From: Tony Houghton on
ISTR reading somewhere that spamassassin's Bayesian learning is
self-reinforcing, ie when it decides that a mail is spam or ham it does
the internal equivalent of feeding it to sa-learn. Is that true, or
would I be better feeding emails back to sa-learn even after it's made
the right decision?

Also, if the report_safe option is 1 (generate a new message with the
spam as an attachment) and I feed a tagged spam to sa-learn, will it
automatically extract the attached spam instead of processing the header
of the wrapper message?

TH *