Aggregation is in my opinion a must. Blacklists alone will miss 40% of the spam we see. Requirements for passing "authentication only" are easily met by spammers. Purely statistical methods do not achieve 100% accuracy because the human judgement and research currently done by the most trusted black and white list operators is needed as an adjunct.
   more »

My favored predictive collectors focus on factors that can be measured before the message is accepted, even before the first spam from a new run is accepted. I also like predictive collectors that focus on positive or "good" characteristics rather than characteristics of "bad" senders.   more »