Retain accepted Unicode input under queue saturation and failure

Record real focused-panel click and ASCII key delivery
Prevent Unicode input startup from deadlocking a headset switch
2026-10-24 16:02:38 +02:00 · 2026-10-24 11:47:47 +00:00 · 2026-10-24 11:23:11 +00:00 · 2026-10-24 11:19:36 +00:00 · 2026-10-24 11:13:47 +00:00 · 2026-10-24 11:10:58 +00:00
34 changed files with 2917 additions and 63 deletions
@@ -2,6 +2,39 @@

 ## Unreleased

+- Use gamescope's input for the keyboard and trackpad, following the focused
+  headset panel without installing an input service. Start KDE Connect only for
+  accents and emoji, keeping queued text ordered and cancelling it on headset
+  switches. Retain intended-headset checks before starting either control and
+  prevent Unicode helper startup from deadlocking with a headset switch.
+- Keep accepted keyboard text when the Unicode queue is full: reject new batches
+  before sending their clicks or keys, and count in-flight text toward the limit.
+  Retain unsent text after helper failures/timeouts, report the error and resume
+  it with Try again. Clean up failed workers without touching a new headset's queue.
+  [Current verification and device limits](docs/evidence/pad-gamescope-2026-10-24.md).
+
+- Keep media playback running when SteamVR refuses the optional theatre dark
+  surround, including a retry after standby. Log the refusal and play without
+  the surround. Overlay teardown attempts every owned handle and always calls
+  OpenVR shutdown even if a handle fails to close.
+
+- Check the desktop server's single-instance lock before starting background
+  downloads, so a second server exits cleanly with the existing "already
+  running" message.
+
+- Recheck saved spatial layouts on SteamOS 0.5.6 / SteamVR 2.18.2 (issue #31).
+  Read-only probes expose some panel geometry, but safe restoration of
+  Steam/gamescope docking and positions remains unverified. Record exact API
+  results and a reproducible probe; saved layouts remain unavailable.
+
+- Record packaged v0.4.17 Desktop Capture succeeding on a mapped nested KDE
+  compositor on a real Frame with SteamOS 0.5.6. Headset overlay visibility was
+  not established; this is evidence, not a new fix or resolution of issue #73.
+
+- Correct the Return to Game Mode help text: Frame Control selects the switch
+  supported by the headset's SteamOS build, including `steamosctl` on current
+  Frames. The button's behavior is unchanged.
+
 - Distinguish an unreadable gamescope panel list from a confirmed empty list in
  Desktop Capture and Control (issue #73). Failed discovery now asks users to
  check SteamVR and report the exact message and SteamOS version, rather than
@@ -75,8 +75,8 @@ Drag files onto the window to send them. Drop a game's .zip, folder or .exe to a
 **📸 Screenshots and recordings**<br>
 Browse the shots you take in the headset and save them to your Pictures folder, or have new ones and your Steam clips copied over automatically. Record what the headset shows, with its sound, or export Steam's clips at any length, as .mp4 files in your Videos folder ([not tested on a headset yet](docs/recording.md)).

-**⌨️ Type on the headset**<br>
-Type and point in the Frame's apps from your computer or phone, through KDE Connect, which Frame Control brings along and sets up on the Frame. Copy text to the headset's clipboard and back again. Nothing else to install, anywhere. Press T to open it.
+**⌨️ Keyboard and trackpad**<br>
+Type and point in the headset from your computer or phone, through Valve's own input path, with nothing to install. Accents and emoji use KDE Connect, which Frame Control brings along and sets up the first time you type one. Copy text to the headset's clipboard and back again. Press T to open it.

 </td>
 </tr>
@@ -1,12 +1,12 @@
 {
  "name": "frame-control",
-  "version": "0.4.17",
+  "version": "0.4.18",
  "lockfileVersion": 3,
  "requires": true,
  "packages": {
    "": {
      "name": "frame-control",
-      "version": "0.4.17",
+      "version": "0.4.18",
      "license": "MIT",
      "devDependencies": {
        "electron": "^44.4.5",
@@ -1,7 +1,7 @@
 {
  "name": "frame-control",
  "productName": "Frame Control",
-  "version": "0.4.17",
+  "version": "0.4.18",
  "description": "Desktop app for managing a Valve Steam Frame over SSH",
  "private": true,
  "main": "main.js",
@@ -25,8 +25,28 @@ tested in another full Plasma session under a three-minute rollback timer.
 `ui/frame_backup.py session` reported Desktop Mode despite gamescope, and
 `ui/frame_backup.py game-mode` invoked the same guarded Return function used by
 the button. Game Mode became active and a second call made no switch. Battery
-remained at 99%; the timer and lock were removed. The literal desktop button
-was not clicked in this test, and older SteamOS builds remain fixture-tested.
+remained at 99%; the timer and lock were removed. That test used the production
+CLI rather than clicking the desktop button.
+
+Later on 2026-10-24, the published **v0.4.17 Linux x86_64 AppImage** passed the
+literal packaged-button check on the same Frame/build. With the shared lock and
+a verified three-minute rollback timer, the headset entered full Plasma Desktop
+Mode. In the packaged Electron window, **Tools → Backup & recovery → Check**
+reported Desktop Mode despite gamescope. A trusted automated pointer click on
+the visible **Return to Game Mode** button, accepting its confirmation, returned
+`switched: true` and `steamosctl switch-to-game-mode`. The systemd targets then
+confirmed Game Mode. A second confirmed button click returned `switched: false`;
+the Game Mode service invocation and SteamVR PID stayed unchanged. **Check**
+then reported Game Mode and disabled Return. The test app/server and virtual
+display were stopped; the timer and shared lock were removed. Steam and SteamVR remained running
+at 99% battery. [Commands, responses, cleanup and limits](evidence/return-game-mode-packaged-2026-10-24.json),
+[before](evidence/return-game-mode-packaged-2026-10-24-before.png) and
+[after](evidence/return-game-mode-packaged-2026-10-24-after.png).
+
+This used the unchanged extracted AppImage payload with its bundled Python on
+Xvfb (`--no-sandbox --disable-gpu`), following the packaged Export test in #156.
+FUSE mounting, desktop integration, hardware rendering and the view inside a
+worn headset were not checked. Older SteamOS builds remain fixture-tested.

