BAYES_00 issue?

["Clay T. Davis" <cl...@cville.net>]

My ISP is almost always allowing messages that have fired the BAYES_00
(scored with -2.60).  These are obvious spam messages, some with porn words
in the subject line others are degree and money scams.  How should I
approach trying to get them to check their BAYES DB for a problem?  The
following score is a fair example of one of the more recent arrivals that
would have been zapped if not for the BAYES_00
 
score=2.695, required 5, BAYES_00 -2.60, DCC_CHECK 2.17,
 DIGEST_MULTIPLE 0.00, RAZOR2_CF_RANGE_51_100 0.50,
 RAZOR2_CF_RANGE_E4_51_100 1.50, RAZOR2_CHECK 0.50,
 SARE_SUB_MONEY 0.62)

Am I missing something?  I can provide the message if you want to run it
through your system as a test.
 
Thanks for the help.
 
Clay

Re: BAYES_00 issue?

[Matt Kettler <mk...@verizon.net>]

Clay T. Davis wrote:
> My ISP is almost always allowing messages that have fired the BAYES_00 
> (scored with -2.60).  These are obvious spam messages, some with porn 
> words in the subject line others are degree and money scams.  How 
> should I approach trying to get them to check their BAYES DB for a 
> problem?  The following score is a fair example of one of the more 
> recent arrivals that would have been zapped if not for the BAYES_00
>  
> score=2.695, required 5, BAYES_00 -2.60, DCC_CHECK 2.17,
>  DIGEST_MULTIPLE 0.00, RAZOR2_CF_RANGE_51_100 0.50,
>  RAZOR2_CF_RANGE_E4_51_100 1.50, RAZOR2_CHECK 0.50,
>  SARE_SUB_MONEY 0.62)
> Am I missing something?
Probably some bad training or autolearning somewhere.

If you want to see what words it was using, try running the message 
through "spamassassin -D bayes". The debug output should have each token 
it found, and what probability it had..

You can try to correct it by feeding several of the messages to sa-learn 
--spam. However, you might want to check if any of your spam emails are 
getting missed and autolearned as nonspam.

>   I can provide the message if you want to run it through your system 
> as a test.
That's not going to generate the same results, as my system has my 
training, not your training..