I think I’ve discovered the reason that SpamAssassin has been letting a lot of spam through – the bayesian filter is effectively neutered if you have the net checks on.  Now I’m sure the SpamAssassin team has a bunch of statistics why this is good in the most common case, it’s not working out so well for me.

I live with my spam filters set to throw anything scoring 3 or higher into the spam bucket (I think SpamAssassin’s default is 10) so I’m ruthless with spam and it tends to classify forwarded jokes as spam which isn’t such a loss (it also classifies emails containing flight details from QANTAS as spam which isn’t so good but QANTAS emails look more like spam than the real thing so what can you do?).  Anyway, from what I can tell, SpamAssassin will be a lot more effective for me if I add the following to my user_prefs file:

score BAYES_99 10

With my setup that pretty much guarantees that anything that the bayesian filter marks definitively as spam gets treated as spam.  It also appears that spam advertising miracle cures for erectile disfunctions have been confusing the bayesian filter a little and scoring only around 40-50% so I’ve added:

score DRUGS_ERECTILE 10

to get rid of them.  We’ll see how the new setup goes.  If anyone from the SpamAssassin team happens to read this and know why the BAYES_99 score was reduced I’d love to find out – particularly if I’m likely to discard too much ham with my new settings.