Noise-Triggered Latent Rank Adaptation / 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": 8
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "epochs": 8
17 },
18 "mean": 0.006717785610817373
19 },
20 {
21 "cfg": {
22 "lr": 0.003,
23 "epochs": 8
24 },
25 "mean": 0.001539975477498956
26 },
27 {
28 "cfg": {
29 "lr": 0.006,
30 "epochs": 8
31 },
32 "mean": 0.0013831749238306656
33 }
34 ],
35 "full": {
36 "mean": 0.001290830463403836,
37 "std": 0.0005098154991553572,
38 "per_seed": [
39 0.001963656395673752,
40 0.0016316777328029275,
41 0.0007375201093964279,
42 0.0011998454574495554,
43 0.0005706549272872508,
44 0.0011663567274808884,
45 0.0010045842500403523,
46 0.002052348107099533
47 ],
48 "n": 8
49 }
50 },
51 "idea": {
52 "mean": 0.0020039106311742216,
53 "std": 0.0006166624712056951,
54 "per_seed": [
55 0.0014695058343932033,
56 0.0029612784273922443,
57 0.002112278714776039,
58 0.0011878599179908633,
59 0.0012330585159361362,
60 0.0023434634786099195,
61 0.0020484665874391794,
62 0.002675373572856188
63 ],
64 "n": 8
65 },
66 "comparison": {
67 "delta_mean": 0.0007130801677703857,
68 "idea_wins": 2,
69 "n_pairs": 8,
70 "per_seed_diffs": [
71 -0.0004941505612805486,
72 0.0013296006945893168,
73 0.0013747586053796113,
74 -1.1985539458692074e-05,
75 0.0006624035886488855,
76 0.0011771067511290312,
77 0.001043882337398827,
78 0.0006230254657566547
79 ],
80 "p_value": 0.0303,
81 "mde": 0.0005597031685205955,
82 "mde_rel_pct": 43.359928696189485,
83 "verdict": "idea worse (significant)",
84 "system_worked": false
85 },
86 "mechanism_signature": {
87 "mechanism_signature": {
88 "prediction": "activation when lambda_next > 2.0 * noise_floor and persistence=2",
89 "observed_mean_active_rank": 16.0,
90 "observed_min_rank": 2,
91 "observed_max_rank": 30,
92 "observed_structural_events": 28,
93 "observed_crossing_samples": 56,
94 "confirmed": true
95 },
96 "idea_sweep": [
97 {
98 "cfg": {
99 "lr": 0.006,
100 "epochs": 8,
101 "con": 2.0
102 },
103 "mean": 0.0019327307236380875
104 },
105 {
106 "cfg": {
107 "lr": 0.001,
108 "epochs": 8,
109 "con": 2.0
110 },
111 "mean": 0.0686129005625844
112 },
113 {
114 "cfg": {
115 "lr": 0.003,
116 "epochs": 8,
117 "con": 2.0
118 },
119 "mean": 0.01151486299932003
120 }
121 ],
122 "math_check": {
123 "predicted_on_boundary": 2.0,
124 "corrected_population_eigenvalues": [
125 0.5,
126 2.0,
127 3.0
128 ],
129 "on_condition": [
130 false,
131 false,
132 true
133 ],
134 "off_condition": [
135 true,
136 false,
137 false
138 ],
139 "boundary_exact": true
140 },
141 "track_match": "dynamics contains recurrent controlled state evolution"
142 }
143}