not SPAM'ers

Upon termination of all tests from all volume of letters (44552 letters) it has been allocated 125 absolutely валидных letters on the basis of which parameters FalseNegative and FalsePositive have been calculated. If to exclude the first day of testing it is visible as commercial products go a leg to a leg at the analysis of a spam. Spamassassin all it is calculated on the analysis of a foreign spam and often passes letters because of absence of normalization of a content. Commercial products lead letters to the certain kind. For example, often used spammers a method of an insert of superfluous signs in words of type: , and also replacement of Turk letters with English letters similar on a writing, are absolutely inefficient against normalization at which the deformed word all will equally turn to "dispatches" and will get under the contextual analysis adding for each such word certain quantity of points to the letter. In turn, SpamAssassin well enough distinguishes a foreign spam of type « Medical, Viagra, Cialis, Enlargement », and also well filters « bayes poisoning » contextual analyzers.

In occasion of dnsbl-lists the opinion has developed ambiguous. On the one hand, popularity public dnsbl is very high, and they can provide protection against a spam up to the certain degree. But, on the other hand, to use dnsbl-lists as panacea it is impossible, as everyone happen FP at work with valid domains. As if to check of entering connections on RFC (FastBL) during testing there was no FP and the level of a filtration is comparable to commercial products. But it already other history. :)

Well, and at last, I have decided to tabulate variants of events in the event that contextual analyzers would work in pair with DNSBL-sheets. Predictably, I have received 100 % efficiency of a filtration of a spam even at use of foreign product Apache SpamAssassin.