When models disagree on predictions, it often signals risky or uncertain...

https://wiki-byte.win/index.php/Counterfactual_Augmentation_for_Disputed_Inputs:_How_Does_It_Work%3F

When models disagree on predictions, it often signals risky or uncertain inputs. By tracking metrics like ensemble variance or prediction margin, you can spot these tricky cases