It's painful, but I have to ask a dumb question again. Why auto-learn messages with, say, a spam score of 9? (Or any score above your spam threshhold?) It seems that to score that high, the types of stuff in that message are ALREADY being recognized as spam. So what's the reasoning behind auto-learn?
Oh, I guess it's for those who start fresh WITHOUT a ham or spam database? That way the Bayes eventually learns >200 each and starts working. OK, so if that's the main reason auto-learn exists, then is there any point in using it once you've built a sufficient bayes base? Thanks! - John
