Sometimes you can produce tests (not unit, but integration and such) from the existing system by treating it as an oracle. That's what I did once, it worked really well and in the end of the differences between it and mine (where mine technically failed the tests), we found out the old system had dozens of critical bugs and mine hand a handful (old system gave many false negatives, which was a major no-no, mine gave some false positives which were tolerable; manual inspection can clear up the latter but not the former).
I'm the person that you're replying to responded to.
This is exactly what I went through during that four month period. The existing tests weren't very useful and I was completely changing the structure of how the program was architected, but the input v output could be measured. I, like you, also discovered critical bugs in the existing implementation because I needed to cover the entire combinatorial set or possible inputs whilst the original implementers presumably went through them as they came up.
