As someone who helped develop full-chassis modelling for racing cars in the mid / late 90s and someone who has done a lot of lab-based testing as well, I'm more aware than most of the gulf that can exist between theory and practice.
Computer modelling (both for structural applications and aero) is brilliant, but it does have its limitations, usually based around the number of iterations the software is capable of, as well as the compromises that have to be made in inputting and choosing the variables. Plus the fact that a computer simulation is very much idealised, which is not the case in the real world.
It's a useful tool, but it's NOT a silver bullet. Unfortunately, with the cap on in-season testing, it has rather become that in recent years. With the inevitable consequences. And that's very much highlighted this season, with such a massive change in the rules.