To be fair I too probably wouldn't accept something as a bug until provided by a trivial reproduction case. I have seen too many developers call things bugs before looking at their own code.
It was a documentation/error response bug in their API, so it was hard to reproduce trivially. I gave them some source that reproduced it (https://github.com/samcgardner/neg-lb-initialiser), which really seems like it ought to be sufficient - it reliably reproduces wrong behaviour.