Some people are trying too hard.

[reynardo]

Two pieces of email arrived in my inbox today marked:

***** SPAM ******

PNC: Planet Netcom's anti-spam server has detected that this email is most
PNC: likely spam. As such, the subject has been altered so that you can
PNC: recognise or block similar unwanted email in the future, using the
PNC: built-in mail filtering support in your mail reader.


I'm amused. One of the emails was a wally laughing about the latest Nigerian scam, and including the phone numbers and addresses.

The other, though, was just an email from a student wanting to sell some textbooks.

The ISP counts up points and calls "SPAM" if they get over 5 points. Points are awarded for things like:

PNC: Hit! (1.6 points) From: ends in numbers
(unfortunately, not unusual in student-number-based email addresses, but also with Yahoo and such)

PNC: Hit! (0.8 points) BODY: Contains a line >=199 characters long
(which happens when autowraps are applied. Also if the twit has left on HTML mail)

PNC: Hit! (2.7 points) BODY: Mutated Nigerian scams
(Ok, this one is fair)

PNC: Hit! (1.8 points) No MX records for the From: domain
(I have no idea what this means)

PNC: Hit! (3.9 points) Forged yahoo.com 'Received:' header found
(Actually, it wasn't forged. The mail had been sent through a mailing list. This means that every time someone from a web-based mail address sends something through the list, it's going to be marked at 1.6 + 3.9 = 5.5 points and thus as possible spam.)

I receive DAILY Asian "hot girls" spam in non-english text. This has never been filtered.

I've asked them to remove this filter. I figure I can deal with my own spam.

Comments

[barberio.livejournal.com]

It looks like they are using a filter called spamassassin, which is a quite good spam filter, if you configure it right. (I use it myself) Its not intended as a blanket multi-user filter, and has to be individualy adjusted for each user.

So, basicaly, good intentions, good tools, fscked up implimentation.

[barberio.livejournal.com]

Also, the recent versions will do adaptive matching to adjust. So if you recive regular e-mails from someone which are obviously not spam it will adjust the score downwards if it gets a suspect one in the future. But this isnt enabled by default, and needs to be run for a while before it starts to work well.

I'm testing a recent update to see if the problem with yahoo mailing lists is fixed.