Am 16.09.2014 um 16:01 schrieb John Hardin: > On Tue, 16 Sep 2014, Reindl Harald wrote: >>> Le 16/09/2014 12:24, Reindl Harald a écrit : >>>> score LIST_PARTIAL 2.000 1.999 2.000 1.999 >>>> >>>> that feels too high, as example we add "List-Unsubscribe" headers in case >>>> of ordiany newsletters to support MUA >>>> which read that header (for TB a extension exists) >> >> i have seen that rule hit one customers newsletter generated here, going >> through a relay responsible for the >> sender domain and coming back to the customer itself why it makes it through >> SA > > Important question: are those messages being scored as spam? > > If not, how close to spam (5.0) are they being scored? > Are you willing to provide a sample of one such?
in that case not, but depends on some wired HTML, bad chosen subject or so hit another rule where 2 points penalty may come closer to spam, keep in mind nobody knows how good or bad a bayes on the RCPT side maybe trained what alarmed me was not that our SA in that specific case had -1 and not -3 but that a message with a header added by good intentions get a high penalty in general seeing other 100% spam make it through the milter because the "score RP_MATCHES_RCVD -3.5" (reduced here to -0.5) and so even outbeating BAYES_99 makes +2 simply too high
signature.asc
Description: OpenPGP digital signature