 ## Back up

@@ -0,0 +1,206 @@
+{
+  "date_utc": "2026-10-24",
+  "base_commit": "de931bc",
+  "issue": 73,
+  "result": "Packaged Desktop Capture passed for a mapped nested KDE compositor on the real Frame; wearer-visible overlay condition was not established. Do not close #73 from this test.",
+  "package": {
+    "tag": "v0.4.17",
+    "asset": "Frame-Control-linux-x86_64.AppImage",
+    "sha256": "b1b8eb82bf73a227b2cd36e8f0ef3a2ebdf7967c5853efb08a9634f38227015d",
+    "release_digest_matched": true,
+    "panel_helper_matches_base": true,
+    "unmodified_payload": true,
+    "host": "lxso2",
+    "display": "Xvfb :139",
+    "flags": [
+      "--no-sandbox",
+      "--disable-gpu"
+    ],
+    "private_profile_and_data": true,
+    "telemetry_and_update_checks_disabled": true
+  },
+  "preflight": {
+    "steamos_version": "0.5.6",
+    "build_id": "20261015.6141207",
+    "battery_percent": 99,
+    "battery_status": "Not charging",
+    "game_mode": true,
+    "full_plasma_session": false,
+    "nested_desktop_running": false,
+    "nested_runtime_present": false,
+    "lock_present": false,
+    "steam_pid": 297826,
+    "vrserver_pid": 297506,
+    "vrcompositor_pid": 297620,
+    "gamescope_service_invocation": "e8c1085b3bbd4e79a323b3e4a9bc6355",
+    "focusable_windows": [],
+    "focused_windows": []
+  },
+  "session": {
+    "started_utc": "2026-10-24T09:43:48Z",
+    "lock_acquired": true,
+    "launch": [
+      "systemd-run",
+      "--user",
+      "--unit=fc-issue73-desktop-20261024",
+      "--collect",
+      "--property=RuntimeMaxSec=180s",
+      "--property=TimeoutStopSec=15s",
+      "--property=KillMode=control-group",
+      "--setenv=DISPLAY=:0",
+      "--setenv=XDG_RUNTIME_DIR=/run/user/1000",
+      "--setenv=DBUS_SESSION_BUS_ADDRESS=unix:path=/run/user/1000/bus",
+      "--setenv=XDG_SESSION_TYPE=x11",
+      "--setenv=XDG_CURRENT_DESKTOP=gamescope",
+      "/usr/bin/steamos-nested-desktop"
+    ],
+    "response": "Running as unit: fc-issue73-desktop-20261024.service; invocation ID: 92cff1e57c794212997d46c0a68877ec\n"
+  },
+  "discovery": {
+    "focus": 62914560,
+    "focus_display": ":0",
+    "panels": [
+      {
+        "window": 62914560,
+        "display": ":0",
+        "name": "KDE Wayland Compositor - Press right control key to grab input",
+        "x": 0,
+        "y": 0,
+        "width": 1920,
+        "height": 1080,
+        "app": 0,
+        "focused": true
+      }
+    ]
+  },
+  "window_evidence": {
+    "xwininfo_map_state": "IsViewable",
+    "width": 1920,
+    "height": 1080,
+    "focused_after_show_request": true,
+    "steamvr_overlay_key": "valve.steam.desktopgame.0",
+    "steamvr_overlay_visibility_after_show_request": "not_visible"
+  },
+  "capture": {
+    "repeat_utc": "2026-10-24T09:44:47.323Z",
+    "route": "/api/screenshot?view=panel&window=62914560&display=%3A0",
+    "http": 200,
+    "content_type": "image/png",
+    "content_length": 1209980,
+    "source_header": "panel",
+    "decoded": {
+      "width": 1920,
+      "height": 1080,
+      "source": "panel",
+      "selected": {
+        "window": 62914560,
+        "display": ":0",
+        "name": "KDE Wayland Compositor - Press right control key to grab input",
+        "x": 0,
+        "y": 0,
+        "width": 1920,
+        "height": 1080,
+        "app": 0,
+        "focused": true
+      },
+      "stamp": "9:44:48 AM",
+      "badge": "Desktop \u00b7 KDE Wayland Compositor - Press right control key to grab input"
+    },
+    "canvas_png_bytes": 2011278,
+    "first_capture_pixel_statistics": {
+      "width": 1920,
+      "height": 1080,
+      "min": 2.3333333333333335,
+      "max": 255,
+      "hidden": false
+    },
+    "first_capture_click_event": {
+      "isTrusted": true,
+      "utc": "2026-10-24T09:44:12.738Z"
+    },
+    "actions": "Two Playwright locator.click() actions on the visible #shotBtn in the actual packaged Electron window; first click recorded isTrusted=true. No direct screenshot API invocation by harness."
