You can do better than this, Norm. Think. And do a modicum of research. OTOH, you're providing a good opportunity to explain why content/context based spamfiltering is so powerful and useful.

Among the rationales behind my article is the fact -- not opinion, not possibility, not future likelihood, but FACT -- that broad-based, multiple-factor, weighted analysis can achieve spam reduction rates of:

\r\n
• Spam detection (true positive): 98%
\r\n
• Ham pass (true negative): 99.8%
\r\n
• Spam pass (false negative): 2%
\r\n
• Ham filter (false positive): 0.2%
\r\n

...or better. In typical use. Not highly rigged "benchmarketing" tests, but real user experience.

"Spam" is of course spam mail, "ham" is legitimate, desired mail.

A full list of rules and scores assiciated with them as used by Spamassassin is [link|http://www.spamassassin.org/tests.html|here]. Briefly, these include:

\r\n
• Header analysis: who the mail is from and to, and what servers it originatef from or was sent from, subject.
\r\n\r\n
• Text analysis: weighted keyword and phrase analysis indicating spam or ham likelihood.
\r\n\r\n
• IP Blacklists: weighted assessments of spamminess based on IP points of origin.
\r\n\r\n
• Known spam checks: Use of Vipul's Razor and other lists which track known spam payloads and reject them.
\r\n\r\n
• Bayesian analysis: A weighted assessment based on the users's own identification of spam and ham messages of whether or not a message is spam.
\r\n\r\n
• Whitelisting / blacklisting: adding of specific addresses to whitelists or blacklist, either automatically or at the users' request, to assure that specific addresses are passed or filtered.
\r\n\r\n

Because of this, a direct, highly specific, individual assessment of spam on given messages can be made, rather than relying on very course methods such as rejecting mail from large blocks of IP addresses for no reason specific to an individual email other than that spam has been received from neighboring addresses. In the real-estate business, this practice is known as redlining, and is illegal.

It also means that systems such as [link|http://kmself.home.netcom.com/Rants/challenge-response.html|challenge-response] which "solve" the spam problem by shoving it off on others cannot be justified.. I'm very much trying to get Earthlink to reconsider its plans to offer C-R antispam service as this will likely cause Earthlink as a whole to be considered a spam source itself by many remote systems.

Not only is this possible, but the processing throughput and overhead are relatively low. In maximum efficiency configuration -- running Spamassassin in client-server mode, with no network checks (these take a significant amount of time and don't markedly improve accuracy), it's possible to handle 10-20 messages/second on reasonable x86 hardware. Say, 256 MB, < 1GHz, and standard 7200 RPM ATA disks.

Even in the case of a monster ISP such as AOL, which claims to have blocked more than 2 billion spam mails (thousand million for EU types) in a single day, or 80% of incoming mail, this load could be handled at low per-user cost. If the bulk of the 2.5 billion daily mails arrives in a 10 hour period, and spam filtering takes 0.1 second per message, the load could be handled with a spam filtering cluster of 700 commodity x86 boxes. Estimating deployment costs generously, at $1000 per box, the total cost of the spam filtering would be $700,000. Which sounds like a lot until divided amongst the 35 million customers of AOL, giving a price of $0.02 per customer. Amortized over the life of the equipment. Note that deployment costs would likely be less, net throughput probably higher, and other functionality could be assigned to the equipment. This is just a rough estimate, someone with a better take on capabilities and costs could dial this in tighter. At this cost, and given the negative costs associated with falsely filtering out legitimate mail, the proposal should be a strongly positive one for AOL customers.

Further, with such an online, realtime assessment, AOL would be btter placed to combat spam in real time by identifying high-volume spam sources, and retaliating either by blocking or delaying mail from such locations for the period of the attack, by identifying persistant sources of spam and denying them service or strongly degrading service to these sources, or by taking legal or other actions against the operators of the systems or networks from which the spam was originating.

Read the fucking manual. Research your problem. Join the mailing list. Find out what you're doing wrong, or if the app works as advertised.

Train the fucking Bayesian features -- they're adaptive, which means you have to give them data to adapt to.

If SAProxy doesn't work, then try [link|http://au.spamassassin.org/where.html|a fucking different product]. There are lots incorporating SpamAssassin and other Bayesian sorters.

Don't be a dumb fuck. Try. Fix. Solve. Get off your ass. Don't whine that it didn't work, you've played that card way too often here, bud.