If the model evidence-based, falsifiable, and tested, applying modeling to a set of conditions is a way of deriving evidence-based conclusions about the likely outcome of those conditions.

No, it's a way to derive evidence based hypotheses. A model that does not include a confounding variable (of which there are many) cannot give any conclusion about that variable - at most a hypotheses.

I've spent some time looking at the models, all of them fail - some spectacularly so (those that inform policy much more so, unfortunately). To the point of being useless, I might say, because their stated assumptions usually apply almost everywhere, but their prediction (and back testing results) match only a small sample (e.g. a few specific countries or cities).