...

... (http://dcc.rhyolite.com/) ...

Indeed, that is a cool idea. I definitely want to look into that a lot more closely. Perhaps we can combine this with deep blacklist checking (beyond just the first hop), tagging, and Bayesian content filtering. Perhaps then we will have a temporary pass at a semi-decent anti-spam filter.

be careful when you gang things together. spamassassin seems to be a considered approach, but that doesn't mean more is always better. as has oft been said of PRNG's, adding complexity usually subtracts from the quality. if you combine too many kinds of spam filtering together then you'll have that much more trouble figuring out what to tune when you get a false positive.