DM> No, I'm not trying any marriage, nor any realtime lookups,
DM> just using its prewired heuristics (from a .forward file, as I
DM> don't own my mailserver).  It's a "look at the message and try
DM> to intuit if it's spam" solution, so it doesn't use any
DM> whitelisting really (well, there's dynamic whitelisting, in
DM> that some of its heuristics are "have I ever noted non-spam
DM> from this address before?")

DM> I love it so far; it's instantly cut my spam to almost nil
DM> (from ~30-40 a day) with very few errors.

I lost track, SpamAssassin or TDMA?

SA

I read about SA briefly and it looks like you could run it in a client/server arrangement.

Yeah; it runs as a relatively-big Perl wad normally, but for efficiency you can start it as a daemon and then run a small C client that behaves the same way, so that the startup costs are one-time instead of per-mail.

If the on-the-wire protocol were simple enough, you could probably write a pure-Python client and then integrate that into Mailman's handler pipeline. If it's too complex, you might be able to wrap SA's C API in a Python extension module.

Yeah, either should be simple enough to code in Python (it's an HTML-like protocol).

I actually hadn't thought about trying to provide spam filtering for lists; I'm just thrilled with what it's doing for my personal email.