Its been some time detached from the mat sci folks deeply involved in the space but its both. There is a bunch of theoretical underpinnings but ultimately a lot of throwing darts on the board as well.
Yeah. It depends a lot on the type of material, too. Conventional metal alloys -- like LPSO-Mg -- are the toughest to model. Too many variables. Ceramics and intermetallics are a lot easier to model in principle, but they can have surprising properties on an atomic level, and there's really no predictive method for that sort of thing. Modeling does get you pretty far with high-entropy alloys -- because to a substantial extent their properties hinge on how a bunch of different atoms might fit together randomly, and that's something that can be computationally predicted. A lot of the recent interest in HEAs is because they're relatively easy to model.