On Wed, Oct 18, 2006 at 02:40:31PM +0200, Renaud Allard wrote:
> Andrew - Supernews wrote:
[...]
> > (I say "perceived" because I am skeptical about the proportion of spam
> > that actually fails callout verification.)
> >
>
> The proportion of spam that actually fails callout verification varies
> from 10% to 99.99% during peaks...
for a data point, here about 50% of my spam over the past
day or so has a valid sender; about 40% has an invalid
sender, and the remainder didn't complete sender
verification within the (fairly aggressive) timeout. By
comparison, about 75% of real mail has a valid sender, and
a bit under 1% an invalid sender. However, if I'd rejected
all the invalid-sender mail I would have lost (among other
mail) an important contract, some correspondence with A
Commercially Successful London Data Center Whose Owners
Are Too Incompetent To Implement SMTP[*] and various
automated mail I really did want to receive.
According to my Bayesian filter, ``SENDER_VERIFY_GOOD''
has spam probability about 0.6 (i.e., of mails given it
for training which passed sender verification, about 60%
were spam). The corresponding probability for
``SENDER_VERIFY_BAD'' is 0.93.
* -- Telecity Redbus.
--
``[Ham the chimpanzee, upon returning from orbit, was] a thoroughly
infuriated space veteran who bared his fangs and bit anything...
he could reach.'' (William Burrows, from `This New Ocean')
