On Tue, 2008-05-13 at 16:16 -0500, Robotech_Master wrote: > I'm using SpamAssassin 3.2.3 w/ Perl 5.8.8 on Linux. I'm not the > sysadmin of the machine, but a user. > > I invoke it through a procmail recipe that says, in part, > > :0fw > | /usr/bin/spamc
> I am getting an immense amount of backscatter spam, and have trained > SA on it until SA gives it a reliable Bayes score of 99%. Please do note, that Bayes will be biased, if you train a LOT more ham than spam. Even though 50 times as much has been reported to work, one should at least expect to see "spammy looking ham" due to excessive, unbalanced training way earlier. This pretty much depends on your own ham and its variety in topic, too. Also, I'm not convinced that Bayes is the correct tool to fight backscatter at all... See your other post for a better way, where you ask about VBounce. :) Since you are using procmail anyway, let me stress a point HOW to handle bounces. Filter them. Into a different folder, for possible later review. Do not just treat them as spam -- keep in mind, the default VBounce scores are LOW, and set to merely have the rules not be disabled (which would be the case with a score of 0). Now, here goes my favorite quote these days: $ grep -A 2 procmail /usr/share/spamassassin/20_vbounce.cf # If you use this, set up procmail or your mail app to spot the # "ANY_BOUNCE_MESSAGE" rule hits in the X-Spam-Status line, and move # messages that match that to a 'vbounce' folder. > However, I'm still ending up getting tons of it passed through into my > mailbox. > > When I check the headers of some of the spams that end up in my > mailbox, I see something like the following: > > From MAILER-DAEMON Tue May 13 13:46:20 2008 > Return-Path: <> ... > X-Spam-Status: No, score=1.2 required=4.0 tests=AWL,BAYES_99 autolearn=no > version=3.2.3 Just a guess, but most likely due to an empty Return-Path. AWL is based on email address and the originating network block. Thus, you might see totally different results for mail sent by the same $address (well, the empty string here) from different net blocks. AWL is not related to Bayes, but all about the average score of mail previously seen by a specific sender (and origin). See also these and probably other articles in the wiki: http://wiki.apache.org/spamassassin/AutoWhitelist http://wiki.apache.org/spamassassin/AwlWrongWay > So, SA is giving it a BAYES_99, which should result in it hitting 5.0 > right off the bat. > > However, apparently the Auto-Whitelist is knocking it back down to > where it still ends up in my mailbox. > > Can someone please tell me how to make it stop? I'm getting a LOT of > these messages that should by all rights be safely filtered into > spammyland. Use VBounce. Filter them (using procmail) into bouncy-land. :) guenther -- char *t="[EMAIL PROTECTED]"; main(){ char h,m=h=*t++,*x=t+2*h,c,i,l=*x,s=0; for (i=0;i<l;i++){ i%8? c<<=1: (c=*++x); c&128 && (s+=h); if (!(h>>=1)||!t[s+h]){ putchar(t[s]);h=m;s=0; }}}
