Mark E. Mallett <mem@mv.mv.com>:
I'm interested in the topic, perhaps not to wrangle over it but at
least to have a standard way of identifying the filter program(s)
being used. If there's already something in practical use, or some
current thinking on the subject, so much the better. I was thinking
mostly of a meaningless "filtered-by" sort of label, but you mention a
score, for which there might be a more structured header or suite of
headers. Personally I think that with a variety of ways of marking
scores (or in fact even having the concept of a "score"), it would be
hard to specify the a particular structure here. Plus you might have
a message passing through multiple filters, each of which would want
to place its mark on the message without disturbing other such
marks (so you'd need a tuple of at least {agent, location, assessment}).
Right. I ran up against this while designing bogofilter. The syntax you
want is something like this:
* Semicolon-separated fields
* Each field to consist of a agent@hostname:evaluation token
* the evaluation field to consist of comma-separated keyword=value
* The agent field should be versioned, so a real example should look like:
bogofilter-5.23@thyrsus.com:spam-probability=90%,forged_address=yes;
Would that mean that the MUAs have to be matched to the user agent of
the filters of the MTA that they use? What about a spammer forging the
header coming from the filter?