Am 01.05.2015 um 20:22 schrieb Matus UHLAR - fantomas:
On 01.05.15 11:17, Kevin A. McGrail wrote:Reindl's test showed it is hitting URIBL and it looks like it's also a good candidate for raising your SPF score and Bayesian training.I wouldn't say so... for getting standard required_score 5.0 is standard 1.7 points for URIBL_BLACK and standard 3.5 for BAYES_99 just enough - properly trained BAYES db should be enough
depends on your setup and *what you do* with high-score spamrequired_score si raised here to 5.5 to not tag much messages as spam just because of some bad template while the most interesting part is reject messages above 8.0
hence a BAYES_99 with 3.5 gains you nothing here and the reject score of 8.0 is careful chosen given a well trained bayes scores 80% of all messages making it to SA with BAYES_00 and so -3.5 by having a zero-FP policy but at the same time the goal to reject as much as possible instead deliver it flagged
BAYES_00 51315 80.37 % BAYES_05 1794 2.80 % BAYES_20 1437 2.25 % BAYES_40 1125 1.76 % BAYES_50 3711 5.81 % BAYES_60 456 0.71 % BAYES_80 386 0.60 % BAYES_95 297 0.46 % BAYES_99 3326 5.20 % BAYES_999 3017 4.72 % DNSWL 55777 87.36 % SPF 39387 61.68 % SPF WL 8159 12.77 % BLOCKED 5102 7.99 % score BAYES_00 -3.5 score BAYES_05 -2.0 score BAYES_20 -1.0 score BAYES_40 -0.5 score BAYES_50 2.0 score BAYES_60 3.5 score BAYES_80 5.0 score BAYES_95 6.5 score BAYES_99 7.5 score BAYES_999 0.4
signature.asc
Description: OpenPGP digital signature
