On Tue, 18 Oct 2011 13:07:21 -0700 (PDT) Mynabbler wrote: > > > RW-15 wrote: > > > > It would hit: > > Re: Did you pick-up the dry-cleaning? > > > Nope. Scores just two (one ':' and a '?') and the rule needs three > different odd characters.
OK the font I'm using makes ~ look very like a -, but the point remains. If a subject starts with FW: or Re: and has a [!?], which is pretty common, you are then triggering on only one extra character. If you look back through this list you will find numerous such replies. > RW-15 wrote: > > > > I think it needs more work, maybe combine it with tests for lots of > > very short words or adjacent punctuation pairs. > > > As I explained, even if the rule would have fired, it adds a whopping > 0.1 score. It only shows teeth when combined with other findings... So, why isn't it worth scoring if it's a useful rule? And why score it so high with FREEMAIL?. The danger here is that you end-up with a lot FREEMAIL && WEAK_RULE metas that are prone to high-scoring FPs that BAYES_00 can't save. If FREEMAIL_FROM is a good indicator then score it up, and score other rules on their merits.
