The default rule scores are generated with an assumed threshold of 5 and a target of 1 false positive in 2,500 non-spams. It sounds like you may be substantially increasing the false positive rate. Which you are certainly entitled to do, but I would not recommend.
http://wiki.apache.org/spamassassin/ImproveAccuracy On 06/03, Bill Polhemus wrote: > Hello. > > I am not a major admin. I have used a Linux box w/ Sendmail + Spamassassin > off and on for years, just for personal and small-biz email. I have only two > dozen or so accounts allocated among three domains. > > Using third-party email service for many years, which supposedly includes > Spam filtering, I noticed that gradually, of ~500 or so mails per account > per day, about 40% are spam. And in fact I noticed perhaps half again as > many spam were getting through as were caught in my email service provider's > Spam trap (I have no idea what they use). > > Decided to take things in hand again. > > After about 3 months of fiddling I've got it to the point where I'm down to > maybe two Spam per account per day getting through. > > Typical SA Bayes files sizes are about 650K Bayes_seen/AWL and 1.2G > Bayes_toks > > Thing is, in order to get this performance I've had to set the threshold for > Spam/Ham at a SA score of 2, after all hand-feeding and tweaking I know to > do. I lowered it gradually over time by 0.5 every two weeks or so, to this > point. > > So far I've found maybe 1 or 2 false positives per account per week at this > scoring. > > I'm fine with it as is, but thought some folks here might find it interesting > to note. > > William L. Polhemus, Jr. P.E. > Sent from my iPhone 5 -- "Believe nothing, no matter where you read it or who has said it, even if I have said it, unless it agrees with your own reason and your own common sense." - Buddha, 563-483 B.C. http://www.ChaosReigns.com
