When models disagree a lot—measured by high ensemble variance or low margin—it...
https://www.primary-bookmarks.win/model-disagreement-helps-spot-tricky-inputs-by-measuring-uncertainty-through
When models disagree a lot—measured by high ensemble variance or low margin—it often signals tricky or risky inputs. By flagging the top 1-2% of these disputed cases for human review, teams can catch errors early