diff --git a/docs/ISSUES.md b/docs/ISSUES.md
index 05c8848b..1cdbcc4b 100644
--- a/docs/ISSUES.md
+++ b/docs/ISSUES.md
@@ -76,6 +76,18 @@ admission: the per-view block edge-plane test (`WalkVisibilityMath.BlockCheck` a
 `LScape::draw_check_blocks` @0x00505f80's clip-height grid — BN elides the 192 m block
 interval to `0f` there) and the exit-view union at this pose (two surviving views).
 
+**Structural comparison (lead, 2026-09-03):** `LScape::draw_check_blocks` @0x00505f80 and
+`WalkLandscape.CheckBlocks` match in structure — the two-parity corner-interval rows
+(`block_interval[((x-1)&1)*(mw+1)+y]` east, `[(x&1)*(mw+1)+y]` west), the per-view
+overwrite of `in_view` (`Render::block_check(west_y, west_y+1, east_y, east_y+1, max_zval,
+min_zval)` @0x005061df), and `landcell_check`'s cell union. The admission of `a9c9` therefore
+comes from the NUMERIC inputs at this pose (the two exit views' edge planes, the CY plane,
+or `Render::get_clip_height`'s per-corner classification) and cannot be settled by reading.
+NEXT STEP: one exact retail capture — a cdb template on `Render::block_check` @0x005061df's
+caller that prints, for block `a9c9` at the doorway pose, the four corner intervals, the
+z slab and the returned BoundingType, alongside acdream's `BlockCheck` inputs for the same
+corners; compare plane by plane.
+
 **Degrade note (2026-09-03 review):** the FW0 sibling row for the SAME pose pins
 `RetailFrameWalk.DegradeMultiplier = 0f` because that capture ran under cdb load with
 `Render::deg_mul` depressed; the OH doorway capture also ran under cdb, so check the