Spectrally identifiable phaseless recurrent layer / 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 },
11 "sweep": [
12 {
13 "cfg": {
14 "lr": 0.001
15 },
16 "mean": 0.0014828930143266916
17 },
18 {
19 "cfg": {
20 "lr": 0.003
21 },
22 "mean": 0.000868373506818898
23 },
24 {
25 "cfg": {
26 "lr": 0.01
27 },
28 "mean": 0.0004153431436861865
29 }
30 ],
31 "full": {
32 "mean": 0.0003938443751394516,
33 "std": 0.00015694612826889552,
34 "per_seed": [
35 0.0004267372714821249,
36 0.00025510918931104243,
37 0.00029441321385093033,
38 0.0006851129001006484,
39 0.00039094407111406326,
40 0.0005367420380935073,
41 0.00014661152090411633,
42 0.0004150847962591797
43 ],
44 "n": 8
45 }
46 },
47 "idea": {
48 "mean": 0.01808006438659504,
49 "std": 0.006827547286877937,
50 "per_seed": [
51 0.015054022893309593,
52 0.01623576134443283,
53 0.03221019357442856,
54 0.00664726784452796,
55 0.021471986547112465,
56 0.021092308685183525,
57 0.016446856781840324,
58 0.015482117421925068
59 ],
60 "n": 8
61 },
62 "comparison": {
63 "delta_mean": 0.01768622001145559,
64 "idea_wins": 0,
65 "n_pairs": 8,
66 "per_seed_diffs": [
67 0.014627285621827468,
68 0.01598065215512179,
69 0.03191578036057763,
70 0.005962154944427311,
71 0.0210810424759984,
72 0.020555566647090018,
73 0.016300245260936208,
74 0.015067032625665888
75 ],
76 "p_value": 0.0081,
77 "mde": 0.006166672867800231,
78 "mde_rel_pct": 1565.763854216973,
79 "verdict": "idea worse (significant)",
80 "system_worked": false
81 },
82 "mechanism_signature": {
83 "mechanism_signature": {
84 "prediction": "global phase is unobservable by magnitude trajectory",
85 "predicted_max_abs_difference": 0.0,
86 "observed_mean_abs_difference": 1.0522416005187551e-07,
87 "operator_pair_sum_gap": 6.656185635112166e-06,
88 "operator_sigma_min_S": 0.009892589246225832,
89 "confirmed": true
90 }
91 },
92 "idea_sweep": [
93 {
94 "cfg": {
95 "lr": 0.01
96 },
97 "result": {
98 "mean": 0.01808006438659504,
99 "std": 0.006827547286877937,
100 "per_seed": [
101 0.015054022893309593,
102 0.01623576134443283,
103 0.03221019357442856,
104 0.00664726784452796,
105 0.021471986547112465,
106 0.021092308685183525,
107 0.016446856781840324,
108 0.015482117421925068
109 ],
110 "n": 8
111 }
112 },
113 {
114 "cfg": {
115 "lr": 0.001
116 },
117 "result": {
118 "mean": 0.01906815404072404,
119 "std": 0.004254108046330375,
120 "per_seed": [
121 0.018856093287467957,
122 0.01202596165239811,
123 0.0262176301330328,
124 0.024596311151981354,
125 0.019076775759458542,
126 0.01587844267487526,
127 0.017323212698101997,
128 0.018570804968476295
129 ],
130 "n": 8
131 }
132 },
133 {
134 "cfg": {
135 "lr": 0.003
136 },
137 "result": {
138 "mean": 0.018212101538665593,
139 "std": 0.0021132105733096982,
140 "per_seed": [
141 0.017741071060299873,
142 0.01399772334843874,
143 0.02053205855190754,
144 0.0214164387434721,
145 0.018905995413661003,
146 0.018360335379838943,
147 0.017668306827545166,
148 0.017074882984161377
149 ],
150 "n": 8
151 }
152 }
153 ],
154 "parameter_counts": {
155 "baseline": 13313,
156 "idea": 24961
157 }
158}