Simulations of certain things are still a problem today, even though less than it was in 1960-s. For example, trans-sonic aerodynamics is still hard. I'd imagine a good simulation of Raptor's inner flow when Starship does active maneuring can be a problem, especially if you also need to take into account material properties under dynamic loads.
That is where they should spend their cash to produce something of real value that can be reused. Simulation close to perfection is the end game. Not sending something chunky into space.
Falcon 9 now flies and lands with no crashes, especially comparing to expendable rockets. Of course a modification big enough will require changing control parameters, but that's a routine task for engineering.
Some day we'll have good enough models for air flows in all necessary modes and good enough computers to run those models; we aren't there yet. Testing is still required.
