Picard's release matching algorithm treats all signals equally in a flat weighted average. This means exact identifiers like barcodes (weight 6 out of ~52 total) only contribute ~11% to the final score — not enough to be decisive.
As a result:

• Two releases with different barcodes can still tie if their titles and artists match
• Catalog number + label information is not used for matching at all
• ISRC is not used for track-level matching
• The totalalbumtracks comparison silently fails for most releases (the track-count key is absent from MB API responses at the release level)
• The trackcount_score function is binary (0.0 or 0.3 or 1.0), so a 1-track difference scores the same as a 10-track difference
• Cluster lookup only sends title/artist/tracks to the search API, missing barcode/catno/date that could narrow results to the correct edition

 

See https://github.com/metabrainz/picard/pull/3166