* Leigh Sharpe wrote (29/06/06 03:03):
This was my first suspicion. I turned off Bayes tests temporarily and it had little effect. I'm seriously considering resetting the bayes and starting again
I can recommend that. I had a situation a while ago where the bayes database got mysteriously corrupted (sa-learn dump magic suddenly showed nspam way way less than nham). I deleted the whole bayes database, did a bit of manual training, let it carry on with the automatic training, and it was all fine again in a day or so.
If spam hits BAYES_00 (which carries a negative score), you're better off without bayes at all.
But with good bayes, most of the spam you've posted will be blocked. The difference between BAYES_00 and BAYES_99 is +6.099. So a small negative score with BAYES_00 will be sent over 5 by BAYES_99.
Chris
