GECC-Gated Loop-Aware Message Passing / bench_report.json
Mechanism confirmed, baseline not beaten
1{
2 "bench_version": 1,
3 "track": "custom:gecc_loop_graph",
4 "model": "GraphNet",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.003
10 },
11 "sweep": [
12 {
13 "cfg": {
14 "lr": 0.001
15 },
16 "mean": 0.11218749638646841
17 },
18 {
19 "cfg": {
20 "lr": 0.003
21 },
22 "mean": 0.09812499862164259
23 },
24 {
25 "cfg": {
26 "lr": 0.01
27 },
28 "mean": 0.09999999776482582
29 }
30 ],
31 "full": {
32 "mean": 0.09812499862164259,
33 "std": 0.016665362924412768,
34 "per_seed": [
35 0.10999999940395355,
36 0.0925000011920929,
37 0.07999999821186066,
38 0.07750000059604645,
39 0.10249999910593033,
40 0.12999999523162842,
41 0.10749999433755875,
42 0.08500000089406967
43 ],
44 "n": 8
45 }
46 },
47 "idea": {
48 "mean": 0.09874999802559614,
49 "std": 0.017001838012669464,
50 "per_seed": [
51 0.11249999701976776,
52 0.09749999642372131,
53 0.07249999791383743,
54 0.07999999821186066,
55 0.10749999433755875,
56 0.125,
57 0.10999999940395355,
58 0.08500000089406967
59 ],
60 "n": 8
61 },
62 "comparison": {
63 "delta_mean": 0.0006249994039535522,
64 "idea_wins": 2,
65 "n_pairs": 8,
66 "per_seed_diffs": [
67 0.002499997615814209,
68 0.004999995231628418,
69 -0.007500000298023224,
70 0.002499997615814209,
71 0.004999995231628418,
72 -0.004999995231628418,
73 0.002500005066394806,
74 0.0
75 ],
76 "p_value": 0.816,
77 "mde": 0.0038301067414134374,
78 "mde_rel_pct": 3.9032935492634633,
79 "verdict": "no measurable effect",
80 "system_worked": false
81 },
82 "mechanism_signature": {
83 "prediction": "trained gate activates more on higher closure; correction is selectively weighted",
84 "observed_model_behavior": [
85 {
86 "mean_C_bin": 0.125,
87 "mean_alpha": 0.0,
88 "mean_correction_gap": 0.0
89 },
90 {
91 "mean_C_bin": 0.375,
92 "mean_alpha": 0.45109381963466777,
93 "mean_correction_gap": 0.3164594641019558
94 },
95 {
96 "mean_C_bin": 0.75,
97 "mean_alpha": 0.5374938638313956,
98 "mean_correction_gap": 0.31253063419590826
99 }
100 ],
101 "predicted_alpha_order": "low < medium < high",
102 "confirmed": true,
103 "custom_track": {
104 "name": "gecc_loop_graph",
105 "file": "gecc_bench.py",
106 "domain": "graph-nn"
107 }
108 },
109 "idea_grid": [
110 {
111 "cfg": {
112 "lr": 0.001
113 },
114 "mean": 0.1184374988079071,
115 "std": 0.02197077209447606,
116 "per_seed": [
117 0.11749999970197678,
118 0.13750000298023224,
119 0.08500000089406967,
120 0.09749999642372131,
121 0.1274999976158142,
122 0.14000000059604645,
123 0.14749999344348907,
124 0.0949999988079071
125 ]
126 },
127 {
128 "cfg": {
129 "lr": 0.003
130 },
131 "mean": 0.09874999802559614,
132 "std": 0.017001838012669464,
133 "per_seed": [
134 0.11249999701976776,
135 0.09749999642372131,
136 0.07249999791383743,
137 0.07999999821186066,
138 0.10749999433755875,
139 0.125,
140 0.10999999940395355,
141 0.08500000089406967
142 ]
143 },
144 {
145 "cfg": {
146 "lr": 0.01
147 },
148 "mean": 0.10374999884516001,
149 "std": 0.017897275587566683,
150 "per_seed": [
151 0.11749999970197678,
152 0.12999999523162842,
153 0.07750000059604645,
154 0.08999999612569809,
155 0.11749999970197678,
156 0.11749999970197678,
157 0.0949999988079071,
158 0.08500000089406967
159 ]
160 }
161 ]
162}