Hi!
I don’t agree with the milisecond time matches: when a recording is integrated into a compilation or a reedition of the disk is make, the recording times can change. An additional problem is that the track times in the booklets can be seriously off (by up to a few seconds) so even if someone carefully copies the track times from amazon/discogs/booklet they can be wrong. Also, acoustids are tricky, as discussed in How reliable is acoustID for merging recordings? (By the way, I’m wondering if it can be a sufficient basis for merging recordings with matching track times, works and where one with [unknown] artist) . I feel concerned because I made a large set of merges for Bach works, and bringing it up here is a good idea I think.
I looked at a few works with a lot of recordings, checked for identical (or mutually containing) recording dates, artists and acoustIDs.
Then I looked at recordings by the same artists and with similar times (±5 seconds), the labels corresponds (on the back cover the phonographic copyrights help also since labels usually pass recordings around), the tracklist were similar enough (several tracks with very close track times in a row), not speaking about additional data like discogs that can most importantly provide the recording dates.
I must admit that I have much less attention to detail when it comes to Alfred Scholz’ 1001 pseudonyms, there I generally do not look further than artist and timing.
For compilations, I tend to merge it with the recording with the closest track times, if there is one (and only one) with close enough track times.
I wonder if there is a set of guidelines specifically for merging classical works, since I believe that there are not that many recordings out there by a given artist, even for famous works, one just needs some merging.
(also, I’m sorry for your relative, that kind of stuff can be pretty soul-crushing)