Hi there, Excuses if this issue is not new for you. SA is new for me.
Running SpamAssissin 2.54 since 3 weeks. Fed the system 2000 hammies and 2000 spammies that I had lying about in the backups of the mailserver. All sorts of timestamps, since beginning of this year. Yes, checked and double checked them all. Boring! Everything seems to be working very nice. All the rules, Bayes, everything but one detail: AWL. As I understand it AWl averages scores. So if a high number of hammies have been received from an address, and for once a possible spammy comes from that address, SA gives it the benefit of the doubt, having AWL subtracting some points. And if a notorious spam sending address for once seems to send some ham AWL will remember the spam and add some points anyway, just in case. (Please correct me if I'm wrong) What I observe are very wrong AWL score, AWL being almost consistently the opposite of what I would expect. The net result is that e.g. Bayes is almost leveled by AWL. See the example below, from a very hammy message: BAYES_01 (-5.4 points) BODY: Bayesian classifier says spam probability is 1 to 10% AWL (3.0 points) AWL: Auto-whitelist adjustment Or from a spammy message: BAYES_70 (2.3 points) BODY: Bayesian classifier says spam probability is 70 to 80% AWL (-1.8 points) AWL: Auto-whitelist adjustment I understand that Bayes and AWL are independent and don't rely on the same corpus. I just contrast the two to demonstrate that where Bayes comes up with very reliable scores, AWL spoils the picture. In the present sistuation I would get a better net result if I would disable AWL. But that is not the intention. The idea behind AWL seems to be a sound one. What am I missing? Perhaps someone can point me into the right direction. Thanks in advance. Sietze Dijkstra ------------------------------------------------------- This SF.Net email sponsored by: Free pre-built ASP.NET sites including Data Reports, E-commerce, Portals, and Forums are available now. Download today and enter to win an XBOX or Visual Studio .NET. http://aspnet.click-url.com/go/psa00100006ave/direct;at.asp_061203_01/01 _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