+  },
+  "device_cleanup": {
+    "utc": "2026-10-24T09:45:33Z",
+    "before_stop": {
+      "exit": 0,
+      "out": "MainPID=300739\nControlGroup=/user.slice/user-1000.slice/user@1000.service/app.slice/fc-issue73-desktop-20261024.service\nActiveState=active\n",
+      "err": ""
+    },
+    "hide_dashboard": {
+      "exit": 0,
+      "out": "HideDashboard",
+      "err": ""
+    },
+    "stop_owned_unit": {
+      "exit": 0,
+      "out": "",
+      "err": ""
+    },
+    "after_stop": {
+      "exit": 0,
+      "out": "MainPID=0\nLoadState=not-found\nActiveState=inactive\n",
+      "err": ""
+    },
+    "desktop_processes": {
+      "exit": 1,
+      "out": "",
+      "err": ""
+    },
+    "nested_runtime_exists": false,
+    "session": {
+      "exit": 0,
+      "out": "Id=gamescope-session.target\nActiveState=active\nInvocationID=a2fee8ebf55c42c7b546390d14cd985c\n\nId=plasma-workspace-wayland.target\nActiveState=inactive\nInvocationID=\n\nId=gamescope-session.service\nActiveState=active\nInvocationID=e8c1085b3bbd4e79a323b3e4a9bc6355\n",
+      "err": ""
+    },
+    "steam": {
+      "exit": 0,
+      "out": "297826 /home/steamos/.local/share/Steam/steamrtarm64/steam -cef-enable-debugging -deckard -gamepadui -steamdeck -steamos3 -vrgamepadui\n",
+      "err": ""
+    },
+    "steamvr": {
+      "exit": 0,
+      "out": "297506 /opt/steamvr/bin/linuxarm64/vrserver -waitformonitor -ipccontrolfilepath /tmp/SteamVR-IPCControlFile-297420\n297620 /opt/steamvr/bin/linuxarm64/vrcompositor\n",
+      "err": ""
+    },
+    "xprops": {
+      "exit": 0,
+      "out": "GAMESCOPE_FOCUSABLE_WINDOWS(CARDINAL) = \nGAMESCOPE_FOCUSED_WINDOW(CARDINAL) = \n",
+      "err": ""
+    },
+    "battery": {
+      "capacity": "99",
+      "status": "Not charging"
+    },
+    "owned_lock_released": true
+  },
+  "host_cleanup": {
+    "electron": {
+      "pid": 1578431,
+      "exists_after": false
+    },
+    "xvfb": {
+      "pid": 1578428,
+      "exists_after": false
+    },
+    "x_socket_absent": true
+  },
+  "delayed_final_readback": "2026-10-24T09:46:39Z\nshared-lock-absent\nLoadState=not-found\nActiveState=inactive\nId=gamescope-session.target\nActiveState=active\n\nId=plasma-workspace-wayland.target\nActiveState=inactive\nInvocationID=e8c1085b3bbd4e79a323b3e4a9bc6355\n297826 /home/steamos/.local/share/Steam/steamrtarm64/steam -cef-enable-debugging -deckard -gamepadui -steamdeck -steamos3 -vrgamepadui\n297506 /opt/steamvr/bin/linuxarm64/vrserver -waitformonitor -ipccontrolfilepath /tmp/SteamVR-IPCControlFile-297420\n297620 /opt/steamvr/bin/linuxarm64/vrcompositor\nnested-runtime-absent\n100\nFull\nGAMESCOPE_FOCUSABLE_WINDOWS(CARDINAL) = \nGAMESCOPE_FOCUSED_WINDOW(CARDINAL) = \n",
+  "validation": {
+    "focused_panel_suite": "42 tests passed in 14.805s (test_touch.py), including production-page Node behavior tests",
+    "website": "2 tests passed (node --test site/test/*.test.mjs)",
+    "node_panels": "Passed independently (node tests/page/desk_panels.mjs)",
+    "metadata_and_links": "JSON parsed; result/cleanup assertions and relative evidence/streaming/testing links passed",
+    "diff": "git diff --check passed"
+  },
+  "limits": [
+    "SteamVR continued reporting not_visible after ShowDashboard, despite mapped X window and confirmed focus. No worn-headset visual assessment; does not establish the reporter's visible-overlay condition.",
+    "SteamOS 0.5.6 only; reporter used 0.4.5 with unknown Frame Control version and no reproduction steps.",
+    "Nested desktop launched with SteamOS steamos-nested-desktop from a bounded transient user unit using Steam display/session variables; not launched through a wearer's Steam UI.",
+    "Extracted unchanged AppImage on a virtual X display with Chromium sandbox and GPU disabled; no FUSE, physical monitor or hardware-rendering test.",
+    "No Live video, Control input or full Desktop Mode test.",
+    "Nested-desktop journal included gamescope WSI/Vulkan-hook warnings. No warning was dismissed; captured compositor pixels were nonblank.",
+    "Initial CDP response.body() yielded zero bytes despite successful image decode. Repeat verified Content-Length, fresh decoded image, panel selection and PNG export from canvas; the initial raw-body size is not treated as capture failure or proof.",
+    "No private desktop pixels, issue screenshot or raw desktop journal committed."
+  ]
+}
@@ -0,0 +1,100 @@
+# Packaged Desktop Capture, 2026-10-24
+
+Base: `de931bc`. Package: Linux x86_64 **v0.4.17**. Device: real Frame,
+SteamOS **0.5.6**, BUILD_ID **20261015.6141207**. Host: lxso2.
+[Sanitized state and response evidence](desktop-capture-packaged-2026-10-24.json).
+
+## Result and issue boundary
+
+The packaged app discovered the nested KDE compositor and both automated
+Desktop **Capture** button clicks succeeded. The repeat returned HTTP 200,
+`Content-Type: image/png`, `X-Capture-Source: panel`, and Content-Length
+1,209,980. The app decoded a fresh, nonblank 1920×1080 image and displayed
+the Desktop panel badge. No further product failure was reproduced or fixed.
+
+This does **not** establish the full visible-desktop condition in
+[#73](https://github.com/saphid/frame-control/issues/73). `xwininfo` reported
+the desktop window `IsViewable`, and gamescope confirmed focus after
+`vrcmd --showdashboard valve.steam.desktopgame.0`. However, SteamVR still
+reported that overlay `not_visible`. There was no worn-headset visual check.
+The observation proves capture of a mapped nested compositor, not its visible
+presentation to a wearer. Do not close #73 from this evidence.
+
+The issue screenshot shows a desktop panel in the headset view while Desktop
+Capture logs “no app panel is open in the headset.” The reporter used SteamOS
+0.4.5, gave no Frame Control version or reproduction steps, and has not replied
+to the maintainer's 0.4.3 update request. The duplicate page function was already
+fixed; [PR #152](https://github.com/saphid/frame-control/pull/152) subsequently
+improved unknown-state diagnostics. Its [earlier investigation](panel-discovery-2026-10-24.md)
+was blocked from starting a desktop by another task's lock. This run adds a
+current-package capture observation, without reproducing the older failure.
+
+## Procedure
+
+1. Read-only preflight found Game Mode active, full Plasma inactive, Steam and
+   SteamVR running, battery 99%, no nested desktop/runtime and no test lock.
+2. Compared the retained AppImage's SHA-256 with the live v0.4.17 release asset
+   digest: `b1b8eb82bf73a227b2cd36e8f0ef3a2ebdf7967c5853efb08a9634f38227015d`.
+   Extracted an independent copy; `cmp` confirmed its `frame_touch.py` matches
+   the base checkout. No payload was edited.
