Confidence Calibration
Maps the distance & sample-size penalty scores (0.0 - 1.0) into discrete buckets to ensure our expected Edge mapping aligns linearly with prediction accuracy over testing.
0.0 - 0.2
41.2%
-0.8%
0.2 - 0.4
50.1%
+0.1%
0.4 - 0.6
54.5%
+1.4%
0.6 - 0.8
61.2%
+3.8%
0.8 - 1.0
68.9%
+7.1%