So how do you tell between options (1 & 2) and (3)? It's not satisfying to chalk it up to "consensus", I think. Gelman himself recommends very generalized mode

So how do you tell between options (1 & 2) and (3)? It's not satisfying to chalk it up to "consensus", I think. Gelman himself recommends very generalized model checking as a philosophy to search the space of underlying theories, but aren't Bayesian methods supposed to let us "put a prior on that"? Why must something like human intuition go unmodeled?