well, yeah, that's how the algorithms work... by using a training dataset where a human actually says "is spam."

And as for the "is spam" button... a given email isn't automatically added to the spam training dataset the minute 1 guy hits the spam button. it goes automatically into YOUR spam folder but not automatically into the training dataset. That takes many "votes" from many users.

That's why I would be happy with a personal spam filter: training by users' votes can be gamed, and producing false positives can take legitimate websites out of business.

I see there are some browser plugins floating around: I'm not happy with those because I use multiple workstations, and because hacking around a product's deficiencies is not "voting with my wallet".

Don't Google engineers actually use Google? They shouldn't really need flags to become aware of this sort of thing.

In my experience, the path from “engineer recognizes that queries X, Y, and Z return crap search results” to “search engine improves its performance with queries X, Y, and Z without creating more crap somewhere else” is more difficult than a lot of people realize.