Thanks jdow
Seems like you were right. My debug session was run as an
administrator but needed to be run as root. Now I just got to figure
out how my bayes database could get so sqewed.
I've got a global SPAM and HAM mailbox for our whole domain that our
users drop their false positives and negatives in. I run sa-learn on
them every other day. Since the HAM box hardly get any messages in it
i have used my own inbox, that I know is spam free, to learn as HAM.
Is that a correct way to do it? Could any of the headers that our MTA
is adding get counted as a valid tokens when I run sa-learn --spam?
Thomas Ericsson
31 aug 2006 kl. 04.56 skrev jdow:
From: "Thomas Ericsson" <[EMAIL PROTECTED]>
Hi
I've gotten a lot of low scoring bayes hits, especially BAYES_00,
so I figured the database is off. I ran sa-learn --clear to start
over, but I still get a lot of BAYES_00. I then ran spamassassin
-D --mbox BUNCHOFSPAM.mbox to see what gives. It turns out I get
a different result from when the mail was delivered in the first
place. Could it be that each mail get scanned more than one time?
Any ideas appreciated
Thomas
I would hazard a guess that two different Bayes databases were used,
one for actual reception and the other for your test. You might check
into that possibility. Be sure to run the spamassassin tests as the
same "user" which maintains the Bayes database you are trying to fix.
{^_^}