+3. Started the package's own AppRun, bundled server and Python on Xvfb `:139`
+   with private config/cache/data/profile, telemetry and update checks disabled,
+   and `--no-sandbox --disable-gpu`. Used the existing authorized `ssh frame`
+   identity. T3 preview opened the package server page but does not attach to
+   the existing Electron window. The literal Electron button test used the
+   same Playwright-over-loopback-CDP workflow as the earlier packaged tests.
+4. Acquired `/tmp/frame-test.lock` with atomic `mkdir` and wrote an owner marker.
+   Rechecked no nested runtime or Plasma was present. Started SteamOS's
+   `/usr/bin/steamos-nested-desktop` in the owned transient user unit
+   `fc-issue73-desktop-20261024.service`, using Steam's display/session variables.
+   `RuntimeMaxSec=180s`, `TimeoutStopSec=15s` and `KillMode=control-group` bounded
+   the session. This launches a desktop panel inside Game Mode; no full Desktop
+   Mode switch was sent.
+5. In the actual packaged Electron window, navigated Tools → Capture & Live →
+   Desktop and clicked Capture with `locator.click()`. The first click event
+   recorded `isTrusted=true`. The page selected the KDE compositor on `:0`;
+   the production screenshot path used window `62914560`. Requested the desktop
+   dashboard, read back focus/visibility, and clicked Capture again. Recorded
+   response metadata, fresh decoded image dimensions and canvas pixel range.
+   No direct screenshot API call was substituted for these button clicks.
+6. Ran the packaged `frame_touch.py panels` read-only over SSH to retain its
+   exact discovery result: one focused KDE compositor panel, app ID 0, 1920×1080,
+   display `:0`, with no diagnostic `why` field.
+
+## Cleanup and limits
+
+The owned desktop session ran **09:43:48–09:45:33 UTC** (105 seconds). Hid the
+dashboard to restore its initial hidden state, stopped only the owned unit,
+and verified it was unloaded/inactive with MainPID 0. No Plasma/KWin desktop
+process or nested runtime remained; gamescope's focusable/focused lists were
+empty again. Released only this task's lock, after checking its owner marker.
+
+Delayed readback at **09:46:39 UTC** confirmed the lock absent, Game Mode active,
+full Plasma inactive and the original Steam, vrserver and vrcompositor PIDs
+still running. The Game Mode service invocation was unchanged. Battery was
+99% at cleanup and 100%/Full on the delayed check. Closed both owned preview
+tabs, Electron and Xvfb; their processes and X socket were gone. No test server
+remained. No reboot, installation, security/global setting change or full
+Desktop Mode transition was performed.
+
+The nested desktop journal included gamescope WSI/Vulkan-hook warnings. No
+warning was dismissed. Capture produced nonblank compositor pixels, but this
+run does not assess their usability to a wearer. Private image files stayed
+outside the repository; only state and response metadata are published.
+
+The initial CDP `response.body()` read yielded zero bytes despite successful
+decoding in the page. The repeat verified the PNG Content-Length, fresh image
+dimensions/source/selection and a 2,011,278-byte PNG exported from the canvas.
+The initial raw-body measurement is a harness limitation, not evidence of a
+product capture failure. No Live video, Control input, physical monitor, FUSE
+mounting or hardware rendering was tested. Older SteamOS and other packages
+remain outside this observation.
+
+## Offline verification
+
+`python3 -m unittest discover -s tests -p test_touch.py -v`: **42 tests passed**,
+including the production page's Node panel behavior tests, duplicate-function
+guard, absent versus empty panel diagnostics, alternate displays, capture/input
+target validation and discovery deadlines. JSON parsing and `git diff --check`
+were also checked. No new regression test was added because this is an
+evidence-only change with no reproduced product defect.
@@ -0,0 +1,114 @@
+# Linux PC host feasibility on lxso2, 2026-10-24
+
+**Blocked on this host's current desktop environment. No feature enabled and
+no feature PR opened.** Investigation for [issue #23](https://github.com/saphid/frame-control/issues/23),
+starting at `4e45648` (main after #154). This concerns flat desktop panels,
+not Linux PC VR streaming.
+
+## Decision
+
+The smallest useful candidate is one explicitly selected **whole screen**,
+captured through the ScreenCast portal's restricted PipeWire connection,
+encoded locally with GStreamer/x264, and sent to the existing viewer through
+the shared SSH controller. Input, individual windows and virtual displays
+would remain unsupported until separately verified. This candidate was **not
+implemented**: lxso2 currently cannot exercise the real capture permission path.
+
+Development packages could unblock compilation, but would not create a desktop,
+portal backend or consented screen. Synthetic encoding, an Xvfb screen grab or
+using the Frame as its own host would not verify Linux PC desktop capture.
+No host packages or services were changed.
+
+## Evidence
+
+Host: `lxso2`, Linux x86-64, kernel `6.8.0-139-generic`.
+
+- `DISPLAY`, `WAYLAND_DISPLAY` and `XDG_SESSION_TYPE` were unset. The user bus
+  was reachable at `/run/user/1000/bus`.
+- `/run/user/1000/pipewire-0` and `/run/user/1000/wayland-0` did not exist.
+  No processes matched Xorg, Xwayland, Xvfb, GNOME Shell, KWin Wayland, Weston,
+  PipeWire, WirePlumber or xdg-desktop-portal at the process check.
+- D-Bus `ListNames` returned D-Bus, systemd and anonymous connections, with no
+  `org.freedesktop.portal.Desktop`. `ListActivatableNames` also had no desktop
+  portal. These calls queried D-Bus itself without activating capture services.
+- `dpkg-query` found no `xdg-desktop-portal*` or `pipewire*` packages.
+  `libgstreamer1.0-0:amd64` was installed at `1.24.2-1ubuntu0.1`, but
+  `gstreamer1.0-tools` was uninstalled. `gst-launch-1.0` and `gst-inspect-1.0`
+  were absent from PATH. Xvfb was installed, but supplies no portal by itself.
+- `pkg-config --modversion` failed for `gstreamer-1.0`,
+  `gstreamer-app-1.0`, `gstreamer-video-1.0` and `libpipewire-0.3`.
+  `gio-unix-2.0` reported `2.80.0`.
+- `desktop/bundle/pc-host.so` did not exist.
