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

https://oliviasinsightfulthoughts.novacrestiq.com/posts/how-to-use-disagreement-to-catch-edge-cases-before-they-cause-harm

When models disagree on a prediction, it often signals tricky or risky inputs worth a closer look. By tracking ensemble variance or margin scores, you can spot these unsure cases