diff --git a/sessions/0022/SESSION-0022.0-TRANSCRIPT-2026-06-26T08-40--2026-06-26T08-50.md b/sessions/0022/SESSION-0022.0-TRANSCRIPT-2026-06-26T08-40--2026-06-26T08-50.md
new file mode 100644
index 0000000..45c3990
--- /dev/null
+++ b/sessions/0022/SESSION-0022.0-TRANSCRIPT-2026-06-26T08-40--2026-06-26T08-50.md
@@ -0,0 +1,79 @@
+# Session 0022.0 — Transcript
+
+> App: human-experience-filter-art
+> Start: 2026-06-26T08-40 (PST)
+> End: 2026-06-26T08-50 (PST)
+> Type: executing-plans (operator-driven follow-up to the audio feature)
+> Posture: yolo
+> Claude-Session: e0cc053a-41fa-416d-a4f5-493f14984e75
+> Status: FINALIZED
+
+## Launch prompt
+
+> (1) "It shouldn't say 'Tap to Start Sound'… it can start with both sound and video
+> off but immediately go to the app. We should have a 'Loading Universe…' loading
+> screen until all media has been downloaded and the experience is ready to run."
+> (2) "I still can't hear the sound, even in private mode."
+> (3) "The first time video is turned on, also turn on audio. If audio is turned on
+> before video after the app loads, just audio should turn on."
+
+## Pre-state
+
+- Sessions 0020/0021 had shipped the audio feature + a 0021 "Safari fix" (tap-to-start
+  overlay that primed A/B `<audio>` elements). The operator reported audio STILL
+  silent on their real browser (even in private mode), plus the UX requests above.
+
+## Session arc
+
+1. **Why the 0021 fix failed.** It primed *srcless* A/B elements in the tap gesture
+   and played the real audio in an *async continuation* (after `await fetch`). Safari
+   unlocks an element only when `play()` runs **synchronously inside** the gesture and
+   on an element **with a src** — so neither condition was met. Headless WebKit relaxes
+   autoplay, which is why automated tests never caught it.
+2. **Redesign (client-only, branch `session-0022`):**
+   - **Single `<audio id="aud">`** element (dropped A/B). It's played **synchronously
+     inside the toggle's own `change` handler** — the canonical Safari unlock. Altitude
+     changes reuse the now-unlocked element with a brief fade-swap.
+   - **"Loading Universe…" splash** (`#loading`, in the HTML so it shows pre-JS): `main()`
+     now `await`s `preloadAllMedia()` (driving a progress bar), then `hideLoading()`.
+     Replaces the tap-to-start wall.
+   - **Both toggles default OFF** (black + silent at start) — no autoplay needed until a
+     toggle is flipped.
+   - **First Video-on couples Audio:** the Video `change` handler, on the first on,
+     sets `#audio` checked and plays — in the same gesture (so it unlocks on Safari).
+     Audio toggled on first stays audio-only.
+   - Server contract unchanged (`{off,soundtrack}`); the client now drives playback
+     directly from the ring's `scale.audio`, ignoring `render.audio`.
+3. **Verify + ship.** Rewrote the 5 E2E for the new flow (single `#aud`, no gesture,
+   loading wait, the coupling, video-off blanking). Caught a test bug — `#black.hidden`
+   is `display:none`, so `wait_for_selector` default (visible) timed out; switched to
+   `state="attached"`. **290 passed, 2 skipped; 5 E2E pass.** Verified in headless
+   **Chromium AND WebKit**: first-Video-on couples audio, soundtrack plays, video shows,
+   zero JS errors; screenshotted the splash + running app. Merged to `main`.
+
+## Deferred decisions
+
+_No low-confidence calls — all three were explicit operator requests. The
+single-element synchronous-in-gesture play is the standard Safari remedy._
+
+## Cut state
+
+- `main` @ `4ac39c9`, clean, pushed. 290 tests + 5 Playwright E2E pass.
+- **Honest caveat — still unverifiable headlessly:** every engine Playwright ships
+  (Chromium, WebKit) relaxes autoplay, so I cannot *prove* real-Safari/iOS now plays.
+  The fix is the canonical pattern (sync `play()` in the gesture, real src) and is a
+  genuine change from 0021's broken approach — but it needs the operator's real-device ear.
+
+## Operator plate
+
+- No tap wall: "Loading Universe…" → app with both toggles off (black/silent). Flip
+  **Video** → video + audio together; flip **Audio** alone → audio only.
+- If sound is STILL silent on the real device, that points to something beyond autoplay
+  (codec/MIME, file, or element error) — next session should capture `aud.error` and the
+  exact browser.
+
+## Next /goal
+
+```
+/goal confirm on a REAL device (esp. iPad/Safari) that audio is now audible when Video/Audio is toggled on and the "Loading Universe…" → app flow feels right; if still silent, capture aud.error + the exact browser. Then the deferred per-altitude music layer per docs/audio-candidate-pool.md
+```
diff --git a/sessions/0022/SESSION-0022.0-TRANSCRIPT-2026-06-26T08-40--INPROGRESS.md b/sessions/0022/SESSION-0022.0-TRANSCRIPT-2026-06-26T08-40--INPROGRESS.md
deleted file mode 100644
index 4aed62b..0000000
--- a/sessions/0022/SESSION-0022.0-TRANSCRIPT-2026-06-26T08-40--INPROGRESS.md
+++ /dev/null
@@ -1,22 +0,0 @@
-# Session 0022.0 — Transcript
-
-> App: human-experience-filter-art
-> Start: 2026-06-26T08-40 (PST)
-> Type: executing-plans
-> Posture: yolo
-> Claude-Session: e0cc053a-41fa-416d-a4f5-493f14984e75
-> Status: **PLACEHOLDER — claimed at session start; finalized at session end.**
->
-> This file reserves session ID 0022 for human-experience-filter-art. The driver replaces this
-> body with the full transcript and renames the file to its final
-> SESSION-0022.0-TRANSCRIPT-2026-06-26T08-40--<end>.md form at session end.
-
-## Launch prompt
-
-_(launch prompt not captured at claim time)_
-
-## Deferred decisions
-
-_Autonomous-mode low-confidence calls the driver made and would have
-liked operator input on. Appended as the session runs; surfaced at
-finalize. Empty if none._