tedster - 4:15 am on Jan 22, 2008 (gmt 0)

I think the real test needs to be comparing content that is semantically well-structured but "natural order" with semantically well-structured content that is also source-ordered. No argument from me that tag-soup is a deterrent to ranking well.

Because of the track record that source-ordering has had for me, I would be very reluctant to ask a client to do less than that, and I don't currently have the bandwidth to launch any new test sites of my own. But in recent months, new pages that are source-ordered do seem to me to have more of a challenge than they used to.

It's just a gut feeling right now - I'm noticing there are more finely nuanced semantic elements in the relevance metrics now that go beyond things like sequence in the source code. Google seems more able to zero on content anywhere on the page. And that's a good thing, IMO. The accuracy of some of the long-tail traffic I'm seeing lately is amazing, and that takes some major IR chops to pull off.