[karanheart96]

WS5 (P1): rooms nobody tested.

What changed

• Wall choice moved into src/wall-choice.ts: pure geometry, unit-tested (18 tests), and shared by the runtime and the emulator tests. For plain walls the rules are unchanged: a vertical wall 0.6–4 m away, turned less than 40° from the view, with the spot straight ahead on its outline, and the stage about 33° wide.
• Plane labels. Walls are accepted as wall, wall face or WALL_FACE in any spelling, or with no label. Before, only an exact wall was accepted, so a Quest that reports wall face would never get the stage on a wall. Doors and windows are recognised by those words. Anything else (invisible wall face, wall art, screen, furniture) is not used as a wall.
• Doors and windows on the wall ahead keep the stage off them. It slides beside the opening when its centre can stay within 20° of straight ahead; otherwise the wall is refused and the paper screen is used.
• The paper screen never sits behind something. If any vertical surface straight ahead is nearer than 2 m (a wall too close to use, a window), the screen stands 15 cm in front of it, never nearer than 0.9 m. Its width scales with distance so it stays at 33°.
• Room data that arrives late still moves the stage from the paper screen onto the wall, but only in the first 12 s, before the story is under way.
• "Move stage here" in Settings, replacing "Re-place stage on the wall ahead". Pinch it, then look where you want the stage: placement waits 1.5 s so you can turn, then runs from wherever you're facing. The Settings panel sits beside the old stage, which is why the wait is needed.
• Diagnostics record the vertical plane labels the headset reports (vlabels, shown on the summary's placed: line), so the first device session settles the label question.
• Automation hooks: __sp.room(planes) for fake planes, __sp.view() for placement and UI angles from straight ahead, and __sp.where(name).

Emulator results (node scripts/sim/rooms.mjs --gpu): 17/17 pass

Case	Result
Ordinary wall at 2 m	wall, ±16.5°
No planes at all	paper screen at 2 m, after 4 s
Only floor and ceiling	paper screen at 2 m
Wall too close (0.45 m)	paper screen at 0.9 m
Wall too far (5 m)	paper screen at 2 m
Wall turned 55°	paper screen (1.85 m, in front of the wall)
Wall turned 30°	wall
Labelled "wall face"	wall
Window just right of centre	wall, stage slid left of the window
Narrow window straight ahead	paper screen at 1.83 m
Wide window straight ahead	paper screen at 1.83 m
Door frame only	paper screen at 1.85 m
Planes arrive after the timeout	the wall takes over
Planes arrive 12+ s later	stage stays on the paper screen
Recenter mid-story (head turned 60°)	the stage was 43–76° to the side, and back to centre after recenter; story continues
"Move stage here" by hand pinch, then look left	stage on the left wall, centred
Field of view, journal open	captions and journal page within ±16° wide, 11° up, 13° down

Bug caught on the way

During the refactor, the new plane reader reused the stage-normal scratch vector. When the room had a floor plane, the paper screen came out facing the ceiling. view() caught it, and it's fixed (planeNormal).

Trade-offs (conservative option chosen)

• Window straight ahead: I chose the paper screen in front of it, rather than sliding the stage more than 20° aside. A seated story is watched for minutes, so a turned head isn't worth it.
• Late room data: only the first 12 s, so the stage never jumps mid-story. "Move stage here" is the escape hatch after that.
• Not used: initiateRoomCapture (asking Quest to run Space Setup when no planes exist). It would interrupt the experience; the paper screen is the fallback.
• Settings panel size kept. It's the one thing outside the comfortable field: 27° up, 25° down and 33° to the right at a 2 m wall. Shrinking it would make the text harder to read, and it's a menu you open on purpose. Device check E9b covers it.

Verified

• npm test (now includes tests/wall-choice.test.mjs) and npm run build pass.
• playtest.mjs --gpu: six personas, no stalls, no errors, 60 fps, performance budget ok.
• live-family.mjs: passes with no errors. The Wi-Fi dropout recovered in about 2.5 s, and the story ended on the child's screen. With 30 s phases, the duet turn reached the child but didn't come back before the phase ended (the known 12 s idle timing). With 45 s phases it went to the child and came back.

Device items added (docs/DEVICE-TEST.md)

• A4b: the plane labels from the summary.
• A4c: window or door ahead.
• E6: Move stage here.
• E9: measured angles.
• E9b: Settings comfort.
• Open question: Quest's label strings.

🤖 Generated with Claude Code