Transfer-Spectrum Pseudo-Transition Scheduler / 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": 0.01,
10 "weight_decay": 0.0
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "weight_decay": 0.0
17 },
18 "mean": 0.0005152521189302206
19 },
20 {
21 "cfg": {
22 "lr": 0.001,
23 "weight_decay": 0.0001
24 },
25 "mean": 0.000567634284379892
26 },
27 {
28 "cfg": {
29 "lr": 0.003,
30 "weight_decay": 0.0
31 },
32 "mean": 4.9621185098658316e-05
33 },
34 {
35 "cfg": {
36 "lr": 0.003,
37 "weight_decay": 0.0001
38 },
39 "mean": 0.00020817978656850755
40 },
41 {
42 "cfg": {
43 "lr": 0.01,
44 "weight_decay": 0.0
45 },
46 "mean": 1.4315585985968937e-05
47 },
48 {
49 "cfg": {
50 "lr": 0.01,
51 "weight_decay": 0.0001
52 },
53 "mean": 5.317307386576431e-05
54 }
55 ],
56 "full": {
57 "mean": 1.35165529400183e-05,
58 "std": 3.8039836159720083e-06,
59 "per_seed": [
60 1.1310443369438872e-05,
61 2.1796071450808086e-05,
62 1.0098720849782694e-05,
63 1.4057108273846097e-05,
64 1.5940304365358315e-05,
65 1.4363434274855535e-05,
66 1.1481065484986175e-05,
67 9.085275451070629e-06
68 ],
69 "n": 8
70 },
71 "idea_grid": [
72 {
73 "cfg": {
74 "lr": 0.001,
75 "weight_decay": 0.0
76 },
77 "mean": 0.001614412700291723
78 },
79 {
80 "cfg": {
81 "lr": 0.001,
82 "weight_decay": 0.0001
83 },
84 "mean": 0.0015862095024203882
85 },
86 {
87 "cfg": {
88 "lr": 0.003,
89 "weight_decay": 0.0
90 },
91 "mean": 0.000994522990367841
92 },
93 {
94 "cfg": {
95 "lr": 0.003,
96 "weight_decay": 0.0001
97 },
98 "mean": 0.0009985761280404404
99 },
100 {
101 "cfg": {
102 "lr": 0.01,
103 "weight_decay": 0.0
104 },
105 "mean": 0.000576967402594164
106 },
107 {
108 "cfg": {
109 "lr": 0.01,
110 "weight_decay": 0.0001
111 },
112 "mean": 0.000580504653044045
113 }
114 ]
115 },
116 "idea": {
117 "mean": 0.000576967402594164,
118 "std": 0.00020098972998599003,
119 "per_seed": [
120 0.0005425920826382935,
121 0.0006804523291066289,
122 0.0004205087316222489,
123 0.0007593809277750552,
124 0.0004023100482299924,
125 0.00024564965860918164,
126 0.0009024846949614584,
127 0.0006623607478104532
128 ],
129 "n": 8,
130 "best_cfg": {
131 "lr": 0.01,
132 "weight_decay": 0.0
133 }
134 },
135 "comparison": {
136 "delta_mean": 0.0005634508496541457,
137 "idea_wins": 0,
138 "n_pairs": 8,
139 "per_seed_diffs": [
140 0.0005312816392688546,
141 0.0006586562576558208,
142 0.0004104100107724662,
143 0.0007453238195012091,
144 0.0003863697438646341,
145 0.0002312862243343261,
146 0.0008910036294764723,
147 0.0006532754723593825
148 ],
149 "p_value": 0.0081,
150 "mde": 0.00017979348963317735,
151 "mde_rel_pct": 1330.1726440982216,
152 "verdict": "idea worse (significant)",
153 "system_worked": false
154 },
155 "mechanism_signature": {
156 "prediction": "low normalized transfer gap with high projection entropy marks a pseudo-transition",
157 "predicted_thresholds": {
158 "gap_lt": 0.12,
159 "entropy_gt": 0.7
160 },
161 "observed_mean_gap": 0.01636176293793846,
162 "observed_mean_entropy": 0.9524553120136261,
163 "gap_entropy_correlation": 0.0913397509063977,
164 "trigger_count": 64,
165 "confirmed": true
166 }
167}