Lunar Color-Connectivity Regularizer / bench_report.json
Mechanism confirmed, baseline not beaten
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 "epochs": 12
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.001,
16 "epochs": 12
17 },
18 "mean": 201.0537109375
19 },
20 {
21 "cfg": {
22 "lr": 0.003,
23 "epochs": 12
24 },
25 "mean": 58.3254508972168
26 },
27 {
28 "cfg": {
29 "lr": 0.01,
30 "epochs": 12
31 },
32 "mean": 15.770377159118652
33 }
34 ],
35 "full": {
36 "mean": 16.54259741306305,
37 "std": 2.506983808304334,
38 "per_seed": [
39 16.479936599731445,
40 13.55350399017334,
41 18.048686981201172,
42 14.999381065368652,
43 14.820082664489746,
44 16.00037956237793,
45 16.12912368774414,
46 22.30968475341797
47 ],
48 "n": 8
49 }
50 },
51 "idea": {
52 "mean": 16.655847430229187,
53 "std": 2.25515162281877,
54 "per_seed": [
55 17.07933807373047,
56 14.434268951416016,
57 16.626239776611328,
58 15.863788604736328,
59 14.660067558288574,
60 16.29985809326172,
61 16.1023006439209,
62 22.180917739868164
63 ],
64 "n": 8
65 },
66 "comparison": {
67 "delta_mean": 0.1132500171661377,
68 "idea_wins": 4,
69 "n_pairs": 8,
70 "per_seed_diffs": [
71 0.5994014739990234,
72 0.8807649612426758,
73 -1.4224472045898438,
74 0.8644075393676758,
75 -0.16001510620117188,
76 0.29947853088378906,
77 -0.026823043823242188,
78 -0.1287670135498047
79 ],
80 "p_value": 0.6628,
81 "mde": 0.6269619015296144,
82 "mde_rel_pct": 3.7899846431282116,
83 "verdict": "no measurable effect",
84 "system_worked": false
85 },
86 "idea_config": {
87 "lr": 0.01,
88 "epochs": 12
89 },
90 "idea_sweep": [
91 {
92 "cfg": {
93 "lr": 0.001,
94 "epochs": 12
95 },
96 "result": {
97 "mean": 207.19963264465332,
98 "std": 20.029543267954704,
99 "per_seed": [
100 213.74232482910156,
101 188.0177001953125,
102 228.82228088378906,
103 174.27357482910156,
104 202.07249450683594,
105 209.18301391601562,
106 200.1274871826172,
107 241.35818481445312
108 ],
109 "n": 8
110 }
111 },
112 {
113 "cfg": {
114 "lr": 0.003,
115 "epochs": 12
116 },
117 "result": {
118 "mean": 61.74348449707031,
119 "std": 23.30208403807316,
120 "per_seed": [
121 52.126747131347656,
122 57.39317321777344,
123 69.72927856445312,
124 56.225345611572266,
125 38.751182556152344,
126 66.92098999023438,
127 36.6944465637207,
128 116.1067123413086
129 ],
130 "n": 8
131 }
132 },
133 {
134 "cfg": {
135 "lr": 0.01,
136 "epochs": 12
137 },
138 "result": {
139 "mean": 16.655847430229187,
140 "std": 2.25515162281877,
141 "per_seed": [
142 17.07933807373047,
143 14.434268951416016,
144 16.626239776611328,
145 15.863788604736328,
146 14.660067558288574,
147 16.29985809326172,
148 16.1023006439209,
149 22.180917739868164
150 ],
151 "n": 8
152 }
153 }
154 ],
155 "mechanism_signature": {
156 "prediction": "training lowers symmetric nearest cross-color embedding distance",
157 "predicted_direction": "lower",
158 "observed_idea_cross_color_nearest_mean": 0.07249955972656608,
159 "observed_idea_cross_color_nearest_per_seed": [
160 0.07713642716407776,
161 0.06891444325447083,
162 0.08666811883449554,
163 0.060866888612508774,
164 0.06561814993619919,
165 0.08363944292068481,
166 0.06347937881946564,
167 0.07367362827062607
168 ],
169 "confirmed": true
170 }
171}