I'm assuming much more limited computational resources than all that are being used in at least some of the simulations

In that case, the simulation wouldn't be accurate enough that you'd have the same test. You write down that you saw a pigeon when you didn't in April. Emulated you writes down that he saw a family of ducks when he didn't in February. He decides his test failed. Several months later, he sees a pigeon and thinks nothing of it.