If I run a message through sa-learn and then run spamassassin -t on the same message, shouldn't it be recognized as SPAM ?
If I add a teaspoon of red wine to a 50 gallon barrel of clear water, should the resulting liquid be red?
Depends on the contents of your bayes DB.. if that particular spam mail had mostly well-recognized HAM tokens with very strong ham scores, a single message isn't going to change much.
However, if the spam has mostly unrecognized tokens that didn't exist before, a single message training is going to make a dramatic difference.
Remember, bayes is all about statistics and spam probabilities of the words and headers that make up a message, it's not about message itself.
You can run the message through spamassassin -tD and see what the bayes tokens in it are, and what their probabilities are.
