With that in mind, I don't see why we would need to drop biblionumber at all - it references exactly what it is, is specific to Koha, and can be understood and trained in that context, fully independent of MARC.