[Spambayes] SpamBayes as a gatway solution

[Richard Jowsey]

> I'm wondering if anyone has some opinions about whether this is
> absolutely set in stone, or whether there might be some good ways to
> implement SpamBayes at the gateway level.

It's not SpamBayes, but I've successfully implemented a Bayesian 
gateway filter (POP/SMTP) which utilizes both "system-wide" and 
"personal" data. The logic for assigning spam probabilities in this 
scenario has required a considerable amount of R&D, experimentation, 
testing and measurement, plus some fairly tricky memory and database 
management. However, it's now working very well, on our public proxy 
servers and at a number of ISP/corporate clients. AFAIK, it's 
currently the only commercial-grade Bayesian implementation which 
does this.

Just FYI...

Cheers,
Richard Jowsey