diff --git a/docs/stage-d-trial-3-and-directors-desk-2026-11-04.md b/docs/stage-d-trial-3-and-directors-desk-2026-11-04.md
index 88fb7d9..cbc1ce5 100644
--- a/docs/stage-d-trial-3-and-directors-desk-2026-11-04.md
+++ b/docs/stage-d-trial-3-and-directors-desk-2026-11-04.md
@@ -87,3 +87,23 @@ One TV screen, in Orbit's navy and gold, using the real numbers from this trial:
 1. **The six fixes above,** then a fourth fresh sample, with samples 1 to 3 kept as regression tests.
 2. **A small read-only release** that reads Emby's own collections, so franchise matching can use verified links.
 3. **The Desk in the app, read-only,** exactly as designed, once Paul approves the design.
+
+## Paul's approval (4 November 2026, 23:42)
+
+**Paul approved the Desk design.** Its home is **Settings › Director's Chair**.
+
+**Orbit learns from the decisions made there,** across:
+- channel placements;
+- films;
+- Drive-In events;
+- What's On's picks.
+
+That learning stays local and explainable, keeps Familiar, Balanced and Surprise Me (Balanced by default), and is never a preference bubble.
+
+**Order:**
+1. The scoring fixes, a fresh sample 4, and a read-only read of Emby collections for verified franchises.
+2. The scoring model moved into the app (Kotlin, on the box, no cloud).
+3. Settings › Director's Chair, read only.
+4. Then interactive decisions and learning, presented for approval first.
+
+Auto-approval stays off.