I spent some time during my travels going over the original layout. I've decided to take another crack at slimming everything down further - this means a main board width of 10mm or less. It's also a good opportunity, based on some previous dummy board assembly, to get the board-to-board approach right. I'm particularly interested in how far I can push down the size of castellations.

Obviously, still a work in progress. The goal is still to get everything on the same board using 0402 passives and OSHPark 4-layer spec DRCs. You can see why a lot of commercial units use flex circuits extensively.