Virtual mailuser with their own bayes db?

[Matt Kettler]

Julian Field wrote:
> MailScanner does not use spamd/spamc at all, it talks directly to 
> SpamAssassin's Perl library for speed and efficiency reasons.
> 
> However, this does mean that you are stuck to 1 bayes db for the whole 
> system, sorry. People such as Matt Kettler (Matt--correct me if I'm 
> wrong) have constructed very reliable spam detection without using bayes 
> at all, so this isn't actually a big problem. 

I use bayes, quite extensively. I think bayes is one of the most powerful and
useful tools in SA.

However, in a corporate environment, I find that a single-site-wide bayes DB
actually works better than individual bayes. Really, individual DBs is only
worthwhile if your users are highly diverse, such as at an ISP.

I'm also one of the proponents of the "hamtrap/spamtrap" automated training
technique.

a "hamtrap" is a secret, obscure email address that you subscribe to trusted
email sources, (industry newsletters, etc) you then use a cron-job to process
its mail with sa-learn --ham.

A spamtrap is a not-so-secret email address that you seed out to the world in
innocuous ways and then configure them to be trained as spam. I generally do
this while posting to various technical mailing lists. Whenever an example needs
an email address, I insert a bogus one at my domain. I leave the account as
nonexistent for about a month, and then create an alias for it so it gets trained.