Michael wrote: > Using sql based AWL, this target: > select * from awl where email like "%peppers%"; > +----------+-----------------------------------+--------+-------+------- > ---+ > | username | email | ip | count | > totscore | > +----------+-----------------------------------+--------+-------+------- > ---+ > | amavis | [EMAIL PROTECTED] | 216.17 | 129 | > 475.064 |
> I would assume the AWL score would be 3.68. > I get AWL score of: -0.961: > (yes, its amavisd-new) Any idea why? And if this is the right score? > X-Spam-Status: No, score=4.629 tagged_above=-999 required=5 > tests=[AWL=-0.961, BAYES_95=3, BR_SPAMMER_URI=2, HTML_MESSAGE=0.001, > SARE_UNI=0.591, SPF_HELO_PASS=-0.001, SPF_PASS=-0.001] Let's see if I understand this myself. The 3.68 is only one part of the formula that calculated the final AWL score for this particular message. http://wiki.apache.org/spamassassin/AutoWhitelist http://spamassassin.apache.org/full/3.1.x/doc/Mail_SpamAssassin_Plugin_AWL.html This message without AWL was over 3.68 (what mail from this sender scores at on average)(5.59), so the intent of AWL is to nudge the score halfway between what the sender usually scores at, and what this particular message scored at. Roughly: (mean - score) * factor AWL = (3.68 - 5.59) * .5 auto_whitelist_factor n (default: 0.5, range [0..1]) Gary V
