On 9/24/2014 7:02 PM, Corey Hickey wrote:
Hi,
Lately I have been getting lots of spam that passes through when
initially received, but which is detected as spam when I test it later.
I guess the blacklists catch up to the spammers' new IPs, etc.
I want to write a script that re-checks recent unread messages in order
to catch this situation after the fact. One concern is that I am running
spamassassin with autolearning and AWL enabled, and I don't want my
checking script to have any effect on those databases.
Why not?
The original (incorrect) score had an affect on AWL. Why not let the
corrected score also have an affect?
The same is true of Bayes. If the original score was low enough to
learn the message as ham, you definitely want to re-learn it as spam.
Bayes, in particular, should not be a problem. Since it tracks the
messages, you can re-learn the same message as many times as you want
without causing any problems. It will either skip the message since it
was already learned, or re-learn it as spam if the score is high enough
on your re-scan.
--
Bowie
