force autolearn

[Matt Kettler]

Dave wrote:
> In the future, I'll set up the MTA to not process mail that doesn't
> have an end user. In the interim, is the original question possible?
> Force autolearn using a custom rule?

As I said before, custom rules are factored into the autolearning, and are
treated no differently that the rules that come with SA.

However, you will need at least TWO rules to force autolearning as spam. One
rule alone cannot force autolearning, no matter how high the score of that rule is.

In order to learn as spam, a message must have at least 3.0 worth of points from
header rules, AND 3.0 worth of points from body rules.  This is a hard-coded
requirement that exists regardless of what your autolearn threshold is.


You also can't force autolearn to learn anything that would have scored very low
on the BAYES scale to begin with. ie: regardless of score, and no matter how
many rules fire, the autolearner will not learn as spam anything that would have
hit BAYES_00 or BAYES_05.

(note: I'm assuming SA 3.1.0 here.. the exact sets of rules that cause this
exemption have changed over time because it's based on the score of the bayes
rules being less than -1.0. )

In your current situation, these features might seem like a pain, but they're
there for a reason. They're all there as safety nets to reduce the chance of the
autolearner polluting the bayes database if one rule starts false-firing.