Gustafson, Tim wrote:
The problem with per-user Bayes for me is that I am running SpamAssassin
with spamass-milter, and my users are not savvy enough to be able to
update their own Bayes databases. They don't use any shell accounts -
it's all POP3 and in some cases, after the e-mail passes through my SA
it gets forwarded to their Exchange server.
In that case, I would definately suggest just toning down the scores on
the BAYES rules so that they alone won't make or break an email. If
you are using autolearning, you may want to tweak the autolearn
thresholds as well.
