At the time, proponents of the Ptolemaic system were able to create a very complicated model that predicted the movements of planets reasonably well (don't know exactly who it was). So judging them by their predictions might have lead to the conclusion that the earth was indeed at the center of the universe.
Also, just try to write unit tests to "prove" that some piece of code is correct with regard to concurrency. What if a particular problem occurs so rarely that it is very unlikely to be caught in any empirical test? What if its occurrence depends on factors that are not well known or hard to replicate?
The problem with predictive models is that they sometimes break down unexpectedly. That doesn't make them useless, but it makes the distinction between logical proof and empirical proof all the more useful.