mime.c

[Matthias Andree]

Graham Wilson <bob at decoy.wox.org> writes:

> spamassasin adds keywords to its X-Spam-Status header that indicate
> certain features of the message (no subject, subject all caps, etc.).
> these are then used by bogofilter as tokens and contribute to the
> spamicity of a message.

I'd rather put bogofilter before spamassassin, because it's so much
faster; and if spamassassin is there, why should I parse the traits it
digged out with bogofilter when spamassassin's scoring scheme is so much
more effective?

Still unaware of the algorithms we use, it looks as though we're still
at "we check for every individual token how likely it is to be spam and
sum up the probabilities". David's objection "we'd need lots of the
tokens to have noticable effect" stands.

-- 
Matthias Andree