> We’re not learning new physics are we

We are - at least in terms of model fidelity. There's limitations to how much CFD/simulations you can do. That kind of data (+other sensors) is used to refine models - thermal, aerodynamics, structures. Especially with starship, they are able to stream out live-video and data so that they get it even if the vehicle breaks up. Controlled hypersonic flight of such structures has been done very few times. There's stuff that can be learned from previous vehicles like the Space Shuttle but there are a lot of things that are very different - different control surfaces, flight profiles, thermal management etc.