Gary G. Taylor wrote: > > And here is a header from a beliefnet (gag) message SA caught: > <snip> > X-Spam-Status: Yes, score=5.4 required=5.0 tests=AWL,HELO_DYNAMIC_DHCP, > HTML_80_90,HTML_IMAGE_RATIO_02,HTML_MESSAGE,MIME_HTML_ONLY > autolearn=no version=3.0.4 > > <snip> > How the fsck did these idiots get into the AWL?! >
Step 1: Ditch any preconceptions that the AWL is a whitelist. It's not, it's a score-averager. It's called AWL due to lack of a better name that's not absurdly long. Step 2: The above example is perfectly normal and expected for the AWL. Note that the message was still tagged as spam. This is perfectly normal. The AWL, in the above example, felt the score of the message should be 5.0. SA scored it 5.8, so the AWL split the difference and made it 5.4 by subtracting 0.4 points. Still tagged as spam, no problem. Step 3: Read the WIKI to get a better idea of what the AWL really is, and what it does. Why the AWL sometimes "Scores the wrong way": http://wiki.apache.org/spamassassin/AwlWrongWay What the AWL is and how it works: http://wiki.apache.org/spamassassin/AutoWhitelist > When I view the AWL file I find probably two or three hundred different URLs > and email addresses. I am running SpamAssassin 3.0.4 installed as an rpm from > Mandrakesoft and I have not designated any block of senders as ham. > > The questions are: > 1) How do I clean out the white list? > It's not really a whitelist, as noted above, but you can clean out any "one off" addresses by using the check-whitelist script that comes in the tools subdirectory of the tarball. Note: most RPMs do not come with this, so you WILL have to download the tarball to get it. However, you don't need to install anything. It's just a stand-alone perl script. Read the top of the file to get usage directions. > 2) The installation of SpamAssassin set up KMail with filters for spam. There > are two actions available: Filter a message as spam, and filter a message as > ham; each goes into its own separate folder within KMail. Is using these > manual filters the right thing to do, and then run sa-learn through them at > the appropriate time? > Sounds reasonable to me.
