I’m not talking about back and forth between true and false, but between two explanations. You can have a multimodal probability distribution and two distant modes are about equally probable, and when you update, sometimes one is larger and sometimes the other. Of course one doesn’t need to choose a point estimate (maximum a posteriori), the distribution itself should ideally be believed in its entirety. But just as you can’t see the rabbit-duck as simultaneously 50% rabbit and 50% duck, one sometimes switches between different explanations, similarly to an MCMC sampling procedure.
I don’t want to argue this too much because it’s largely a preference of style and culture. I think the discussions are very repetitive and it’s an illusion that there is much to be learned by spending so much time thinking meta.
Anyway, I evaporate from the site for now.