On Tuesday 02 June 2009, Adam Katz wrote: > Larry Starr <lar...@fullcompass.com> wrote: > >> I have been using the AWL ( --add-addr-to-blacklist ) for some > >> time, to bump new spam senders above the "Bayes-99" score. > > Theo Van Dinter responded: > > Well, the first problem is that the AWL has no impact on Bayes. > > They're totally independent. > > Perhaps you want "sa-learn" ? > > I believe Larry, like myself, trains both Bayes and AWL (I also > report, and my SpamCop account copies KnujOn since bug 6085 is yet to > be addressed). The AWL bump gives me a head start on that senders' > future spams since they obviously beat SA in the past ... without that > bump, AWL would actually be subtracting points! > > Note, I just started this practice, and the recent "83MB > auto-whitelist?" thread's trim_whitelist script looks like it will > help me tremendously.
You are correct, I use it as a temporary bump to score spam higher, which will come down if I receive any volume of HAM from the sender, or take care of itself as the sender is added to various DNSBL's. The AWL is used for those messages that have been fed to Bayes, and still score below a threshold of 6.0 (an arbritrary number to optomize the process. -- Larry G. Starr - lar...@fullcompass.com or sta...@globaldialog.com Software Engineer: Full Compass Systems LTD. Phone: 608-831-7330 x 1347 FAX: 608-831-6330 =================================================================== There are only three sports: bullfighting, mountaineering and motor racing, all the rest are merely games! - Ernest Hemmingway