I think that simply distinguishing "correct" and "useful" dissolves the debate.

No, I think it's more complicated than that.

For example, imagine a complex decision, say what college to go to. Can you write out a Bayesian model that will tell you what to do? Well, kinda. You can, but it's going to be woefully incomplete and involve a lot of guesses without much support from data. A set of heuristics will do much better in this situation. Are you going to say that this Bayesian model is "correct" regardless? I don't think it's a useful application of the word.