If I get any false positives or slips that aren't obvious bayes problems (like random words or long strings of non-words etc) I run sa-learn on a pine created ham and spam box using the following commands:

sa-learn --showdots --mbox --ham Ham sa-learn --showdots --mbox --spam Spam

Every time I do that it reports an extra message "learned from" than was in the box. I'm assuming that this is that pseudo message at the top of every pine mailbox. Is that going to cause problems with the bayes databases?