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

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

When models disagree on predictions, it often signals tricky or risky inputs worth closer look. Tracking metrics like ensemble variance or prediction margin highlights uncertain cases