VCF features only on first exon after loading VCF to coding sequence

a) fetch Uniprot OGT1_HUMAN, Get Cross-References | Ensembl
  - add VCF data from file to CDS CDS|ENST00000373719
  - 5 features are added (2 missense, 2 synonymous, 1 frameshift)
compare to:
b) fetch Ensembl ENST00000373719
  - add VCF from the same data file
  - Get Cross-References | Uniprot
  - CDS CDS|ENST00000373719 has 227 VCF features
or
c) fetch Ensembl ENST00000373719
  - Get Cross-References | Uniprot
  - add VCF to CDS alignment from the same data file
  - CDS CDS|ENST00000373719 has 5 VCF features

Some VCF is being lost when it is loaded to CDS rather than transcript - why?