Yes — that non-locality on the surface is because the braiding structure is non-local when projected.
You are correct and I misspoke:
You don’t lose the information, it becomes non-local on the surface — and so if you’re building a model of the interior from a local sampling of the surface, you get a statistical model built on pseudoknots (again, like a shadow from a lamp).