> This is what I do from .procmailrc. I have autolearn turned off.
>
> I have a Junk Maildir and Junk/Bad Maildir nested inside of Junk. I
> never verify things in Bad, and verify everything in Junk. I can
> manually learn non-Bad things in Junk by dragging to one of my Auto
> folders. I have Auto/Ham, Auto/HamBayes, Auto/Spam, Auto/SpamBayes.
> A cronjob will take things in these dirs and if they are Ham,
> classify them as such and redeliver them to my Inbox, Spam gets
> classified as Spam and moved to a non-mail directory to be archived.
> The "Bayes" folders are for classifying but not scoring (using a
> homegrown stats program... can't have my numbers getting thrown off).
>
> # if spamassassin thinks it's > 10, lets classify it as spam and mark
> bad
> :0
> * ^X-Spam-Checker-Version: SpamAssassin .* on .*\.myhost\.com
> * ^X-Spam-Level: \*\*\*\*\*\*\*\*\*\*
> {
> # learn as spam if > 10 and not image spam
> :0c
> * ^X-Spam-Status: .*autolearn=disabled.*
> * !^Content-Type:.*(multipart/related).*
> | $HOME/local/bin/sa-learn --spam --no-sync
>
> # score as successful spam catch
> :0ci
> | $HOME/bin/sa_score.pl -s
>
> # put in Bad Junk dir.
> :0
> /home/bdwilson/Maildir/.Junk.Bad/
> }
>
> # if < 10 in SA score, don't bayes learn and only put in Junk.
> # user can drag to Auto/SpamBayes if it is truly spam to get added to
> # bayes without affecting sa_score stats.
> :0
> * ^X-Spam-Checker-Version: SpamAssassin .* on .*\.myhost\.com
> * ^X-Spam-Status: Yes
> {
> # score as spam catch
> :0ci
> | $HOME/bin/sa_score.pl -s
>
> # only put in Junk
> :0
> /home/bdwilson/Maildir/.Junk/
> }
>
> # everything else I manually mark as ham.
>
All that is ok for end user training bayes
But an ISP with 7Million mails a day for ~1000 domains cannot do that.
Autolearn is the option I can see.
What I wanted to know was Can I do an optional autolearn
