The amazon-for-source is a cool idea that could also be applied to navigating docs. It's a 1st order Markov process (predicting the next link based on the current link), but could be extended to 2nd order (predicting the next link based on how you got to the current link), or <i>n</i>th order, of course. This context might capture the reason you were looking at the current page. We could do a little statistical AI, and combine the predictions of different sequences that are similar enough.<p>The problem that source is changed (unlike books) as it is fixed/refactored/deleted, probably wouldn't hurt much.
Great writeup. I'd add a lesson I'd learned coming out of lab research as well. "Don't develop in a vacuum, involve the end users early and often." which is a bit of a corollary to "Lesson learned: Customers may have problems, but they also have solutions." I see this also outside of research, in general practice also. You seem to have figured this out naturally as part of your idea development process, but it's amazing how often this doesn't happen.
In a different world(java) the social aspect of mylyn/tasktop is pretty similar - <a href="http://live.eclipse.org/node/573" rel="nofollow">http://live.eclipse.org/node/573</a>