Users are reporting the existence of incorrect pairings of accessions and types generated by other tools. This isn't a problem with hgvs, but it would be nice to provide a mechanism to warn about these.
For example, SnpEff current returns p. variants with NM (i.e., mRNA/cDNA) accessions. A similar problem was also reported for an unspecified LOVD installation for BRCA1.
This issue proposes a new intrinsic validation step that will validate accessions against types.
1) Known good pairs, e.g., NM w/ c. 2) Known bad pairs, e.g., NM w/ p., NP w/ c., etc. 3) Everything else.
Two modes seem likely. In a strict mode, cases 2 and 3 raise exceptions because they are not known to be valid. In a relaxed mode, only case 2 raises an exception and case 3 is presumed okay. In other words, strict is pessimistic about unknown pairs and relaxed is optimistic about them.