Stephane Bortzmeyer <bortzmeyer(_at_)nic(_dot_)fr> wrote:
Wether MARID encourage or discourages it will make no
difference. Scores of the various tests in SpamAssassin are not
determined by an human trying to guess wether "sex" should be more an
indication of spam than "viagra", or less. Scores are determined by
using the actual tests against a corpus of spam and a corpus of
ham. See http://wiki.apache.org/spamassassin/HowScoresAreAssigned.
I am saying that we know a-priori, that giving positive marks to
messages which pass SPF checks is not a good idea.
Your statement also appears to contradict Wayne's assertion about
Spamassassin.
Do note that, at the present time, the scores of all SPF-related tests
are very low (probably reflecting the small number of SPF-enabled
domains). See http://spamassassin.apache.org/tests.html and search for
SPF.
Which has a "-0.001" for SPF_PASS, or marking the message as somehow
"better" for passing SPF.
While this score is low enough to not effect much, I think it should
be forced to zero, otherwise spammers will *gain* by publishing
MARID-style records. If the neural net method is still used, I'll bet
that the score for SPF_PASS won't get significantly far from zero, for
the reasons I outlined previously.
Passing MARID checks is *no* indication of whether or not the
message is spam. And if, as we've seen, spammers publish lots of
MARID records, then there's a non-zero chance that a neural net may
decide there's a correlation between spam & MARID records, which means
MARID won't be as useful for everyone else.
Alan DeKok.