Almost all (>95%) of my spam is tagged as BAYES_99 by SA (which is
great), but only approx. 60% of my spam is classified as spam by my
Thunderbird 2.0.0.12. Thunderbird also uses a a bayesian filtering
system. I always learn all of my spam and all of my ham in both systems
perhaps once a week, not only FP/FN.

I'm just curious: is there any explanation for this big difference? Why
is the SA implementation that much better? One year ago Thunderbird
scored much better than nowadays, while SA is still at maximum efficiency.

Tschau
Alex