+  `python3 desktop/build.py --with-x264 --out /tmp/frame-linux-pc-feasibility`
+  exited **1** before compilation: required GStreamer pkg-config modules were
+  missing. No native build or encoder runtime success is claimed.
+
+Reproduce the non-activating portal checks in the intended desktop user's
+session, using that session's actual runtime directory and bus:
+
+```sh
+gdbus call --session --dest org.freedesktop.DBus \
+  --object-path /org/freedesktop/DBus --method org.freedesktop.DBus.ListNames
+gdbus call --session --dest org.freedesktop.DBus \
+  --object-path /org/freedesktop/DBus --method org.freedesktop.DBus.ListActivatableNames
+pkg-config --modversion gstreamer-1.0 gstreamer-app-1.0 gstreamer-video-1.0 gio-unix-2.0
+```
+
+## Existing implementation reviewed
+
+- `ui/frame_pcview.py` selects PCView only on Windows and rejects Linux even
+  when enabled. Linux installers do not carry its native bundle.
+- `desktop/portal.c` creates a **RemoteDesktop** session, requests pointer and
+  keyboard devices, and selects monitors and windows (`types=3`). It is not
+  ScreenCast-only or screen-only. Its restricted fd/node and fresh-connection
+  handling are reusable.
+- `ui/frame_pc_agent.py` calls the selection a shared window or screen and
+  picks VA-API when the element is present, falling back to x264. A bounded
+  x264 slice needs explicit encoder policy and unsupported-input behavior;
+  exposing the existing Linux path alone would not meet that scope.
+- `app/main.js` starts Python with the inherited environment; it does not
+  supply a compositor, portal backend or capture permission.
+- `.github/workflows/pc-host.yml` installs native dependencies and tests
+  synthetic x264/protocol behavior with fake capture consent/input. This does
+  not establish interactive Linux desktop capture.
+- Issue #23 and its six existing comments distinguish prior synthetic
+  helper/viewer results from untested Linux desktop capture and consent.
+
+## Verification and limits
+
+`python3 -m unittest discover -s tests -p 'test_pc*.py' -v` exited **0**:
+27 tests ran successfully, with one additional native-class setup skip
+(`native PC libraries not built on this host`). Coverage includes the compiled
+C controller trace, Python adapters, grants and default-off/Windows-only gates.
+It does not prove native GStreamer capture or local-client streaming.
+
+No fake/local native stream, consent prompt, real desktop capture, Frame viewer
+run, benchmark, input or window mode was tested in this investigation. No SSH
+connection to the Frame was needed; no lock was acquired and no headset install,
+launch, stop or settings change was attempted. Issue #113 was left alone.
+No issue comments or other public messages were sent.
+
+## Concrete unblock and next slice
+
+1. Supply an interactive Linux desktop, physical or VM, with a working
+   ScreenCast portal/backend, PipeWire and accessible consent dialog. Installing
+   the portal frontend alone is insufficient. A VM can establish software
+   capture/x264 behavior while physical-host behavior remains unverified.
+2. Install the native dependencies listed in `pc-host.yml` in that environment
+   and build a local `--with-x264` bundle. Resolve x264 distribution
+   licensing/source obligations before shipping it: the current build labels
+   that option test-only and the Windows installer omits it.
+3. Implement a ScreenCast-only session with monitor-only selection, no
+   RemoteDesktop/device request, disabled input in protocol and viewer, hidden
+   window/virtual-display modes, and explicit x264 selection. Preserve the
+   default-off gate and existing Mac/Windows paths.
+4. Verify consent acceptance/refusal/cancellation, Stop/reopen, connection loss
+   and portal closure with a local client decoding actual captured content.
+   Run the shared benchmark and record results without inferring hardware or
+   physical laser behavior.
+5. Only after local capture works, acquire the shared Frame lock once and
+   follow [the headset procedure](../testing.md#headset-smoke-test) for a brief
+   viewer test. If occupied, leave the lock alone and defer device testing.
+
+These are prerequisites and proposed scope, not a ready feature. The Linux
+host request in #23 remains open.
@@ -0,0 +1,142 @@
+# Media surround hardening: 2026-10-24
+
+PR [#51](https://github.com/saphid/frame-control/pull/51), branch
+`theatre-media-surround-hardening`, merged main `2ded4c5` without rebasing.
+Both fixes remain necessary on that main: an optional-surround error escapes
+`play()`, and a non-`RuntimeError` during overlay teardown skips the remaining
+handles and OpenVR shutdown.
+
+## Offline checks
+
+- Focused media suite: **23 tests passed** on Linux / Python 3.12.3.
+- Final full suite: **1,381 tests, OK (15 skipped)** in 145.384 s with
+  `PYTHONFAULTHANDLER=1 python3 -m unittest discover -s tests -v`.
+  The service-regression module passed **33 tests**. Player/server compilation
+  and `git diff --check` passed. Current macOS/Windows validation is left to CI.
+- The startup/upload, creation/setup, deferred-surround refusal and teardown
+  regression tests were also run with main's player implementation: they
+  failed as expected (six errors across subtests). The screen-refusal test
+  passed, confirming that actual screen failures still propagate.
+- The historical macOS failure was job
+  [109239733717](https://github.com/saphid/frame-control/actions/runs/36516462623/job/109239733717),
+  on `dac9dbc`, Python 3.12: 619 tests, three failures in artwork resolver
+  tests (`test_resolver_slot_released_when_thread_cannot_start`,
+  `test_timed_out_lookups_are_capped`,
+  `test_url_fetch_refuses_private_hosts_and_honours_deadline`). Media tests
+  passed. Late DNS workers retained semaphore slots; the first failed
+  assertion could also leak acquired slots into later tests.
+- Main already contains `8051687`'s wait-for-idle and unconditional slot
+  cleanup, preserved by `d546568`. The current artwork module passed all
+  **40 tests** locally. No additional artwork changes were needed.
+- The first full run completed **1,380 tests, 15 skipped**, with one failure:
+  `test_a_second_server_is_refused_until_the_first_exits` observed SIGSEGV
+  (`-11`) instead of exit 1. All 41 server tests passed on a focused rerun.
+  Main started daemon index/TLS workers before refusing a second server;
+  its existing shutdown notes document a race with OpenSSL interpreter exit.
