My lab is running GBrowse v2.14 and I'm observing some strange behavior from the CDS glyph. It seems to happen when the view spans an intron. When the view completely spans an intron, one of the exon's CDS sequence changes to something that no longer corresponds to the 6 frame translation. When the view spans only a splice junction and not a whole intron, the CDS sequence returns to something that corresponds to the 6 frame translation. I have observed this on all of the introns that I've tried (screenshots attached).
Zooming out more, it appears that the CDS sequence is repeating itself from one exon to the next. This can be seen when multiple introns are spanned as well -- the same CDS sequence appears in all three exons when two introns are spanned (screenshot attached).
I've also attached a small set of fake gene info on a 10kb scaffold. Because I randomly picked CDS start-stop positions, the translations don't make much sense (i.e. there are stop codons in the CDS). But the behavior is consistent.