Sabine's argument is basically that many predictions are conditional, and the conditions often aren't met in practice, therefore we shouldn't judge scientific m

Sabine's argument is basically that many predictions are conditional, and the conditions often aren't met in practice, therefore we shouldn't judge scientific models based on predictions. Well, the conditions are met sometimes, and those are the times when the models can be fairly validated or not. Also, you can make "unit tests" that examine submodels with less conditions/assumptions. Getting all the conditions/assumptions right for the "integration test" is of course hard. If you can show that you model can make many blind predictions that end up being correct, that's a good model! I don't think Sabine denies that, but she sees predictions as impractical. Her alternative is not better as-is to me, though. In my experience, physicists tend to prefer "explanatory power" or other qualitative metrics to more objective ways to determine whether a model is validated. In contrast, my experience is that engineers and statisticians tend to prefer objective measures. Explanatory power is a useful concept, but at present it's rather vague. I did recommend AIC as a similar objective measure to what Sabine argues for on her blog, but no one responded: http://backreaction.blogspot.com/2020/05/predictions-are-overrated.html?showComment=1588873188265#c6900263718570345454 http://backreaction.blogspot.com/2020/05/predictions-are-ove...