SpaceX does do simulations, and tons of it. SN8 and 9 wouldn't have performed that well during the descent stage without amazing guidance software, which could only have been developed through extensive simulations.

Simulations are always going to be "assume a spherical cow in a vacuum" to a degree, as the real world is unbelievably complex, and you have to use models at some point. These models _can_ be wrong, and the only way to test them is through experimentation.