First off, I have to say I really like the antispam plugin. I am
running a (relatively) new freebsd 8.0 server with
postfix/dovecot/dspam/squirrelmail and it works great. Create a SPAM
folder and tell the dovecot antispam plugin to use INBOX.SPAM as the
spam folder and drag messages from INBOX to the spam folder and they are
trained as spam. Here is the problem though: after training dspam for
a couple of weeks, I had a couple hundred messages in the SPAM folder,
so I selected them and hit the DELETE button and they were gone. Except
they weren't. I had forgotten that by default squirrelmail moves
deleted messages to a Trash folder under INBOX. So the messages were
moved out of the SPAM folder, resulting in them being retrained as ham.
Oops :( I then hit the purge link in squirrelmail and the messages were
forever gone, so I couldn't rectify my gaffe. I have temporarily
reconfigured squirrelmail to delete messages outright rather than save
them in the Trash folder, but I don't like that as a long-term
solution. What occurred to me is this: the current algorithm is: if
message moves from * => SPAM, it is trained as spam, if it moves from
SPAM => *, it is trained as ham. It would be nice if you could
elaborate on that such that the ham rule was "if message moves from SPAM
to X, it is ham" (and in my case, I would defined X as INBOX. Thoughts?