On Tue, 25 Sep 2007, feral wrote:

> Whatever the case, global bayes or not, or even bayes or not, how
> could an email with the obvious porn words in the subject (as in
> my examples) NOT get flagged?

If bayes was mistrained to consider such words hammy, then BAYES_00
could drag the score back down below the threshold, cancelling out the
points added by HOT_NASTY and PORN_16.

One response would be to make the HOT_NASTY and PORN_16 rules "poison
pills" by raising their scores well above the threshold (i.e. to 20 or
30 or even 100) - but you would have to *really trust* those rules to
do that.

And I note that those rules didn't even hit on your first two 
examples.

Both of the domains in those spams are listed in SURBL (but may not 
have been at the time you received them). URIBL network tests probably 
would have hit.

So it looks to me like two major problems are present:

1) mistrained bayes

2) no network tests occurring (DNS RBLs, URI BLs, razor, etc.)

And possibly:

3) not enough rules - add some from SARE? 
http://www.rulesemporium.com

--
 John Hardin KA7OHZ                    http://www.impsec.org/~jhardin/
 [EMAIL PROTECTED]    FALaholic #11174     pgpk -a [EMAIL PROTECTED]
 key: 0xB8732E79 -- 2D8C 34F4 6411 F507 136C  AF76 D822 E6E6 B873 2E79
-----------------------------------------------------------------------
  Pelley: Will you pledge not to test a nuclear weapon?
  Ahmadeinejad: CIA! Secret prison in Europe! Abu Ghraib!
                   -- Mahmoud Ahmadeinejad clumsily dodges a question
                                    (60 minutes interview, 9/20/2007)
-----------------------------------------------------------------------
 243 days until the Mars Phoenix lander arrives at Mars




Reply via email to