On Thu, 11 Feb 2010 12:26:03 +0100 Bernd Petrovitsch <be...@petrovitsch.priv.at> wrote:
> On Don, 2010-02-11 at 10:38 +0000, Mike Cardwell wrote: > > On 11/02/2010 08:27, LuKreme wrote: > > >> At SMTP time I return a 5xx code during the "DATA" phase for > > >> messages classified as Spam. However, I also deliver the message > > >> into a read only "Junk E-Mail" folder for the user, > > > > > > This is just wrong. Either accept the message, or reject the > > message. Rejecting the message while secretly accepting it is just > > completely wrong. > [...] > > > Let's say your filter catches a legitimate message to > > u...@yourdomain.tld from b...@mydomain.tld. Bob gets an erro saying > > the message was spammy and didn't go through, so he goes to his > > gmail account and sends it again, hoping for better results. This > > time it goes through. > > > > Bob could also have just clicked the link in the NDR. > Some people - e.g. /me - do not try to pass Turing tests. Obviously > you are not interested in my mails anyway .... But it's only applied to mail classified as spam, and unlike CR it generates no additional backscatter. > Apart from that why should I decode captchas from some random site? > After all, they could come from a third site so that people solve them > to the the other can log automatically into the third one ... Because the NDR is generated by *your* mail server in response to *your* email. I think it's rather a good idea.