A sampling strategy that prioritizes tasks with high disagreement or uncertainty in performance, focusing evaluation effort where it matters most.