+  This crash mechanism is inferred; the first run did not retain a native
+  fault trace. The subprocess assertion now retains stderr for diagnosis.
+  A deterministic regression now checks that rejection starts no listener,
+  cleanup sweep or background workers. It failed before moving the ownership
+  check ahead of startup work and passed afterwards.
+
+The surround retry now logs a refusal before discarding that cosmetic work.
+Creation, setup and first-upload failures also allow the screen to play;
+standby remains retryable, and screen errors remain fatal. Teardown tests
+exercise OpenVR errors, unexpected exceptions and a late `InterruptedError`,
+checking that all handles are attempted and shutdown runs.
+
+## Earlier blocked real Frame attempt
+
+Read-only SSH preflight: SteamOS **0.5.6**, BUILD_ID **20261015.6141207**;
+SteamVR server PID **297506**, compositor PID **297620**; existing media
+service `inactive/dead`, MainPID 0; battery **99%**, fuel gauge `Not charging`,
+charger `Full`. The shared lock was absent at preflight.
+
+The subsequent atomic `mkdir /tmp/frame-test.lock` failed with
+`FileExistsError(17, 'File exists')`: another session acquired it in between.
+The probe stopped before uploading files, launching media, creating overlays,
+or changing any state. The foreign lock was neither modified nor released.
+All generated fixtures remained on the computer and were then removed; no
+headset cleanup was necessary. Steam/SteamVR and the prior player state were
+left alone.
+
+## Follow-up real Frame smoke at the exact PR head
+
+**Passed, 2026-10-24 10:35:02–10:35:23 UTC**, with byte-identical media modules
+from **`5d551a81658d9ff14f49285c28609d506cbc20aa`**. This later run resolves
+the earlier lack of fresh playback evidence. No production code changed.
+The headset was unworn. SteamOS **0.5.6**, BUILD_ID **20261015.6141207**,
+SteamVR **2.18.2**; battery **100%, Full** before and after.
+
+Read-only preflight found Game Mode (`XDG_CURRENT_DESKTOP=gamescope`,
+`start-gamescope` running, no `plasmashell`), Steam/SteamVR running, no media
+overlays, and the installed media service inactive/not loaded. The test then
+acquired `/tmp/frame-test.lock` with atomic `mkdir`; it did not take over a
+foreign lock. Under the lock, those safety conditions were checked again.
+
+Only generated media and the PR's four unchanged modules were uploaded, under
+the test-owned, uid-1000 directory
+`/tmp/pr51-5d551a8-4c0976a52564-utfbk_m0/`. The clip was a **12 s, 30 fps,
+960×540 half-SBS H.264** MP4 with solid red/cyan eye images and **silent AAC**.
+It was generated with FFmpeg's `color`, `hstack` and `anullsrc` sources; no
+personal or external media was used.
+
+The exact `frame_media_remote.run()` Play/status/Stop path was exercised by
+importing its unmodified module and changing only its in-memory `ROOT`,
+`RUNTIME`, `STATUS` and `UNIT` constants to the isolated test locations and
+`frame-control-media-pr51-5d551a8-4c0976a52564.service`. The real
+`frame_media_player.py` ran unchanged with `--layout auto --theatre`, using
+`h264_v4l2m2m` and Pulse output. This checks the player and its systemd lifecycle;
+it does **not** reverify the HTTP/UI upload or runtime installer, nor the
+literal production unit name. The installed media runtime, log, status and
+library were preserved.
+
+- First Play: `starting` → `playing` → **`ended`**, **360 frames**, **160 dropped**,
+  **12.0397 s**, `running=false`. The drops were the normal unworn/standby
+  `SetOverlayRaw` retry path; no fault was forced. At 60 frames, dropped was 0.
+  `vrcmd --overlays` listed `framecontrol.media.surround` as **1×1 visible**
+  and `framecontrol.media.screen` as **visible**. These are SteamVR metadata,
+  not a wearer or eye-capture verification.
+- After EOF: unit **inactive/dead, not-found, MainPID 0**, no test player or
+  decoder, and neither owned overlay remained. Stop after EOF succeeded and
+  retained `ended`.
+- Second Play reached `playing`, 60 frames, 0 dropped. Explicit Stop returned
+  **`stopped`, 72 frames, 0 dropped, `running=false`**. Again the unit was
+  inactive/dead/not-found with MainPID 0, no player/decoder and no owned overlays.
+- The test unit's `ExecStart` and cgroup identified the unique temporary
+  runtime. First-run player/decoder PIDs were **309650/309655**; second-run
+  PIDs were **309805/309810**. Stop controlled only this dedicated test unit.
+  Its player log was empty; no surround refusal or teardown error was logged.
+
+### Uploaded SHA-256 values
+
+```text
+generated_SBS.mp4      1b39cac1d3b092b89657250c817019b68e87a474976ccd44aeb38e8eee44ffca
+frame_media.py         001e3158f4c3931794846b21f1e4cb553349d9326b4db1454a01976d446ad97a
+frame_media_player.py  86b0a140ebff857c2819c74d3ac45d1897735c585d44c7360ad6f18c2ecf2645
+frame_media_remote.py  b60386eb192d32594358246a4e665474c543600f2c564c2efd554ef38d7a631d
+frame_splat.py         f35750926f1149c41b95617a302012406d355c0dfdd4530438b0c42ee6d55daa
+```
+
+### Cleanup and limits
+
+The temporary directory, including its fixture, code, bytecode, status and log,
+was removed only after verifying that its player and decoder had exited. The
+transient unit was collected. The test released only its own lock after cleanup.
+All installed media runtime file hashes, including `player.log` and
+`status.json`, were unchanged. The installed `frame-control-media.service`
+remained inactive/not loaded. Steam **297826**, vrserver **297506**,
+vrcompositor **297620**, vrdashboard **297721**, and start-gamescope **296997**
+kept the same PIDs. The pre-existing overlay key/dimension/visibility/type
+inventory matched the final inventory, and all dashboard main panels remained
+`not_visible`. No dashboard command, wake request, global setting, reboot or
+Steam/SteamVR stop was issued.
+
+Cosmetic surround refusal and exceptional teardown remain **fake-tested**;
+no real refusal or teardown fault was induced. This short unworn-headset
+smoke does not verify appearance, eye isolation, comfort, audible sound/lip
+sync (the AAC was silent), sustained display performance, or the issue #28
+report of laggy 1080p video. It does not cover new H.265, photo or splat cases.
