Modifiable mbox?

[Geoff Kuenning]

I procmail presumed spam into an mbox file, and then periodically scan
it for ham.  Right now, I use rmail to move the ham elsewhere and
delete it from the mbox; then I convert it from Babyl back to mbox and
feed it to spamassassin for Bayesian learning.

The problem with that method is that rmail isn't my favorite
interface, and in fact it breaks after 10,000 messages.  I'd prefer to
use gnus to do my filtering and processing.  The problem is that I
can't figure out how to make gnus descend into an mbox in a writeable
fashion.  gnus-group-make-doc-group is great for browsing the group,
but doesn't let me delete ham after I've written it to a different file.

Is it possible to get gnus to descend into an mbox without
reformatting it into nnml, but still being able to delete messages?
I'd rather not move thousands of messages into separate files, since
then I'd just have to collect everything together again to feed it to
Spamassassin.

I'm also open to alternative suggestions.  My requirements are that I
need to save semi-spam in a procmail-friendly way, use gnus to move
the few real ham messages to a different place, and easily feed the
remaining stuff to spamassassin.  Other than that, I'm willing to
change how I do things.
-- 
    Geoff Kuenning   geoff@cs.hmc.edu   http://www.cs.hmc.edu/~geoff/

If you can't measure it, it's not science.
        -- Robert A. Heinlein, "The Door Into Summer"