When models disagree, it signals uncertain or risky inputs worth a closer look....

https://seo.edu.rs/blog/counterfactual-augmentation-for-disputed-inputs-how-does-it-work-11189

When models disagree, it signals uncertain or risky inputs worth a closer look. Tracking metrics like ensemble variance or prediction margin helps spot these cases