Directional Hölder Step Controller / bench_report.json
Beats tuned baseline
1{
2 "bench_version": 1,
3 "track": "tabular",
4 "model": "mlp_tiny",
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": 187.7414321899414
17 },
18 {
19 "cfg": {
20 "lr": 0.003
21 },
22 "mean": 19.22788381576538
23 },
24 {
25 "cfg": {
26 "lr": 0.01
27 },
28 "mean": 15.669923067092896
29 }
30 ],
31 "full": {
32 "mean": 16.668858289718628,
33 "std": 2.7275460099983677,
34 "per_seed": [
35 15.673293113708496,
36 11.123649597167969,
37 17.505971908569336,
38 18.37677764892578,
39 14.736111640930176,
40 21.006744384765625,
41 17.87649917602539,
42 17.05181884765625
43 ],
44 "n": 8
45 },
46 "idea_parity_sweep": [
47 {
48 "cfg": {
49 "lr": 0.001
50 },
51 "mean": 8.131636142730713
52 },
53 {
54 "cfg": {
55 "lr": 0.003
56 },
57 "mean": 11.476103067398071
58 },
59 {
60 "cfg": {
61 "lr": 0.01
62 },
63 "mean": 9.458438754081726
64 }
65 ]
66 },
67 "idea": {
68 "mean": 9.025225281715393,
69 "std": 1.7380340756046946,
70 "per_seed": [
71 8.675573348999023,
72 6.953583717346191,
73 8.074329376220703,
74 8.823058128356934,
75 13.168746948242188,
76 8.126770973205566,
77 8.566627502441406,
78 9.813112258911133
79 ],
80 "n": 8,
81 "selected_cfg": {
82 "lr": 0.001
83 }
84 },
85 "comparison": {
86 "delta_mean": -7.643633008003235,
87 "idea_wins": 8,
88 "n_pairs": 8,
89 "per_seed_diffs": [
90 -6.997719764709473,
91 -4.170065879821777,
92 -9.431642532348633,
93 -9.553719520568848,
94 -1.5673646926879883,
95 -12.879973411560059,
96 -9.309871673583984,
97 -7.238706588745117
98 ],
99 "p_value": 0.0081,
100 "mde": 2.941158171817024,
101 "mde_rel_pct": 17.644628808387758,
102 "verdict": "idea better (significant)",
103 "system_worked": true
104 },
105 "mechanism_signature": {
106 "alpha": 1.0,
107 "rho": 0.9,
108 "prediction": "accepted_eta ~= rho / observed_directional_Lhat",
109 "per_seed": [
110 {
111 "predicted_eta": 0.004098492150815332,
112 "observed_eta": 0.021143166023050074,
113 "ratio_observed_to_predicted": 5.1587669916225085
114 },
115 {
116 "predicted_eta": 0.004114928975852553,
117 "observed_eta": 0.015304616531639226,
118 "ratio_observed_to_predicted": 3.719290568913971
119 },
120 {
121 "predicted_eta": 0.003983274857888152,
122 "observed_eta": 0.0124441517321635,
123 "ratio_observed_to_predicted": 3.1241006900440524
124 },
125 {
126 "predicted_eta": 0.004497933758441668,
127 "observed_eta": 0.031266616003165305,
128 "ratio_observed_to_predicted": 6.951328694977888
129 },
130 {
131 "predicted_eta": 0.0055342093536037895,
132 "observed_eta": 0.015162049792218038,
133 "ratio_observed_to_predicted": 2.7396957403399913
134 },
135 {
136 "predicted_eta": 0.004462357038057677,
137 "observed_eta": 0.02690663006815206,
138 "ratio_observed_to_predicted": 6.029690102937991
139 },
140 {
141 "predicted_eta": 0.003190808863205975,
142 "observed_eta": 0.021336048993906148,
143 "ratio_observed_to_predicted": 6.686721113237942
144 },
145 {
146 "predicted_eta": 0.0022129334132261974,
147 "observed_eta": 0.012751969561600147,
148 "ratio_observed_to_predicted": 5.762473233665567
149 }
150 ],
151 "mean_ratio": 5.0215083919674885,
152 "confirmed": false
153 },
154 "runtime_seconds": 7.080888032913208,
155 "protocol_notes": {
156 "epochs": 8,
157 "batch": 128,
158 "dataset_sizes": [
159 400,
160 200
161 ],
162 "structural_match": "optimizer -> tabular"
163 }
164}