Stephane Bortzmeyer <bortzmeyer(_at_)nic(_dot_)fr> wrote:
Such analysis are always dangerous ("We all know that spams are in
text/html only")
That's not a good analogy. We *know* SPF_PASS says nothing about
whether a message is spam or not. We *know* SPF_FAIl says that the
domain publishing SPF has not authorized the message.
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.
If the neural net decides so, fine with me.
Assigning negative spamassassin scores to SPF_PASS means that
spammers can benefit from using SPF. Assigning positive spamassassin
scores to SPF_PASS means that non-spam will get punished for using
SPF.
We want neither to happen, therefore anti-spam systems should assign
no benefit, and no penalty, for publishing MARID-style records. If
you want to allow neural-net systems to assign scores, then you're OK
with either, or both, of the two scenarios from the preceding
paragraph.
Alan DeKok.