Thanks, everyone, for your suggestions. They’re very helpful. > On 4 Jan 2026, Peter Nicolai Mathias Hansteen via mailop wrote: > If you want to run analyses I can make at archived spamd logs available. I > have weekly archives going back to late 2021, but I suppose a smaller sample > might be just as useful?
That’s a very kind offer, which I may very well take you up on later this year if I find myself with some free time. > On 5 Jan 2026, Niels Dettenbach wrote: > I think it makes no real sense to run own classical spam traps just to get > info about potential spammers to block them. The amount / share of spam you > get down from that is comparably low. > On 5 Jan 2026, Laura Atkins via mailop wrote: > IMO, there are more effective (catch more spam, cost lest to maintain) ways > to filter spam locally. You are both probably right, but I’m trying to solve a specific problem: the spam/ham classifier training set is currently heavily biased towards HAM + false-positive SPAM, so I am attempting to reduce this bias using spam traps. > On 5 Jan 2026, Niels Dettenbach wrote: > Beside this, spamtraps could be used > (if known to someone bad) as a tool to block valid email traffic. How? By poisoning the training set with HAM-like messages? > On 5 Jan 2026, Niels Dettenbach wrote: > Especially for more small volume email hosts using other datasources instead > makes much more sense / is way more efficient - i.e.: > > - multiple DNSBLs > - DCC > - Pyzor / Razor etc. > - commercial data sources(for small volume / private use often free entry > products available) > > to weight it out wih i.e. SA or spamd > > a additionally well configured / adapted fail2ban may help as well very far > against brute force stuff and even somewhat of (D)DoS. I have already implemented all these, except DDC, and they work very effectively — which leads to a paucity of spam and the bias problem mentioned above. Dan
smime.p7s
Description: S/MIME cryptographic signature
_______________________________________________ mailop mailing list [email protected] https://list.mailop.org/listinfo/mailop
