I've been discussing in some detail here a mathematical model of the response of Earth's climate to radiative forcings, trying to address some of the concerns expressed elsewhere on the need for such a model to be "physically realistic". In the case of the two-box model, a given fit of the response function to a two-time-constant decay curve could come from one of many different underlying physical models that correspond to a partitioning of Earth's climate system into two parts with different response rates. So the question has been whether any of these possible underlying physical models are in some way "realistic" or not. That essentially reduces to criteria on the magnitude of the various constants and partial outcomes in the model relative to real components of our planet.

This will necessarily be a somewhat rambling collection of my thoughts as I don't feel I've settled down to any solid conclusions on the matter. Perhaps just writing this down will clarify some of my thinking, or perhaps the ideas here will fetch some comments from others that will help point me in better directions. This is the development of some of the thinking from my earlier comments on measuring wrongness - the science I'm familiar with centers on measurement and quantification, and it certainly seems potentially fruitful to consider ways in which one could impose measures (of "wrongness" or "rightness" or just "uncertainty") on the world of ideas. In a sense that kind of measurement is what we impose under the banner of peer review, though the visible outcome is more a binary (publish/don't publish) than continuous measure.