diff --git a/docs/ISSUES.md b/docs/ISSUES.md
index 3ae572ae..eb8aa142 100644
--- a/docs/ISSUES.md
+++ b/docs/ISSUES.md
@@ -55,6 +55,19 @@ three kit poses (terrace-edge, cathedral-arrival, foundry-deep) reproduce exactl
 `WalkTraceConformanceTests.Oh_doorway_still_first_frame_diff` carries the row tagged
 `Status=KnownFailure` with the position documented.
 
+**Lead experiment (2026-09-03):** with `DegradeMultiplier = 0f` (the FW0 sibling's
+cdb-depressed value) the row diverges at the SAME token 165 — the degrade level is NOT the
+cause. A token diff of the full frame shows THREE divergences, not one: (1) token 165:
+acdream inserts `LC/SC a9c90001` (block `a9c9`, ring 21 straight north of the viewer's
+`a9b4`; retail admits `a8c9` and `95c6` around it but not `a9c9` — a `draw_check_blocks`
+block test difference, likely the block's MinZ/MaxZ clip heights against the exit-view
+edge planes); (2) retail tokens 1093–1107: two look-in floods acdream never enters —
+`DC:ov=0:a9b40103,a9b40100` (+EC/OC) and `DC:ov=0:a9b40100` (+EC/OC); (3) retail tokens
+1112–1115: a third look-in `DC:ov=0:a9b40124` (+EC/OC). (2)/(3) are the PRE-EXISTING FW1
+'near-win' look-in gate state at this pose (`WalkLookInGateSweepTests` is still skipped;
+the FW0 `Doorway_still_first_frame_diff` was a diff printer, never an exact pin), now
+exposed by the eight-kind comparison. Signature lengths: retail 1185 tokens, acdream 1170.
+
 **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