On 08/13/2014 09:37 AM, Axb wrote:
the so called "legit" will set your addr flag as unsubbed
I see a significant amount of "spam" to my users from truly legitimate sources. Where "truly legitimate" doesn't mean that they are legitimately the USDA or Merrill Lynch. These can be fire arms ads from small companies I've never heard of, going to people whom I could already have guessed belonged to gun clubs and probably missed unticking a checkbox somewhere during sign-up.
IMO, Bayes has enough attacks going on against it that we need to give it all the help it can get. And that means that when we tell it something is spam, that something really needs to be spam, by anyone's definition. When a message can't be unsubscribed from, the DNSBL's miss it, and the other rules miss it, I want a Bayes with maximum specificity.
I also up the bayes scores. I believe in Bayes. But "Garbage In, Garbage Out" is particularly appropriate for Bayes' inputs and outputs.
