Well data extraction from PDF is always tricky and there isn’t a defined way of how to translate PDF to EPUB 1:1 so I don’t think it’s calibre is the problem. It’s difficult to program how to reverse engineer automatically.