Is the model's output distribution healthy, are confidence scores trustworthy, and is there subgroup skew?
Actual positive rate
0.63
Predicted positive rate
0.49
Prediction bias
-0.13
under-flags
ECE (calibration)
0.14
Brier 0.15
Confidence & calibration
Confidence distribution for current predictions.Calibration check: below the diagonal means confidence is too optimistic.
A reliability curve below the diagonal = over-confident model. ECE is the average gap between confidence and observed accuracy; keep it under 0.10 before letting confidence auto-route cases.
Subgroup skew
Lower-recall segments should be routed to review or targeted for data refresh.