A situation I often encounter is that the files are well clustered but the album names are not suitable for lookup (e.g. clustered album names are romanized or in a particular language when only names from some other languages are in musicbrainz, or the album names contain weird characters). And when I run scanning on the files, the most common outcome is the detected matches go to a bunch of single releases with a few hits in the correct album, then I have to drag'n'drop the files from the singles into the album for full album match. Still gets the job done, but could be faster with cluster scanning.

Since the fingerprint lookup results already contain a bunch of potential matches, I think it would be neat if files in a cluster can do a majority vote and pick a release that is most commonly shared between all of them, then fall back to other releases when unable to.