While the tokens in my incoming mails have very different probabilities, the final KS_BL_PROB is almost always either 0.000 or 1.000. There are very few mails with a value between these. With this I think setting KS_BAYESIAN_BOUNDARY is useless, isn't it?