Theo Van Dinter wrote: > On Tue, Oct 14, 2008 at 09:32:42PM -0400, Matt Kettler wrote: > >> Agreed. However, there are times where it's not worth fixing a FP case >> because its a rare case caused by a small-scale broken tool that >> violates the specs. >> > > FWIW, this case is specifically looked for by the rule. > > In general, I'm not advocating dropping this aspect of the rule, just moving it to a separate test where it can stand alone and be scored on its own merits. >> Of course, I lack the time right now to write it, so unless Theo or >> someone else wants to jump on it, I'm sure patches are welcome. >> > > I'm positive we've gone through this before, but I lack the time to go > searching for the discussion/bugs. One that came up in a quick bugzilla > search: > > https://issues.apache.org/SpamAssassin/show_bug.cgi?id=4234 > > The results at the time were that spam hits went down by so much, the rule > became worse. Score generation is expected to deal with the FPs. > >
Well, that alone sounds like a good reason to split the rule. Those results suggest that SINGLEPART_MULTIPART would be a better performing rule, and probably deserves to be singled out so it can be scored higher. Of course, the split would have the opposite affect of what the poster desires, but it would be a good, conclusive point to stand on. This test shows that this exact property occurs rarely in nonspam, and frequently in spam. It also would allow the score to readily adapt to changes in the FP ratio.
