The tree composition algorithm in section 5.5 appears only to act on the first shadow insertion point found in a shadow root.

If this is correct (which is probably is and should be*), to avoid confusion it should be explicitly stated that subsequent shadow insertion points are inactive.

* If subsequent shadow insertion points distributed some older shadow root, we'd have to decide who got first dibs on the shadow stack (depth or breadth first). It would also mean that you wouldn't know which shadow root you'd get when including a shadow insertion point, which seems like pretty undesirable behavior.