Maximum-Entropy Relational Block Kernel / bench_report.json
Mechanism confirmed, baseline not beaten
1{
2 "bench_version": 1,
3 "track": "relational_block_graph",
4 "model": "shared_node_encoder",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.003,
10 "weight_decay": 0.0001
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "weight_decay": 0.0
17 },
18 "mean": 0.21859551593661308
19 },
20 {
21 "cfg": {
22 "lr": 0.001,
23 "weight_decay": 0.0001
24 },
25 "mean": 0.2186652570962906
26 },
27 {
28 "cfg": {
29 "lr": 0.003,
30 "weight_decay": 0.0
31 },
32 "mean": 0.19878213480114937
33 },
34 {
35 "cfg": {
36 "lr": 0.003,
37 "weight_decay": 0.0001
38 },
39 "mean": 0.19875044003129005
40 },
41 {
42 "cfg": {
43 "lr": 0.01,
44 "weight_decay": 0.0
45 },
46 "mean": 0.21140450984239578
47 },
48 {
49 "cfg": {
50 "lr": 0.01,
51 "weight_decay": 0.0001
52 },
53 "mean": 0.21123026683926582
54 }
55 ],
56 "full": {
57 "mean": 0.194888049736619,
58 "std": 0.01561744050651578,
59 "per_seed": [
60 0.17818164825439453,
61 0.2278975248336792,
62 0.2048080563545227,
63 0.18411453068256378,
64 0.2039090394973755,
65 0.18120922148227692,
66 0.18494731187820435,
67 0.194037064909935
68 ],
69 "n": 8
70 }
71 },
72 "idea": {
73 "mean": 0.19668109714984894,
74 "std": 0.01395659198231783,
75 "per_seed": [
76 0.18862538039684296,
77 0.22365549206733704,
78 0.20406734943389893,
79 0.1854495406150818,
80 0.20570962131023407,
81 0.18311351537704468,
82 0.17992492020130157,
83 0.20290295779705048
84 ],
85 "n": 8
86 },
87 "comparison": {
88 "delta_mean": 0.0017930474132299423,
89 "idea_wins": 3,
90 "n_pairs": 8,
91 "per_seed_diffs": [
92 0.010443732142448425,
93 -0.004242032766342163,
94 -0.0007407069206237793,
95 0.0013350099325180054,
96 0.0018005818128585815,
97 0.0019042938947677612,
98 -0.005022391676902771,
99 0.008865892887115479
100 ],
101 "p_value": 0.38515,
102 "mde": 0.004627252745979375,
103 "mde_rel_pct": 2.374313228662745,
104 "verdict": "no measurable effect",
105 "system_worked": false
106 },
107 "mechanism_signature": {
108 "prediction_vs_observed": {
109 "baseline_pred_mean": 0.36992982029914856,
110 "idea_pred_mean": 0.35136160254478455,
111 "observed_edge_label_mean": 0.36666667461395264,
112 "baseline_abs_mean_calibration_error": 0.003263145685195923,
113 "idea_abs_mean_calibration_error": 0.01530507206916809
114 },
115 "trained_model_parameter_counts": {
116 "baseline": 1996,
117 "idea": 937
118 },
119 "confirmed": true,
120 "note": "Signature is measured on held-out predictions of trained paired systems; confirmed means the block system produces finite, label-calibrated relational predictions."
121 },
122 "idea_sweep": [
123 {
124 "cfg": {
125 "lr": 0.003,
126 "weight_decay": 0.0001
127 },
128 "result": {
129 "mean": 0.19953397288918495,
130 "std": 0.012231897269365018,
131 "per_seed": [
132 0.1849757432937622,
133 0.225642591714859,
134 0.20098033547401428,
135 0.1886790692806244,
136 0.20687870681285858,
137 0.1936967372894287,
138 0.19104768335819244,
139 0.20437091588974
140 ],
141 "n": 8
142 }
143 },
144 {
145 "cfg": {
146 "lr": 0.001,
147 "weight_decay": 0.0001
148 },
149 "result": {
150 "mean": 0.22159693762660027,
151 "std": 0.035181132803443205,
152 "per_seed": [
153 0.19724169373512268,
154 0.22439901530742645,
155 0.3056352436542511,
156 0.20666202902793884,
157 0.2244376391172409,
158 0.19750705361366272,
159 0.18519867956638336,
160 0.23169414699077606
161 ],
162 "n": 8
163 }
164 },
165 {
166 "cfg": {
167 "lr": 0.01,
168 "weight_decay": 0.0001
169 },
170 "result": {
171 "mean": 0.19668109714984894,
172 "std": 0.01395659198231783,
173 "per_seed": [
174 0.18862538039684296,
175 0.22365549206733704,
176 0.20406734943389893,
177 0.1854495406150818,
178 0.20570962131023407,
179 0.18311351537704468,
180 0.17992492020130157,
181 0.20290295779705048
182 ],
183 "n": 8
184 }
185 }
186 ],
187 "custom_track": {
188 "name": "relational_block_graph",
189 "file": "graph_track.py",
190 "domain": "graph-nn"
191 }
192}