+Historical checks remain in the earlier PR #51 description and
+[media playback notes](../vr-video.md). The PR remains draft and unmerged
+for parent review and CI.
@@ -0,0 +1,178 @@
+# Keyboard and trackpad through gamescope, 2026-10-24
+
+PR: [#66](https://github.com/saphid/frame-control/pull/66), `pad-gamescope`.
+Parent/main: v0.4.18 commit `0da1fe5`. Merge commit: `1737366`, with parents
+`cba23fb` (the existing PR) and `0da1fe5` (main). No history was rewritten.
+
+## Merge and implementation
+
+Both `/api/input` (keyboard/trackpad) and `/api/touch` (live-view Control)
+start and poll the gamescope helper. KDE Connect starts only when Unicode
+typing needs it. The existing ordered, bounded queue and switch cancellation
+remain in place; cross-service typing order is an estimate, not an application
+acknowledgement.
+
+The merge keeps main's intended-device validation and work reservation before
+a GET starts an input helper, immutable SSH routes, launch generations and
+serialized writes. Switching headsets also stops the power report and resets
+both input agents and queued typing before publishing the new route. The page
+keeps main's Tools sections, clipboard controls and single instances of the
+keyboard/trackpad and VR tools, with the PR's gamescope help text.
+
+The start/status regression tests now exercise the gamescope helper for both
+endpoints. An explicit `/api/input` test checks that start and poll use that
+helper without starting KDE Connect. Existing stale/missing-headset and
+route-reservation tests remain active.
+
+## Real Frame: focused click and ASCII keys received
+
+Host: lxso2, `ssh frame`. SteamOS **0.5.6**, BUILD_ID **20261015.6141207**,
+SteamVR **2.18.2**. At **11:05 UTC**, Game Mode was active, full Plasma was
+inactive, Steam and SteamVR were running, battery was **99%**, and gamescope's
+focusable/focused window lists were empty. The shared test lock was absent.
+Existing X11 tools and libX11 were available; Tk could not load its library.
+
+At **11:09:23 UTC**, an initial read-only preflight saw no focused window
+but activity level **0** (idle), rather than the required **3** (standby), and
+aborted before lock acquisition or input. At **11:09:57 UTC**, another task
+held the shared lock with a focused panel; that lock and panel were left alone.
+
+At **11:20:22 UTC**, read-only checks found the lock absent, no focused or
+focusable windows, activity **3** and battery **98%**. The bounded probe ran at
+**11:20:37 UTC** against implementation commit `5641d3c`:
+
+1. Rechecked empty focus and standby, acquired `/tmp/frame-test.lock` with
+   atomic `mkdir`, wrote a unique owner marker and rechecked both conditions.
+2. Created only a disposable 640×400 X11 event-recorder window on `:0`, tagged
+   with test app ID `2000999066`. It had a 45-second alarm and closed on stdin
+   EOF. No existing application, full desktop or dashboard was started.
+3. Waited for gamescope to report that exact window (`62914561`) focused.
+   Started the production server's `TouchAgent` over SSH; it reported ready.
+4. Called production `server.remote_touch` for positioning, then
+   `server.remote_input` with a left click, `Frame66` and the Enter key button.
+   Both responses reported `sent: true`. More importantly, the **window itself
+   recorded ButtonPress/Release (button 1), the ASCII key events in order and
+   Return**. `XLookupString` decoded exactly `Frame66\r`. The events were
+   delivered through gamescope's EIS socket, not fabricated with XSendEvent.
+5. Closed the helper by EOF and waited for exit, closed the recorder, checked
+   empty focus and standby, then removed only the owner-matching test lock.
+
+[State and received events](pad-gamescope-2026-10-24/readback.json) and the
+[probe](pad-gamescope-2026-10-24/probe.py)/[X11 recorder](pad-gamescope-2026-10-24/recorder.py)
+are retained. The archived scripts use checkout-relative source paths; the
+recorded run used the same harness from `/tmp`. Motion records omit an unused
+field that only has meaning for key/button events.
+
+Delayed readback at **11:21:46 UTC** confirmed no test lock or recorder process,
+empty focused/focusable lists, Game Mode active, full Plasma inactive, activity
+3 and battery 98%. The original Steam and vrserver PIDs were unchanged. No
+install, security/global setting change, dashboard/session switch, reset or
+reboot was performed. No headset helper or owned lock remained.
+
+This establishes **focused-panel click and ASCII key delivery on `:0`**.
+It does not establish wearer-visible presentation, Unicode delivery or GUI
+interaction. The absolute-position setup reached the recorder's edge
+(639,399), rather than its centre; pixel-accurate absolute pointing is **not**
+proved by this small-window probe. The trackpad click and typed keys still
+arrived in the owned focused window. The original PR's earlier acceptance-only
+observations remain historical; this observation adds actual window events.
+
+## Offline checks and remaining limits
+
+- `python3 -m unittest discover -s tests -p test_input.py -v`: **58 passed**.
+- `python3 -m unittest discover -s tests -p test_touch.py -v`: **43 passed**.
+- `python3 -m unittest discover -s tests -p test_service_regressions.py -v`:
+  **34 passed**.
+- `python3 -m unittest discover -s tests -v`: **1,403 ran, OK, 15 skipped**
+  (Linux x86_64, Python 3.12; 150.825 seconds).
+- Full fake-Frame e2e in task-specific Docker project `frame-control-pr66`:
+  **47 passed** after the lock fix (192.270 seconds). Both task containers,
+  their network and all three volumes were removed afterward. The run emitted
+  an unclosed test-socket ResourceWarning; the suite returned OK.
+- Node page suites: `ctrl_gestures.mjs`, `pad_queue.mjs`, `desk_panels.mjs`
+  passed, using the production page functions.
+- `node --test app/test/*.test.js`: **3 test files passed**.
+- `node --test site/test/*.test.mjs`: **2 test files passed**.
+- Python compilation, desktop/site JavaScript syntax and `git diff --check`
+  passed. Fake-Frame e2e and exact-head CI results are recorded in the PR after completion.
