While this code has been manually tested a great deal, it lacks tests. This is not a simple issue, as unit testing anything that's centrally built around an RNG is far from easy.
Comprehensive testing is not necessarily possible, but some review by folks other than the author is needed before wider use. This review is in two forms: Assuring that the language definition pieces are correct and up to date, and verifying that the code is accurately expressing those definitions.
The more eyes on this the better, as this project's usage could greatly assist in ensuring the correctness of any existing or new parsers.