I think we actually agree - if competing theories both expect a lot of "normal" to happen, you make much smaller updates on "normal" than on "abnormal". What I

I think we actually agree - if competing theories both expect a lot of "normal" to happen, you make much smaller updates on "normal" than on "abnormal". What I take issue with in particular is: >I think this happens because people generally believe that a anecdote confirming the null hypothesis serves as evidence for that hypothesis. It is evidence. It's not much evidence, and it's evidence you should expect from "unusual things sometimes happen"-type hypotheses, but it is evidence. Maybe the phrasing I want is something like "People generally believe that anecdotes consistent with multiple theories help distinguish them". Dunno. Probably a lot of this is just Bayesian vs Frequentist worldviews talking past each other. >It is now logically impossible to prove the machine works perfectly at all times It's logically impossible to prove anything at 100% certainty. The best you can do is gather more observations and make incremental updates each time. I can give you what my belief is that my next observation will say that everything is fine (if I don't know anything else, it's Laplace's Rule of Succession - if I observe it 100 times at it all works, my confidence level for the next time is 101/102).