On Sun, 15 Feb 2004 19:10:47 -0500, John Fleming wrote: >� 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?
A fictional example: Evil SPAM 1-59 comes in and got high scores because the relay was found in 8 RBLs, the sender matched 5 rules and the software that sent it is buggy and therefore inserted a bunch of spam signs matching another 4 rules. These mails had a high enough score to be automagically sent to the black hole, and nobody even saw them. Evil SPAM 60-99 are almost identical, but: They came from other relays that wasn't in any RBLS, the sender address was different and didn't match any rules, and the bug didn't trigger either so the it didn't insert any spam signs. Thanks to our friend Bayes, SA noticed their similarity to Evil SPAM 1-59 and they were sent to the same hole. Regards /Jonas -- Jonas Eckerman, [EMAIL PROTECTED] http://www.fsdb.org/
