Projections aren't the same as actual performance, and moreover, actual performance will involve quirks. Hence a simulator cannot accurately simulate the aerodynamics of the J-20, especially since aerodynamic flow is computationally difficult to predict.
What you have in mind is a training simulator, which is mainly used to train pilots to operate an existing plane. In this case, a simulator is simply a replica of an existing equipment and is only used by operators (pilots, in this case) of the equipment.
However, during research and development, simulations are mainly used to predict various parameters that cannot be or too expensive to or too difficult to be experimentally obtained. The main purpose of simulations at this stage is NOT simply to replicate the actual performance, but to explore what we don't know and to see how much we can push the envelope without actually physically doing the testing, which can be dangerous and expensive.
In this case when the Chinese are testing thrust vectoring using their simulators, the first and the foremost thing that they must do is to establish ALL the parameters to completely replicate the fly performance of their existing J-20. They will not do anything else until they can tweak their simulators to fully replicate 100% of their existing J-20.
Once they have done that, they will move forward and add the thrust vectoring based on their goals and initial plans. They will use this new sets of thrust vectoring parameters to test their imagined J-20 on the simulator. With a highly experienced test pilot, they can get a good idea of what they need to actually physically do to modify their J-20. With the help of the test pilot, they continue to tweak their parameters until they are satisfied. Then they move on to manufacturing the actual parts based on the parameters they get from simulations. Then they move on to testing these parts on an actual J-20. Of course, the actual performance of the new J-20 won't be exactly as simulated. No one expects it to be. But that's not what the simulators are for.
No simulator is perfect. So once they actually test the new parts, they will get actual experimental data. they will then use these experimental data to further modify their mathematical models and conduct more simulations to further push the envelope. So experimental testing and simulations always go hand-in-hand.
And this kind of simulation is very different from the training simulators used by pilots.