in general, this sounds good (particularly the big data approach - gmail.com v gmali.com based on stats).

but i am concerned that they don't mention RFC 3696 when describing their grammar. it's all there (and implemented in - the now unsupported - lepl).

http://www.faqs.org/rfcs/rfc3696.html

http://www.acooke.org/lepl/rfc3696.html