sa-learn MD filtered mail

[Keith Whyte]

Hi I have a setup with MimeDefang and Spamassassin.

I reject mail scoring over 10 at SMTP level, then i tag the rest with
X-Spam-Level Headers and a
Subject header change - via md's action_change_header(), for spam
scoring over 4.

So now I have a huge junk mail box (procmail filtered) full of spam.

I'm wondering can I feed this to sa-learn.

My doubt is over whether the changed Subject: headers and the
X-Spam-Level header will confuse the bayes system.

for example, a mail in my junk mail box now may have a Subject of:
**SPAM (6.50) this is spam **
where as when it arrives the subject would just be "this is spam"

surely this would confuse bayes, no?

any suggestions and help greatly appreciated, thanks.