From: Bill Wohler <wohler@newt.com>
Subject: spamassassin: Bayesian documentation weak
Date: Mon, 17 Mar 2003 16:57:01 -0800

Package: spamassassin
Version: 2.50-1
Severity: normal

  I just submitted the following to debian-user, but I thought the
  upstream spamassassin folks might be interesting in getting this
  feedback too:

  Compare the spamassassin Bayesian documentation with that
  documentation in the spamprobe and bogofilter man pages.

				 -----

  It wasn't clear to me from the man page that you could build a "good"
  word lists as well as a "spam" word list. The man page implies that
  only a spam word list would get generated with -r and these words
  would be removed with -k, which doesn't work at all. Nor did the man
  page suggest how to prime the database with "good" and "bad" messages.
  I was leery of the -k option since I don't want my personal mail going
  to the Razor (although I think the --local option could be used to
  suppress that). The man page didn't mention whether the database was
  updated every time spamassassin was called or not.

  Thus, the man page raises more questions than it answers and implies a
  suspect implementation.