I'm not really sure why these mails seem to not get hit
I started running 2.61 on my real mail feed just about a week ago, initially training Bayes on the most recent 2000 spam and about 500 ham that I had received. I also have been learning every piece of mail that comes in as either spam or ham. Since then it has scored BAYES_99 on every one of the "poisoned" spam with random words after the /HTML tag.
Could there be something about the way you train or use Bayes that is causing it to miss on these "poisoned" spam? Why would we see such different results?
-- sidney
