Selectively bypassing Bayes

[Steve Freegard]

Kevin Miller wrote:
> I'm fighting a lot of joe-job blowback from Russia (still), and have been posting to the spamassassin list for some help.  I enabled/configured the VBounce ruleset but it was suggested to me that I use that as a trigger to quarantine the messages but to not feed them to Bayes, since it could skew the results of legitimate NDRs.  It was suggested that I use procmail to pre-sort the messages.  My MailScanner boxes are just gateways, forwarding to an internal Exchange server and doesn't have any local mailboxes so I don't think procmail is the right tool.  (If I'm wrong let me know.)
> 
> I have a custom rule that examines the headers for .ru and ANY_BOUNCE_MESSAGE which comes from the VBounce ruleset.  How to I tell MailScanner and/or spamassassin not to feed those messages to Bayes but still block them?  Or can I even do that?
> 

Add:

tflags CUSTOM_RULE_NAME noautolearn

after your rule definition; it will not prevent auto-learning but it
will not include the score of your custom rule when calculating the
auto-learn thresholds.

See man Mail::SpamAssassin::Conf