Re: Yet another thread about AWL

On 2/21/2012 10:11 AM, Antonio Gutiérrez Mayoral wrote:
> Hi all!
>
> I'm afraid this is yet another thread about Auto white lists module
> :-( Sorry if this question is yet
> solved on the list.
>
> I have a lot of Spam and backscatter throwing high spam score on my
> actual spam assassin system (postfix,
> spamassassin with amavisd-new and Maia). Tracing these messages, I see
> on the amavis log that the
> rule AWL is triggered with a negative score. Reading the documentation
> I think that the problem
> was a wrong auto-learn thresold for HAM, the first week the system
> starts to work. The initial
> thresold for Ham was -0.001 and I think that this thresold causes a
> lot of spam and backscatter addresses
> was learned as non-spam addresses. Could be possible?
>
> Now the ham autolearn thresold is on -5 points. So I guess now its
> auto-learning right... I think.
> But, what can I do with the wrong addresses learned before? Could I
> simply delete them from
> the MySQL database?

You are confusing AWL and Bayes.  The autolearn thresholds apply to the
Bayes DB.  IIRC, Bayes scores do not contribute to the AWL.

A negative AWL score simply means that previous emails by this sender
had a lower score, so AWL generates a negative score to bring the
overall score closer to the previous average.  For a pure spam sender,
this effect should be minimal.

Other than disabling AWL, the only way to fix a score is to remove it
from the database.

$ spamassassin --remove-addr-from-whitelist [EMAIL PROTECTED]

-- 
Bowie