On 2022-04-15 at 14:20:28 UTC-0400 (Fri, 15 Apr 2022 19:20:28 +0100) Laura Atkins via mailop <la...@wordtothewise.com> is rumored to have said:
> .eu.org <http://eu.org/> is, essentially, a tld. And .tlds have their own > reputation, too. Just this week a few of us were talking about ‘weird’ tlds. > One of the participants works at a filtering company, checked their stats and > said “this particular tld is 9x% spam.” So 9+ times in 10 when they see a > domain registered in that tld, it’s spam. This is consistent with the SpamAssassin RuleQA stats. https://ruleqa.spamassassin.org/?daterev=&rule=%2FTLD shows the stats for rules with 'TLD' in their names. (N.B.: the SA sample size is nowhere near what the big providers have, and it probably skews differently) The default rules channel includes a list of 'suspicious' TLDs that have been used predominantly in spam, at least at some point in the past. Reliably 99%+ spam & with some minor FP protection that goes over 99.9%. Some of those have been pulled out as test rules (T_*) in response to complaints by "legit" (stipulated, unchecked) senders. Every TLD tested individually that way has been found to be persistently associated at least 95% with spam, except for .space which has a slightly better record hovering around 90%. It is entirely reasonable to see a TLD (or any zone used as a registry) as a meaningful attribute in spam filtering. There is a correlation in many cases. I cannot nail down what causes that correlation, but that doesn't affect its utility. OTOH, even with the spam/ham reduction in recent years, the majority of mail is still spam so a 90% correlation isn't as significant as it sounds. A few years ago, a 90% spam source would have been *better* than the aggregate mean... -- Bill Cole b...@scconsult.com or billc...@apache.org (AKA @grumpybozo and many *@billmail.scconsult.com addresses) Not Currently Available For Hire _______________________________________________ mailop mailing list mailop@mailop.org https://list.mailop.org/listinfo/mailop