Sounds like map-vs-territory confusion to me. A model of a mouse in a computer isn't a mouse, the point of doing the experiment is to uncover whether what you thought the model predicted was in fact what happened, because it's possible you didn't incorporate some aspect of the real mouse.

Simulations are useful for making calculations. Say I'm convinced that buildings work a certain way and I have a finite element model of one. I don't know what the predictions are for a new building of a certain shape, so I shove it in a computer to tell me. The computer tells me what I should predict assuming that my finite elements ideas were correct. What it won't tell me is whether there's some aspect of that model that doesn't correspond to what the building is actually gonna be made of.