 0.0002372822896855463,
179 "confirmed": true
180 }
181}