Model disagreement helps spot risky inputs by measuring uncertainty like...

https://emilioezho292.theglensecret.com/counterfactual-augmentation-for-disputed-inputs-how-does-it-work

Model disagreement helps spot risky inputs by measuring uncertainty like ensemble variance or prediction margin. When models don’t agree, it flags cases that need extra attention