´Rule possible to autolearn 'really local mail' as ham?

["Chr. v. Stuckrad" <st...@mi.fu-berlin.de>]

Hi!

Sorry, its me again, bayes learning seems again 'biased',
this time to spam, may be by picture spams?

After installing new modules to catch picture spams, the
scores of those seem to go high enough to autolearn those
included random texts.  So I seem to be in a fix, eighter
learning 'everything is spam' (because random tokens hit
randomly -> everything) or switching off 'bayes'(learning)
and letting through a lot of Spam.
Spammers would love both, my users will complain.

I have so far no chance to get enough spam+ham from the users
to 'hand-train instead of autolern'.  (Not allowd to read userboxes,
no time to coordinate and crosscheck extra ham-folders of users).

Is there a chance to create a rule to autolearn mail which is surely
local (and not forwarded by local users, lest they forward spam).

I looked into the rules-files and found lots of 'relay-checks', but
could not find out, how to adapt one to this idea.  I could add
a negative score to mails which contain anly out mailhub's headers.

Pointers, Examples, any Help welcome!
   Yours  Stucki, postmaster at math/inf/mi.fu-berlin.de

-- 
Christoph von Stuckrad      * * |nickname |<st...@mi.fu-berlin.de>   \
Freie Universitaet Berlin   |/_*|'stucki' |Tel(days):+49 30 838-5 57 78|
Mathematik & Informatik EDV |\ *|if online|Tel(else):+49 30 77 39 66 00|
Arnimallee 6 / 14195 Berlin * * |on IRCnet|Fax(alle):+49 30 838-75 454/