A large number of EPUBs in the wild are not valid, so the team implemented a fallback strategy: first attempt strict XML parsing, and on failure use the lxml.html parser.