OK - I had thought that bayes poisoning was making bayes useless but I'm
now thinking that I might be wrong. Most of the spam and ham are
processed by Exim rules, callback verification, and greylisting trics
and good blacklists that bayes was information starved. So what I'm
doing now is piping my whitelisting and blacklisting into bayes and with
saved messages I've seeded bayes with about 90k ham and 104k spams and
I'm going to see if that fixes the problem.
