When models disagree a lot—measured by high ensemble variance or low...

https://raymondsinterestingchat.quantlynix.com/posts/when-should-a-model-abstain-instead-of-guessing-understanding-selective-prediction-and-escalation-strategies

When models disagree a lot—measured by high ensemble variance or low margin—it’s a red flag that the input might be risky or unusual. We can catch those tricky cases early by monitoring disagreement scores