SA Bugzilla – Bug 2534
AWL applied to identified as spam messages
Last modified: 2007-04-16 05:17:37 UTC
huge amount of spam with same From produce noticeable AWL compensation (and can "clean" spam), if recipient can't modify own prefs by adding "blacklist_from" I think, appling AWL shifting _after_ calclulationg final score only for non-spam messages can be reasonable add-on
One of the main reasons for AWL is to keep messages from regular correspondents from being erroneously marked as spam. When a friend sends you a joke about Viagra that would pass the spam threshold, the score gets reduced because of the friend's long-term average. With your suggested change, that wouldn't happen. Have you read the FAQ at http://spamassassin.taint.org/faq/index.cgi?req=show&file=faq06.002.htp ?
I just ran into this myself - the AWL scanning report is simply unclear and causes confusion. The issue described pops up when you run multiple tests on the same/similar spam. Higher bayes scores, for example, will then cause your current score to be higher than the averaged mean score, which result in a negative adjustment (delta) for the mail being scanned. Which will look like this in the report: -1.9 AWL AWL: From: address is in the auto white-list Leading anybody not clued into the deep mechanics of the AWL into thinking "hey i didn't whitelist those suckers did i?" I suggest changing the report line to something like "AWL: From: Adjustment towards mean score" to limit confusion.
actually, that's a *really* good point that we've missed... All the way along, we've been referring to the AWL as an auto-whitelist; we've been considering how to fix that, but renaming the rule etc. seems like a lot of work as it's pretty pervasive. We don't actually need to change anything except for the rule description: -1.9 AWL AWL: sender historical score adjustment would do it IMO.
move bug to Future milestone (previously set to Future -- I hope)
this is a FAQ: http://wiki.apache.org/spamassassin/AwlWrongWay