Theo Van Dinter wrote: > On Thu, Dec 13, 2007 at 11:29:21AM -0500, Randy Ramsdell wrote: > >> I have doing some checking of spam messages that make it through our >> mail filtering systems and noticed that the spam score does not reflect >> what I get when checking manually. >> >> An example spam report: >> X-Spam-Status: No, score=3.068 tagged_above=-9999 required=5 >> tests=[BAYES_50=0.001, HELO_DYNAMIC_DHCP=3.066, HTML_MESSAGE=0.001] >> X-Spam-Score: 3.068 >> >> But when using "spamassassin -D -lint < $message" it hits more rules: >> > [...] > Are you *SURE* that works Randy?
note that --lint specifies rule-test only mode, and message scanning is dsabled. lint also (in recent versions) force disables any network tests, so hitting RCVD_IN_XBL would be impossible with the --lint parameter. >> 3.5 BAYES_99 BODY: Bayesian spam probability is 99 to 100% >> 3.9 RCVD_IN_XBL RBL: Received via a relay in Spamhaus XBL >> 0.0 RCVD_IN_PBL RBL: Received via a relay in Spamhaus PBL >> >> That is a big difference! >> Any ideas about why this is? >> > > It appears that the first results are a) using a different Bayes DB, > and b) not using network tests (aka: local mode). > > Also: c) performed using amavis, which will force local-only mode via the *sa_local_tests_only *option in the amavis config. If you want network tests, set this to 0. Also, amavis will run the tests as whatever user amavis runs as, so if you want to do any sa-learning, or valid spamassassin tests, do so via something like: su $amavis_userid -c 'spamassassin -t < $message
