> X-Spam-Status: No, score=3.964 tagged_above=-100 required=5 > tests=[BAYES_99=3.5, HTML_30_40=0.463, HTML_MESSAGE=0.001]
To me, it looks like enough tokens were seen to flag it as BAYES_99, but that the host and IP it came from didn't trigger any RBL hits, which left your point score well under your 5.0 tolerance.
I can't speak for others, but I personally set my 'required' level to 3.5, so anything SpamAssassin flags as BAYES_99 automatically gets flagged as spam. And since I use {SPAM _SCORE(0)_} as a subject rewrite, I can catch any messages that score over another threshold (for me, it's 10.0) and filter them to /dev/null if they score that high -- the only things that score that high are messages with bayes scores of 80-100 plus RBL hits.
Just my $0.02.
