I've been getting a bunch of spam hawking mortgage rates. You may have seen it, it starts with "Dear Homeowner." Tthe only test that flags this message is "BAYES_50", for all practical purposes a score of 0.

What concerns me the most is that this triggers "autolearn=ham". I later feed this back through sa-learn as spam, but what I'm wondering is whether this undoes the damage to the Bayes databases caused by the autolearn=ham.

I'm considering lowering the autolearn threshhold to less than zero. I wonder if anyone else has any thoughts on this as well.

Thanks

Jack




Reply via email to