I got all fired up about decision theory this past weekend, and wrote something about it today, in the context of the Astrometry.net
verify step. Here is the first paragraph:
All science is really decision theory, not inference! After all, the end result of an inference is a paper, and what one writes in the paper depends on the result of the inference, and not at all trivially. That is, the content of the paper is not simply a statement of the posterior distribution function (obviously). Despite the obviousness of this, most discussions of scientific data analysis end at (or even before) the inference step, and don't give quantitative guidelines for anything that follows, except perhaps for the construction of confidence intervals.