What's the litmus test for an economic theory?

Very interesting comment by Mankiw reacting to Peter Diamond's criticism of his paper on 'height tax';

The Times also quotes a critic:

Peter Diamond, an economist at M.I.T., says the paper’s basic mistake is the notion “that if you can draw a silly inference from an approach, then that discredits a model.” He comments: “I think there is probably no model that passes that test."

I wonder what Peter's alternative approach is. If economic theorists are allowed to embrace inferences from a model that they like and cavalierly reject those that they consider "silly," what is the point of theory? That discretion gives the theorist the freedom to always confirm his priors. The economist ends up using theory like a drunk uses a light post--for support rather than illumination.

It seems to me that if you are going to reject a logical inference from a model, you have to explain why. That is not so easy for a height tax, which is precisely the point of the paper.