# Nested cut/refill restoration defect

Before any kernel change, `test_replace_remove_and_regenerate_restore_skin_without_stale_window_owners`
failed: removing `/lightwell` left two missing host roof cells. The complete first
test output is preserved in `first-tests.txt` (12 passed, 2 failed; the other
failure was a test fixture accidentally calling nonexistent `View.cells`).

Concrete sequence: parent P replaces host H's cell with air; child C writes its
wall into that air; grandchild W replaces C's infill with glass. Removing P first
removes W and C. The old loop then skipped P's displaced H record because the cell
was empty and its erasure marker had been consumed by C's write. The original H
record was already retained correctly; the restoration traversal lost its eligibility.

This violates the intended R12/R19 removal behavior. It is an ordinary implementation
bug: the existing displacement journal contains the required history and the failed
removal rolls back atomically. No new architecture, authority relaxation or persistence
format is needed. The fix lets the traversal restore a recorded prior cell if that
coordinate was just vacated by the same branch removal. Foreign surviving owners and
erasures remain protected by the existing checks. A geometry-independent nested-cut
test and paired-reload removal test accompany the lantern case.