A better criterion, I think, is if it's consistently correct. I know, if it isn't consistently correct, it won't be useful; but it's easier to measure "correct

A better criterion, I think, is if it's consistently correct. I know, if it isn't consistently correct, it won't be useful; but it's easier to measure "correct" than "useful." (And correct with respect to what is generally known. If there had been a pre-Einstein's Theory of General Relativity, "correct" would have been "conforms to Newton's theory of gravity.")