Hello John, Wednesday, January 12, 2005, 4:54:24 AM, you wrote:
JF> I've never seen this before - I don't know squat about AWL, but I JF> sure need help understanding the following headers! JF> X-Spam-Status: Yes, hits=26.1 required=5.0 JF> tests=AWL,BAYES_00,SARE_HEAD_XBEEN autolearn=ham version=2.64 JF> X-Spam-Report: JF> * -4.9 BAYES_00 BODY: Bayesian spam probability is 0 to 1% JF> * [score: 0.0000] JF> * -0.6 SARE_HEAD_XBEEN Mailng list header found, frequent ham sign JF> * 32 AWL AWL: Auto-whitelist adjustment JF> 1. If hits = 26, why was it auto-learned as ham????? I didn't JF> specifically manually white_list the sender. auto-whitelist (more accurately: auto-sender-adjust) is not manual, so it's nothing YOU have done. See http://wiki.apache.org/spamassassin/AutoWhitelist This score suggests that emails from this source in the past have scored quite high, and so it raises the score on this email based on that history. Neither AWL nor Bayes determines the autolearn. Therefore the only rule that applied was SARE_HEAD_XBEEN, and that flagged the email as ham. JF> P.S. The msg is NOT spam in general. The subject matter of the JF> mail is stuff I've learned many times. The sender might be listed JF> somewhere bad, I don't know him. Please step me through the JF> headers. Tnx So that means apparently that both Bayes and SARE_HEAD_XBEEN were correct this time, and the email was correctly auto-learned. The sender likely sent you a goodly amount of spam in the past, and will possibly do so in the future. That sender's AWL on your system has been lowered somewhat, but not horribly (the next AWL should still score well over 5), and if that sender continues sending spam, the AWL will climb again. Well, to revisit: > and the email was correctly auto-learned A copy of a perfect ham email sent UBE to you, is spam, even if the content isn't. In this case Bayes and SARE_HEAD_XBEEN were correct, in that everything but the sender indicated it was ham, and AWL was also correct, in that the sender suggested strongly it was spam. Bayes in the current version will not autolearn against itself (will not auto-learn as ham something it thought was spam, or v.v.) -- it might be a good enhancement to also have bayes look at AWL if active, and if AWL disagrees with the auto-learn judgment, then do not auto-learn. Looking at http://bugzilla.spamassassin.org/show_bug.cgi?id=3418, maybe a tflags entry for nolearnconflict (do not conflict with auto-learn) could be used for any given rule, indicating that if this rule indicates spam (or ham), then do NOT auto-learn as ham (or spam) because of this "difference of opinion." It would be set on by default for Bayes itself, IMO AWL, and IMO blacklists and whitelists, and we could turn it on/off for these rules and any others as desired for our sites and users. Bob Menschel
