diff --git a/HANDOFF.md b/HANDOFF.md
index 2816eaa..26536fc 100644
--- a/HANDOFF.md
+++ b/HANDOFF.md
@@ -1,4 +1,4 @@
-# HANDOFF — 2026-11-06 (Starting art + site scenes)
+# HANDOFF — 2026-11-09 (TASK 33 policy; vertical checks passed)
 
 ## State
 
@@ -11,7 +11,10 @@
   bundled background needs no version bump.
 - Left panel alert stack shipped: Steam licence indicator pinned at row 1, event rows LIFO, click →
   About or Logs with the event's lines painted. Detail: `docs/ai/ui-surface.md`.
-- Site: the Starting screen is now in the scenes section (`scenes-shot`), deployed 2026-11-06 12:05.
+- Site: demo video (silent loop) is the hero; Starting screen is in the scenes section.
+- **TASK 33**: policy + planner done (`bc62bd8`, `555ff5c`); remaining — VM subscribe, encoder
+  restart, toast, capture-cost shed. **Creator hands cleared 2026-11-09**: vertical camera check 1–6
+  pass, vertical cadence 29.8 distinct/s (no defect).
 - Tests: 575 green; `FAST=1 ./scripts/verify.sh` ≈ 62s. Background suite 45 green.
 
 ## Next action
diff --git a/docs/ai/frame-pipeline.md b/docs/ai/frame-pipeline.md
index c2530f6..3362596 100644
--- a/docs/ai/frame-pipeline.md
+++ b/docs/ai/frame-pipeline.md
@@ -24,6 +24,18 @@ what makes a face judder while the file looks flawless. (The "webcam renders at
 used to sit in this map was WRONG: it came from a different session and counted every newly rendered
 layer, not webcam arrivals. Corrected 2026-10-05.)
 
+**Vertical-tier take (2026-11-09).** `ty-20261109-1142-036-0005.mp4`, Release, 1080×1920 @ 30 fps,
+300 s, webcam in its default bottom-right tile. Cropped the tile, downscaled to 160×88 grey and
+diffed frame to frame: **29.8 distinct images/s overall, median 30-second window 30.0/s, worst 24/s,
+one 4-frame repeat run in 9026 frames.** The app's own lines agree — `webcam 28.4–30.0/s` against
+`arrivals 29.9–30.1/s`. So the 25.6/s failure above does **not** reproduce in the vertical tier, and
+the vertical take therefore does not block the cadence units.
+
+⚠️ Threshold matters. A naive `mean abs diff > 1.0` on a downscaled tile reports long "0.0/s"
+stretches during low motion — that is the measurement failing, not the camera. Duplicates encode to
+near-zero (<0.02); a still face with sensor noise sits around 0.3–0.7. Use ~0.05 and read the
+histogram before believing a count.
+
 **`Helpers/FrameCadence.cs`** — one pure counter, three feeds. `Add(timestamp)` on any shared
 monotonic clock; `Summarize()` → frames, rate (gaps ÷ span, NOT frames ÷ span — N frames span N-1
 periods), gaps, hitches (default threshold 50ms ≈ 1.5× a 30fps period), longest and average gap;