From: "Steve [Spamassasin]" <[EMAIL PROTECTED]>
I'm using spamassassin (Razor, Pyzor, DCC) and procmail to filter all my
mail on my (Gentoo) linux-server, to which I connect from a number of
Windows (XP/2000) machines using Mozilla Thunderbird to access my
(dovecot) IMAP folders on the linux server. I configured spamassassin
to use "Rulesdujour" and to regularly update those rules - and I was
very happy... at least 99.99% of spam was correctly marked with only one
incident of false positives (for which spamassasin wasn't entirely to
blame.) in several months.
You do not say which version of spamassassin you are using. If it is not
3.04 an upgrade might help.
Lately I've been less lucky - only ~99% of my spam is marked as such...
which sounds good but the remaining 1% gives me up-to a dozen bogus
messages each day... which is frustrating. To the naked eye the missed
spam is obviously spam - but typically the only significant rule it
triggers is the Bayesian rule... As I've stuck to the default settings
this alone is insufficient to identify a mail as spam.
So far Bayes 99 triggers ONLY on spam here. I use a per user Bayes. I do
not use autolearn (or autowhitelist) and I basically train only with
caught spam that is not up to Bayes 99 or with spam that escaped. It has
taken some time; but, I've made it to one in one thousand spams or less
escaping detection with about the same level of false alarms.
I'm left with several questions...
* Is there somewhere where I can report spams which aren't caught by
the default configuration in order to feed-back into future
improvements?
There are places to report them manually.
* Is there an easy way to report spam explicitly to the checksum
services (Razor/Pyzor/DCC)?
I have a strong personal bias against automating anything related to
spam REPORTING. Please examine the downsides of automatic reporting
before proceeding.
{^_^}
