[Spambayes] How are good messages treated?

[Meyer, Tony]

> OK, here's a dumb question--when SpamBayes (or SpamAtBay)
> correctly leaves a ham message in the inbox, and if I do 
> nothing to that message (except read and delete it), does SB 
> add that message's words/tokens to the ham database?

No.

> To phrase it another way, are the only ham words added to the
> database the ones from messages that I manually indicate as ham?

Yes.

Note that the same applies to messages classified as spam.  No matter
what the classification, incoming messages are not used for training.
The only way the message is trained is by using the "delete as spam" or
"recover from spam" buttons, or via the training dialog.

Note that this follows the 'train on unsures' training regime - i.e.
after setting up an initial database, the only training that is done is
on messages incorrectly classified or classified as unsure.  Seems to
work well for most people (but then, so do other regimes!).

(I had thought that a message was also trained if it was moved into a
folder being watched, but that doesn't seem to be the case from the
trace).

=Tony Meyer