I agree. I think a system whereby the weighted average of long term scores dictates how well each email is received. ie, a friend that sends you 20 good emails should have a heavily weighted negative average to offset the occassional spam type email they may send you. And a spammer that sends you 10 spams should not be allowed to send you anything in the future, or at least not have the AWL score help it through.
With the IP number matching as well, it should prevent the AWL from blocking email from yourself, as spammers often forge the From: field to match the To: field, but their IP will not match up. This should give us the best of both worlds. One other observation over the last few days of receiving 200+ spams a day is that 2.42 seems to be tagging spammy emails just below 5 many times, as the spam getting through seems to be slowly increasing. Maybe the spammers are getting smarter these days, and the older corpus of spam is shifting the emphasis to older spam blocking instead of newer spam blocking. Rob M. > Well, the initial idea behind the AWL was to exploit the fact that > spammers typically forge their From addresses randomly, whereas legit > senders do not. > > This was changed around 2.31, and I put it back in for 2.42 since I was > seeing some nonspams that could benefit. > > But to tell the truth, I'm having second thoughts about it too, now. ;) > > I put it in a few weeks ago, and since then I've noticed a couple of > spammers in my feed who are getting through because of this, so it may be > time to recant on this point, and remove this behaviour again. The bad > points are outweighing the good, unfortunately. > > I'll check in the reversion now... I guess this means we'll need to > do a 2.43 soonish. ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
