claaarky - 7:31 am on Jul 13, 2012 (gmt 0)

Here's my take on it....

1) Google gathers user metrics via the browser (possibly in conjunction with data from ISP's etc.) as people visit and use your site (the sort of metrics you see in Google Analytics - I'm not saying they use Google Analytics, I'm just mentioning that to demonstrate the sort of user metrics I'm referring to). They do this for a period of weeks (the period between Panda refreshes). This tells them where the 'potentially' low quality pages are.
2) As googlebot crawls your site it gathers information that helps it make sense of the user metrics. For example, some pages will have low time on page and high bounce rates but if the page provides the perfect answer very quickly that is a good page. So it needs to see if the content (text, image, video, etc.) explains the user metrics and all this combined together tells Google if a page is high or low quality.
3) Having identified the low quality pages of your site, they then do a massive crawl just before a Panda refresh to see if the low quality pages still exist on your site, and if there is a high proportion of them still in existence at that point, a demotion factor is calculated for the low quality areas of your site and, to a lesser extent, areas of your site that link to the low quality pages.

I have a theory that you could get a page to the top of Google with just a few words on it and terrible design, IF those few words were so astounding and intriguing that it created superb engagement.

For example, if I created a website about the solution to Panda and the home page had nothing except one sentence in the middle that said "CLICK HERE FOR THE SOLUTION TO PANDA" that would be a pretty engaging home page for people looking for a Panda solution. If the website really did contain the solution, it would then get links, likes and shares like crazy. The whole website could be ugly, but if it contained the solution to Panda then human reaction to it (user metrics) would tell Google all it needs to know.

I'm with Kendo on this one. You can't get a machine to think and make all the decisions and judgements a human does in a split second based on our entire life's experience. How can a machine possibly learn at the rate humans do. A machine can only experience the web - humans experience everything in the world and it all affects our thinking. That thinking changes every day as well. How can a machine hope to keep up with that.