Jean-Paul Natola wrote:
> If I set the use_bayes 0
>
> Will that negate the training of spam?
> =20

this will disable the use of bayes completely. After setting that, no=20
BAYES_* rules should match any email.
>
> I've trained sa with approx 7000 spam messages , yet I still get the
>
> -2.6 BAYES_00 =20
> =20

Interesting. How are you gathering spam for training? One thing to know =

is it's fairly important that the message match the original, including=20
headers. Most forms of forwarding, and some clients (ie: eudora,=20
outlook, etc) irreparably mangle the message. Usually this is done in=20
order to save space by stripping out parts that the client won't ever=20
use in rendering, etc. However, what's left doesn't always resemble the=20
original spam.
------------------------------------------------------------

I have users DRAG not FORWARD them to a public folder, all headers are =
kept,
I even see after running sa-learn xxx tokens learned from xxx messages..

But for some reason A LOT of spams get the -2.6 and subsequently making =
to
the inbox