On 18/12/2009 2:13 AM, Christian Brel wrote: > On he subject of Spammy whitelists... > > * -1.0 RCVD_IN_DNSWL_LOW RBL: Sender listed at http://www.dnswl.org/, > low > * trust > * [212.159.7.100 listed in list.dnswl.org] > > Yet the same IP is on and off SORBS and part of an ongoing spam > problem. Perhaps this can be reviewed and given a zero score by default?
Forgot individual occurrences of FPs or FNs. They're statistically meaningless. In last week's net-enabled mass-check the -1.0 score for RCVD_IN_DNSWL_LOW RBL caused only 10 of 148025 (0.00675%) spams to fall below 5.0 (and that could have happened with as small as a -0.1 score, I don't have data, so at approx -0.5 the same thing could have happened). On the other hand, it moved 101 of 199558 (0.05061%) hams below the 5.0 mark. That's an S/O of 0.035 which is pretty good (we wouldn't be questioning a spam hitting rule with an S/O of 0.965, at least not at a score of 1). http://ruleqa.spamassassin.org/20091212-r889898-n/RCVD_IN_DNSWL_LOW/detail Again, to anyone, if our statistics are way off from the reality our users are seeing we need more mass-check contributors. Daryl
