Our code sometimes matches one column, sometimes the other.
I imagine that by no means is the dropby data is canonical but we should document clearly in the tests and in the code what is what.
Right now, I do not know if our implementation is correct or buggy for the commented out tests.
Granted I've not spent much time on this.
The unit tests do not get 100% code coverage either, which should be a goal for a new codec. 100% coverage does not guarantee correct implementation of course, but it's a start to make sure we at least test what we have.
At least, the new tests brought the coverage from 98%/93% line/branch to 100%/94%, so we've got that going for us ;)