wayne <wayne(_at_)midwestcs(_dot_)com> wrote:
In Spamassassin, the more postivie the score, the more spammy the
message was found.
I understand. My comments were about other anti-spam systems, too.
The intention was to describe the "spamminess" of a message, so
positive scores are positive indications of spam. I should probably
have used the word "benefit" rather than "score", but I can't keep
track of which system uses what vocabulary.
For implementation reasons, Spamassassin has chosen to invert the
"score", so "ham" is given negative scores. It amounts to the same
thing, with different vocabulary.
Stephane explained *how* Spamassassin comes up with their numbers, I
just mentioned the result.
My point then, as now, is that passing SPF or MARID checks means
*nothing* about the "spam" or "ham" nature of a message. Any
anti-spam system should give *zero* additional benefit to messages
which pass those checks. If those systems choose to implement zero as
a small negative number, that's their business.
Having spamassassin use a neural net to determine the score for
SPF_PASS is wrong. We *know* what the value is, from an analysis of
As others have pointed out, this has no real effect, it only means
that the results gets put into the headers.
That's an implementation decision for Spamassassin. My comments
about scoring were completely independent of reporting of that score.