I've heard of that scenario, but I'm curious about what the best way to produce a bug-for-bug replacement would be. Should you do it right, and then put a layer in front of it that switches/transforms the output to match the old errors?