I already discussed this in IRC, but I thought that some others might want to weigh in, and to store the decision here for others to refer to in similar situations.
We have a collection of traditional music from Morocco (https://musicbrainz.org/collection/142ea0d7-7fdf-4ea5-9b04-219f68023d01, https://compmusic.upf.edu/corpora), and we spent some time with a collaborator digitising recordings (and uploading them to internet archive) and transcribing the recordings to scores (uploaded to musescore.com).
Because of the nature of this music, a work doesn’t really represent a traditional “composition”, instead it’s a combination of a melodic and a rhythmic mode. These can be combined as the performer wants.
In our case, we have some works (https://musicbrainz.org/work/7f3b9596-a028-41e2-a1a9-9243b47886bb), for which we have more than one recording by different artists (https://musicbrainz.org/recording/3e5a82a2-d806-45cc-876e-6fa8a2b5a61d, https://musicbrainz.org/recording/44183247-4857-40cd-82bc-b4e9e3f458f1). The music in each of these recordings differs significantly, and so we have two score transcriptions, one for each recording.
The relationship “get the score” is only valid between work-url, not recording-url, so our plan is to make an additional work for each recording, and join these works to the existing work with a work-work relationship (arrangement, or based on?). This will allow us to attach the url of the score to the recording (indirectly).
Does this sound like a good plan? Are there existing situations in the MB database that does something similar? Does anyone have any other suggestions?