So for a particular MbRecording Id you probably have multiple documents. Each document represents data for the same recording submitted by different people, but of course they may have have incorrectly matched their song to the wrong MbRecoridingId and so in fact in some cases the data represents a different song entirely.

When we have a handful of documents for a Recording Id it should be possible to compare and pick the one that most documents are aligned top and disregard one that massivley outlie the norms.

Its theoretically possible for an end user to do this but the large number of calls required to receive all documents, plus the computational effort required to work out the valid documents make this not feasible in practise.

Make much more sense for this to be done by AcousticBrainz server itself, but too my knowledge it is not currently happening.