Affects Version/s: None
Fix Version/s: None
Component/s: Schema Change
If a band has several distinct versions of a song that they've released over the years, and a user wants to enter the existence of a compilation including that song into MusicBrainz, but they've never HEARD the compilation, they'll have no way to know which recording to select. Their only choice is to either guess at random (which stands a decent chance of being wrong in cases of several distinct versions) or create a new recording (which will incorrectly imply that another version of the song exists besides the ones already known).
The general problem here is that, the "recordings" page for any artist who features on a large number of compilations, is flooded with duplicates—and while it's possible to just go through and mass merge them and hope that you're right, that's fixing one bad situation with another bad situation: the root cause is that people who don't know about a compilation's recordings are forced to weigh in nonetheless, and the only safe default (which they then go for) is to create a new recording which likely doesn't really exist, and can't really be told apart from a "real" recording. If the database allowed them to indicate that they don't know yet, then the "recordings" page wouldn't fill up with spurious entries that are indistinguishable from the real entries.
There are a few ways that this could be approached:
A) users could be given a choice between tying a track to a Recording (if they know which it is) or merely to a Work (which should be virtually unambiguous even from the tracklist alone)
B) they could be given an option to leave a track's Recording field blank (the way they can for example leave a release's Year field blank)—which would of course be a less-than-ideal type of data, but the crucial thing is that it would ACKNOWLEDGE that the data is un-ideal rather than simply masking it and diluting genuine data ambiguously (an "implicit recording" could be generated from the track's title and artist, that would allow people browsing an artist's recordings to know that the compilations exist, while being aware which of the recordings are confirmed unique and which are unknown)
C) we could approach the problem from the other direction, and have a 'data quality' rating for Recordings the way we do for Releases, and then freely auto-generate Low-quality Recordings for the first Release of any new Release Group (compilation or otherwise), and allow users to manually create High-quality Recordings for situations where someone has actually listened to multiple releases and would like to indicate that they're definitely the same
Any of these approaches would still allow us to implicitly link tracks to Artists and Works with the same level of certainty that we do now (remembering that this change would ONLY affect poorly-sourced data!), without requiring people to use the Recordings feature with spurious specificity as the only way to connect the ambiguous tracks to their Artists and Works.
Obviously this isn't a small or a simple change, but I at least thought I'd see what people thought; it's frustrating that the Recordings feature has no real way of distinguishing researched information from the flood of people who are trying to say "I don't know" but are not allowed to by the system. People wanting to enter partial information is a fairly common case, that MusicBrainz mostly allows for very well, but the Recordings feature just sticks out as something that could be very powerful if people weren't also required to use it to address situations that it isn't suited for.