Is there a good methodology to figuring out appropriate scores? I was thinking of writing a script that would look at the proportion of
ham/spam that this rule applies to, but that gives me a percentage, but not a value.
Don't bother writing your own script for this.. mass-check already comes with SA and you can feed it mailboxes full of spam and nonspam. From there, use hit-frequencies to make a pretty report..
As for score assignment, I personally like the "start low and tune up until it works but doesn't hurt" approach.
