Fran Fabrizio wrote: > We've been running SpamAssassin on a Fedora server successfully for > about a year now. About a month ago, we all noticed a spike in the > number of spam getting through. I upgraded to 3.1.0 but that does not > seem to help. I'm beginning to suspect that some local database or > the learning has somehow gone awry. Here a sample score from an > obvious spam... > > ---------- > Subject: Investment Ideas > X-Spam-Status: No, hits=-1.098 tagged_above=-999 required=2 > tests=BAYES_00, RCVD_NUMERIC_HELO, UNPARSEABLE_RELAY > > Lots of similar spam is getting through. Is this an example of where > they have just figured out a way through the tests, or do I have a > local database or learning system that needs retraining? I've got > SpamAssassin to check Razor and DCC, you'd have thought those tests > would hit but I guess not. :-/
It looks like you have a Bayes problem. These spams score at least BAYES_90 here: OUR OCTOBER HOT PICK IS: CWTD. SYMBOL: CWTD CURRENT PRICE: $2.07 5 Day Target: $5 WE GIVE IT TO YOU AS A GIFT AND THIS IS WHY !!! *************PRESS RELEASE*****************PRESS RELEASE************* China World Trade Corporation Attends The World Trade Centers Association's 36th General Assembly in Shanghai ... Content analysis details: (6.6 points, 6.0 required) pts rule name description ---- ---------------------- -------------------------------------------------- 0.5 PT_CWORDS Subject contains command verb 2.0 FROM_M2 relayed from mail2 when mail1 is up 2.1 BAYES_90 BODY: Bayesian spam probability is 90 to 99% [score: 0.9313] 2.0 DCC_CHECK Listed in DCC (http://rhyolite.com/anti-spam/dcc/) If you don't use DCC, check it out. But the main thing will be getting Bayes fixed. It might require deleting the whole DB and re-training. Good luck Pierre