It seems to me to be more of a spectrum, where a 'well-understood' system is one we can simulate accurately with extremely pared-down models. (I.e., we can simu

It seems to me to be more of a spectrum, where a 'well-understood' system is one we can simulate accurately with extremely pared-down models. (I.e., we can simulate it much more cheaply than running the whole atomic simulation) because we found symmetries/simplifications/shortcuts. Approximating ballistics arcs with parabolae, for example, is sort of a compression algorithm on the full simulation.