From: "Steven Dickenson" <[EMAIL PROTECTED]>
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On Jul 5, 2005, at 8:37 AM, Jerry Van Brimmer wrote:
> >> I also set my bayes to score 5 points for a 99 score, that gets
> >> the stock
> >> scam guys who think they are just too slick by using a different
> >> clean IP
> >> every time.
> >
> > I'm just learning how to use SA. Would you please tell me How to
> > set this option?
>
> It's not a wise idea to set any rule to score this high, particularly
> something as variable as a Bayes rule. In general, it's considered
> bad form for any one rule to score a message as spam by itself.
>
> With that being said, you're going to want to add a SCORE entry in
> your local.cf. This page should help you get up to speed with
> writing SA rules (or just re-scoring existing ones).
>
> http://wiki.apache.org/spamassassin/WritingRules
>
> Steven
> - ---
> Steven Dickenson <[EMAIL PROTECTED]>
> http://www.mrchuckles.net
OK OK - I made it 4.999 just for you. {^_-}
If a rule has not shot me false alarms in a VERY long time and it has
caught spams nothing else seemed to touch I like to give it a nice solid
score. I don't believe fighting by "Roberts Rules" or even "Marquis of
Queensbury" rules makes sense if things are bad enough to make fighting
a good idea. But in deference to the one rule bigots out there I just
lowered it a tenth of a point. But if I see that on ANYTHING that has
no other markups it goes back to 5.0. Otherwise, what is the point of
Bayes if you are not going to believe it?
I prefer to see how high I can get scores to go with collections of rules
that show very low false alarm rates when used. But then I also like nice
"sure thing" rules such as my addresses but at mail.earthlink.com. And
another good one for awhile was "X-Mailer =~ /MOM Agent/i". I never
received a real mailing from it. But for a couple months it sent me a
lot of spam. Why cripple yourself when you have a working killer stroke?
(That said, the rules for Bayes need bayesic reworking. There are two
modes in which it works, per user or global. In per user mode with some
decent manual training it is quite good. In global with hundreds of
users and "autolearn" it is probably at best an 80% thing. Some personal
for the per user with no autolearn is quite worthwhile.)
{^_^} <- Joanne believes in the "largest weapon that has zero collateral
damage as she can bring to bear. 'Fair' is for losers.")
