Keith Moore wrote:
Just imagine how many wrongly rejected emails aren't reported.
Stupid spam filtering mechanisms are a DoS attack on email.
The problem is nearly all of our anti-spam measures are empirical. We
all know a lot of people who swear by this, that, or the other check,
even if the only supporting evidence is that using that particular
mechanism cut down the number of complaints. And what works on one
stream fails miserably on another.
FWIW, I last looked at this problem about three years ago. I
specifically wanted to know if some form of RDNS checks might be useful
in cutting down the load on the content-based spam filters. (I was also
checking effect and utilization of SPF and TLS.) Note that the purpose
here was not to improve the catch rate. This was on a stream that was
already filtered by a short-lived (60 minute) IP reputation filter, so
that would reduce the message count from some known spam sources; and
most of the recipients were business users.
As I recall, my sample size was something around 10 million E-mails from
a single MX server in a load balanced cluster over a 24-hour period.
There was a weak correlation between spam and a simple existence check
for the PTR record. There was no correlation at all for a stronger
check, e.g., A record matches; a message that failed the strong check
was as likely to be judged ham as spam by the content filter.