Is it me or does this article end right when some actual implementation details would have been interesting to see?

Yeah I would've loved a little more detail on how they architected the move to NSCoding / serialization straight to disk. There's a lot more ways to do this wrong than right from how open-ended a 'schema' like that is. Any speculations?

A while back, I came to the same conclusion as Facebook's iOS engineers for some apps that I work on. I based the model layer of the apps on Github's Mantle: https://github.com/Mantle/Mantle

Facebook's implementation probably ended up looking somewhat similar.