tedster - 4:58 pm on Jul 2, 2011 (gmt 0)

It doesn't need to know that YOUR page has a high bounce rate or low CTR, only that it looks like one that would.

Yes - and that's the essential idea behind machine learning. That is the approach Google has preferred for many years, rather than human composed "guesses" at would be good algo factors. The idea is to use a hand-chosen seed set and then isolate the factors that describe just those URLs and not others. Then those factors can be used to build a predictive algorithm. Panda does it for "quality" but "trust" has long been done that way - building out from a seed set.

The early TED interview about Panda 1.0 [wired.com] essentially describes the process of machine learning in this case. I've returned to that interview many times in recent months.