Ideally you would re-implement the code to the same spec and compare the outputs. This, too, is a form of replication.

If the results are sufficiently different, then look at the source?