Dara's Pearl B reports: from inside the B Hall tunnel you could not see out
its west end, and from the Hall above you could not look across the tunnel
into the yard beside it. The layers themselves were already right: the
painted ground is the tunnel floor, the Hall is a standing surface that
automatic standing prefers, and explicit elevation reaches the tunnel. The
walls were wrong.

* remeasure_above_openings.py: a tunnel-review piece keeps its reviewed
  ceiling but the blanket filled above it is intersected with the ray
  derivation, so only the slab edge and the structure from 11 m remain.
* apply_reviewed_openings.py with reviewed-openings-2026-09-19.json: the
  tunnel's west mouth takes the derived bands, a passage under a 7 m
  header that a protected record had kept solid.
* seal_void_walls.py: opening the roof let the upper level look over the
  yard's 6.9 m wall into the unpainted block behind it. A hole in the
  painted floor whose scene roof stands 4 m over the ground beside it is a
  building; every unprotected piece bounding it carries at least
  [floor, roof], unless a standing surface overlaps the piece, which marks
  an arch or overhang over walkable floor.

The tunnel test now accepts the measured slab edge or upper structure at
any of several heights. The Breeze facade regression detector checks the
lowered facade no longer stops its ray, since the wall behind it is a
building face. tool/svg_height_layer_view_test.dart renders ground height
with walls coloured by which of two eyes they block.

Gate: all tests pass.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>