[abstrakraft]

I agree that you can't know what is "real" without looking at our universe from outside it, but that, in and of itself, doesn't imply that every model must be flawed, in the sense that its predictions must not be 100% consistent with observation. We could stumble across the "real" model, or something equivalent to it (in the sense of identical predictions) - we'd have no way of knowing whether the model is "real", but it could still be right.

[strogonoff]

Such a model would look extraordinarily compelling to rely on all the time for all purposes, all the while remaining capable of being false in a critical and very difficult (impossible with technology of the time) to detect way.

I’d rather prefer a Unix way: a plurality of explicitly more limited but more numerous and conceptually diverse models.