Accessions historically associated with multiple species can cause invalid editions to appear in results.

[JacobsonMT]

Some accessions (~300) were in early editions associated with multiple species (ex. http://www.uniprot.org/uniprot/P84099) . The current assumption is that accessions are specific to a protein product and therefore a gene and a species, this assumption (apparently) only holds in recent editions.

A hypothetical situation where this can cause an issue: Asking for all annotations associated with one of these multi-species accessions might return data from its early years where some is from Human edition 14 and some is from Mouse edition 14, these editions are not concurrent and counting them together is incorrect.

Fix can either involve filtering out the annotations not associated with the desired species or modifying the logic of the algorithm to rely on the goa_release edition which was created to have an edition that satisfies the requirement of: goa_release1 == goa_release2 <-> goa_release1_date == goa_release2_date.