Well, I can say Picard has been pretty well flawless for me. And in those few instances where it misidentifies something, you can always do a manual search and match.
Nine times out of ten my process is to load the tracks into Picard, cluster them, look them up, do a quick scan to confirm it looks good, and then save the updated metadata. For those few times it messes up, I just reload the files, cluster them, then do a manual search to find the appropriate release. It really is very good at its job.