[Mimedefang] MimeDefang and SpamAssassin

alan premselaar alien at 12inch.com
Mon May 19 10:02:01 EDT 2003


On 5/19/03 10:45 PM, "VCI Help Desk" <admin at vci.net> wrote:

>   Ok, I added these entries into my "border mail servers" as the older
> message that said Bayes was disabled described and have primed it with
> sa-learn. In my /var/spool/MIMEDefang/bayes/ folder I have 3 bayes_ files
> that are being updated after an email arrives. The concern I have is that I
> don't see any BAYES_* entries in ANY of the emails that pass into my system.
> I have mimedefang adding a header that resembles the following. None of the
> rules over the past 3 days have ever said anything about BAYES_*   Shouldn't
> there be a BAYES rule added in every message?
> 

the following BAYES tests have 0 scores and therefore won't return in your
list of tests:

50_scores.cf:score BAYES_40 0.0
50_scores.cf:score BAYES_44 0.0
50_scores.cf:score BAYES_50 0.0
50_scores.cf:score BAYES_56 0.0

so basically if bayes is on the fence about the spaminess of an email, it
won't score it.  also, you have to make sure you have at least 200 spam and
200 ham learned before the bayes filters will become active.

>   X-Spam-Level: * 1.5 HTML_10_20,HTML_MESSAGE
> 
>   As another post mentioned I'm using a site wide installation that
> probably makes bayes testing useless. I'm trying to block more spam because
> over the past 10-15 days the amount of spam has gone up a LOT. A lot of the
> spam emails that come in seem to have one big picture and little text in
> them. If all spam emails start doing that they we may as well start
> converting our spam reduction services to services that block all emails
> except those from people we specifically authorize - which I've considered
> also.

I'm using bayes site-wide on my personal domain as well as my office domain.
My personal domain doesn't have many users, so perhaps bayes is more
effective there. My office domain has relatively few users (compared to a
lot of places) at approx. 150 unique email accounts.  I've just recently
turned on the bayes auto-learning, although it hasn't kicked in yet.

my take on the whole site-wide bayes thing is, there are some common spam
techniques that may be caught by bayes filters yet too complicated or
inconsistant to be caught by rules alone. every little bit helps, no?

this may be a topic you'll want to discuss on the SpamAssassin (SATalk) list
however. 

hope this helps,

alan




More information about the MIMEDefang mailing list