Matt Garretson wrote:
Despite conventional wisdom to the contrary, I have been training Bayes on bounces (both spam and ham) for years with at least semi-decent results when it comes to backscatter. That'd be one potential way to get at the original content (when it's available). But I'd advise against doing it blindly.
*nod* I've been doing the same for quite a while; on a much smaller mail system (~500 accounts at peak IIRC, SA2.63) I was seeing pretty good results for a while. Unfortunately I'm still experimenting with the Bayes settings on the much larger mail system, and a lot of reported missed-spam (NDR or otherwise) is found to have hit BAYES_50 originally.
-kgd
