Matt Kettler <[EMAIL PROTECTED]> writes: > Dan.. AFAIK that's an outright bug in spamassassin. One of the rules of > autolearning introduced in the 2.61 series is to never autolearn anything > that strongly contradicts existing learning (ie: never learn a message as > ham that would be bayes_99)...
I'm not sure we really fixed it to begin with. > Did that intentional feature somehow get reverted out of 3.x? > > http://bugzilla.spamassassin.org/show_bug.cgi?id=2437 *shrug* Daniel -- Daniel Quinlan http://www.pathname.com/~quinlan/
