How much can we trust neural simulation strategies?