著者: Tom Bombadil 日付: To: Renaud Allard CC: Exim Mailing List, Chris Edwards 題目: Re: [exim] Estimating spam deliveries
> The hard part here is to detect false assumptions from users. For example, some
> will say that a mail is a spam while it is not, it is just a mailing list from a
> shop where they bought a product and clicked on the "I want to receive your
> mailings". The reverse is also possible (think phishing). As strange as it may
> sound, "un-spam-educated" users are often less accurate than automated spam filters.
Yes. Even trained eyes can make mistakes when sorting spam/innocent.
Anyways, you can fork a random sample of the messages, lets say 1% to a
spam counting mailbox:
#------------------------
# fork a random sample of the messages
#------------------------
random_sample:
driver = redirect
local_part_suffix = +*
local_part_suffix_optional
condition = ${if eq {${eval:($tod_epoch+$pid)%100}}{0} {yes}{no} }
data = random.sample@???
redirect_router = next_router_after_sampler
unseen
no_verify
#------------------------