I think the bayes filter just goes bad if you ignore it

[Cathryn Mataga]

And that no amount of training will fix it once it goes bad.  I was 
getting maybe 30-40 spams through
every day and many of them marked BAYES_00 -- and the last month I've 
been carefully feeding the
software spam and ham lists, all carefully checked, but so much of it 
just kept coming in.

But last night I backed up bayes, reset it, and then retrained with my 
old spam and ham email
collection, and difference is stunning.  Many fewer spam are coming in.