Nice to see you wanting to dev in PIcard.
Adding multiple releases UI would be useful, compilation matching issues are some of the most common reported about Picard.
You seem to be focusing on if you already have assigned MBID info to your files.
If you have the wrong information there then technically you have used Picard incorrectly in the first place.
Maybe it is an idea to look at getting that correct information in there earlier on (I am not sure if you are suggesting that)
Here are a few suggestions:
Often the recommended workflow is
Cluster --> Lookup (the Cluster)
so maybe that is an area to improve.
Possibly look at more elements here.
Track number, track lengths, etc. And try better matching as a group. Does the best match release have the same about of tracks, do the track names look similar, etc
tracks times within a few percentage and/or within a few seconds for all the tracks.
This might make the searching widely complex
Also it does still get a little confused for multiple artists in a cluster. Sometimes for a releases with multiple artists can be an artist name or Various artists in Musicbrainz.
In the cluster code it fixes the choice to Various Artists or a fixed artist name if they appear multiple times. It is a best guess with multiartist variable, but sometimes this isn't correct and there is no way of doing a clustered lookup for the other option.
Sorry if that isn't within the scope of what you want to achieve, just a few ideas.
(A more crude solution to getting re-running the checks would be to an option to intentionally ignore all MBIDs but that is a option in Option-->General)