Instead of expecting everyone to train their spam filter from scratch, M&N could come with pretrained filter, the way e.g. Safari does. That would increase download size somewhat, but also save a lot of time for many users.

I don't think that this is a good idea. The spam filter is quite individual for a specific user/mailbox. If M&N came with a pre-trained junk control, the error rate would be quite high, because the user most probably receives other kinds of spam than the pre-trained filter would recognize. It's important to understand how the Bayesian filtering works.

1. Most of the spam is quite similar. 2. If it works in OS X' Mail, it can work in Mozilla. 3. "the error rate would be quite high": you just think so, or do you have any data to back this up?

I do not have any data to back this up, but I have read Paul Graham's essay about how Bayesian filtering works. This is why I _imagine_ that sharing training.dat files would cause problems. Actually, I can not prove this, but according to the theory, it would not work very well.

*** This bug has been marked as a duplicate of 179999 ***