2003-05-14»
new spamassassin out»
I have to write about something other than spam sometime. Anyway, Justin Mason wrote in, to say the new version of SA (v.2.54) severely punishes those Pine-'n'-Mozilla
pretending creeps, and adds: "overly cautious my ass". Apparently Razor
doesn't get high scores, because the genetic algorithm that they use to
calculate the scores doesn't think it needs it. In other words, most of the
Razor stuff gets caught by other traps. He also reckons I've got BAYES_60 and
BAYES_70 a bit too high, and maybe he's right - I got one FP from LinkedIn the other day. On the other
hand, I have problems separating
give-all-your-friends-email-addresses-to-a-third-party services like LinkedIn
from spam.
2003-05-10»
spam»
Wow. I now get over 20,000 spams
a month. Maybe that explains why I seem to get a lot more false
negatives (spam that gets through) via Spamassassin. It's the same small
percentage as ever, there's just a lot more of them.
Some spammers are definitely working on evading SA's filters. I see a
lot of spams that pretend to come from Pine and Mozilla
simultaneously - scoring them a massive minus twelve on Spamassassin's
anti-spam-o-meter. I've tweaked the reward for these tests down a bit in my
own user preferences. I've also nudged up the penalties for falling foul of
SA's new Razor2 and Bayesian modules. Spamasssassin's still a bit hesitant to
trust these much. But with the amount of spam (and normal mail) I get, my
Bayesian filters get all the training they need. Don't know why it gives Razor
such a low score - perhaps Spamassassin's overly cautious approach to false
positives is leading it to punish for accidentally catching some major mailing
lists?
% grep score ~/.spamassassin/user_prefs
score USER_AGENT_PINE -0.5
score USER_AGENT_MOZILLA_UA -0.5
score BAYES_99 8.0
score BAYES_90 8.0
score BAYES_80 8.0
score BAYES_70 8.0
score BAYES_60 4.0
score RAZOR2_CHECK 3.0
score RAZOR2_CF_RANGE_91_100 3.0