On Tue, 2005-08-16 at 16:39 -0700, jdow wrote:
> On the whole these are not very high scores. SORBS is not regarded as
> being all that reliable. So far this week SORBS_DUL is triggering on
> only 22% of spam and is triggering on 0.29% of ham. The other SORBS
> rules aren't even in my top 100 rules.
Unfortunately SpamAssassin's default scores are absolutely meaningless
in this regard. The people behind it are good coders but have shown
little understanding of statistics and abuse the scoring system as a
catch all facility.
What I am trying to say is, that for example about 90% of the messages
from a dynamic IP are spam which in turn means that each message
matching this criteria has a probability of about 90% of being spam.
But instead of assigning a score of about 4 to messages coming from a
dynamic IP a separate score is used for each DUL list. The genetic
algorithm tries to compensate by distributing the total probability
among these lists according to their completeness which means, that a
message only gets the score it deserves if the sender is on all lists.
Tom
--
T h o m a s Z e h e t b a u e r ( TZ251 )
PGP encrypted mail preferred - KeyID 96FFCB89
finger thomasz@xxxxxxxxxxxxxx for key
Chemists don't die, they just stop to react.
Attachment:
signature.asc
Description: This is a digitally signed message part
