I totally rewrote that note about focusing. It is still missing some details I didn't feel like grinding through, but it's much more readable, and makes the point I didn't realize I should have been making all along, that 90% of the reasoning is just about adjoints, and only 10% of it is about focusing in particular.
The punch line is still that focusing discipline is nothing other than "inverting discipline" (where invertible rules must be eagerly decomposed) where up and down shift are interpreted as p-tensor and p-lolli.