+
+The first full regression run executed 1,401 tests and found two tests still
+expecting `/api/input` to start the old KDE Connect helper. They were updated
+to the intended gamescope route. The final full run includes the lazy-KDE
+regression above and the lock-order regression below. The initial restricted-sandbox input run could not open
+loopback sockets; the focused rerun with managed network access passed.
+
+Review of the merged code then found a lock inversion: the Unicode worker
+held the pad-send lock while `InputAgent.start()` acquired the route lock,
+whereas a headset switch held the route lock before resetting the pad. A
+bounded subprocess regression reproduced the deadlock with the real
+`InputAgent.send/start` methods and an offline launch stub. Unicode delivery
+and resets now take the route lock before the pad lock; the route lock is
+reentrant so an agent can capture its route inside that reservation. The new
+regression passed after the fix, as did all **58 focused input tests** and
+the **1,403-test full suite** (15 skipped). The local fake-Frame suite also
+passed **47 tests after this fix**. Exact-head CI results are recorded in the PR.
+
+Remaining device work requires an idle, available Frame and the shared lock:
+verify Unicode in a real text field, ordering under real typing load and
+precise absolute pointing for small panels. Worn-headset presentation,
+packaged Electron pointer lock and Windows GUI execution remain separate
+checks. Automated Windows/macOS CI does not establish those interactions.
+
+
+## Review round 2: queue saturation and failure recovery
+
+The coordinator identified silent eviction in both `_queue_for_kde()` and the
+worker's retry path. Both sliced the oldest accepted events off at 200 entries,
+even though the page had already received `sent: true` and discarded its copy.
+
+Admission now serializes concurrent requests, checks capacity before any pointer
+or keyboard send, and returns `sent: false, state: waiting` if the whole batch
+will not fit. Text stays in the bounded queue while in flight, and is removed
+only after the helper accepts it. Rejected mixed batches send no clicks; retrying
+them cannot duplicate clicks already sent by this server. The page polls until
+it can retry. Pointer input stays usable while accepted text is pending.
+
+The old worker also left its worker marker and one-hour timing sentinel behind
+if `send()` raised. Failure/timeout handling now retains unsent text, clears the
+worker/timing state and exposes an error through input status. The page polls
+accepted text without requiring another keypress. **Try again** explicitly
+restarts the retained queue. Thread-start failures follow the same retention
+rule. A stale worker's exception cannot change the new route's state.
+
+The proposed send-to-enqueue headset-switch race was not reachable through the
+normal HTTP switch path: `Handler.do_POST()` reserves `working(meant)` across
+both operations; `devices_post()` shares `_work_lock` and refuses a moving
+operation while work is active. Connector configuration/reconnect paths defer
+changed headsets while work runs. A production-handler regression attempts a
+switch at both send and enqueue and passes even on the previous implementation.
+No separate connector change was needed. Atomic queue admission now uses the
+existing route-before-pad lock order as well, because starting either input
+agent can capture its route synchronously.
+
+Three deterministic saturation regressions fail on `215db7b` and pass with the
+fix: capacity during the typing delay, capacity during an unsuccessful-send
+retry, and two concurrent producers competing for the last slots. They check
+exact accepted event sequences, reject a mixed click/text batch before its
+pointer effects, and verify delivery exactly once after retry. Additional tests
+cover explicit recovery after timeout, helper errors, send exceptions and
+thread-start failure, stale-worker isolation, HTTP status/start behavior, and
+the production page's polling/error behavior.
+
+No new active headset test was performed: this change affects host queue
+admission/recovery and status handling, not Frame event translation or injection.
+The earlier real click/ASCII evidence remains scoped to its recorded commit.
+Real Unicode, packaged Electron pointer lock, Windows GUI, wearer-visible
+presentation and precise small-window absolute positioning remain unverified.
+`sent: true` means accepted by this server, not application delivery. The queue
+is in memory; switching headsets cancels it and server exit loses it. A transport
+failure after a partial write can still make delivery ambiguous; this is not an
+end-to-end exactly-once protocol.
+
+Round 2 local verification: **141 focused input/touch/service tests passed**
+(62 input, 43 touch, 36 service; 20.061 seconds), and **1,409 server tests ran, OK, 15 skipped**
+(Linux x86_64, Python 3.12, 153.198 seconds). All three production-page Node
+suites passed; app tests passed (3 files), site tests passed (2 files), Python
+compilation and `git diff --check` passed. The existing skipped tests require
+zsh, macOS/Swift or built native PC libraries. Current-head CI is tracked in
+PR #66; the previous 47-test local fake-Frame run above is historical.
@@ -0,0 +1,129 @@
+"""Active real-Frame evidence probe: aborts unless idle, then owns the shared lock.
+
+Run manually on an authorised host: python3 docs/evidence/pad-gamescope-2026-10-24/probe.py
+The recorded run used /tmp paths; archived paths are relative to this checkout.
+"""
+import base64
+import datetime
+import json
+import os
+from pathlib import Path
+import queue
+import shlex
+import subprocess
+import sys
+import tempfile
+import threading
+import time
+import uuid
+
+ROOT = Path(__file__).resolve().parents[3]
+data = tempfile.TemporaryDirectory(prefix='pr66-live-')
+os.environ.update(FRAME_CONTROL_DATA_DIR=data.name, FRAME_CONTROL_TELEMETRY='0', FRAME_PRIVATE_SSH='1')
+sys.path.insert(0, str(ROOT / 'ui'))
+import server
+
+SSH = ['ssh', '-o', 'BatchMode=yes', '-o', 'ConnectTimeout=5', 'frame']
+owner = 'pr66-' + uuid.uuid4().hex
+result = {'time': datetime.datetime.now(datetime.timezone.utc).isoformat(), 'owner': owner,
+          'base': '0da1fe5', 'route': 'production server.remote_input -> TouchAgent SSH -> frame_touch.py -> gamescope EIS',
+          'events': []}
+recorder = None
+locked = False
+rows = queue.Queue()
+
+def ssh(cmd):
+    return subprocess.run([*SSH, cmd], capture_output=True, text=True, timeout=15, check=True).stdout
+
+def focus():
+    return json.loads(server.ssh('python3 - focus', stdin=(ROOT/'ui/frame_touch.py').read_text(), timeout=15))
+
+def activity():
+    return json.loads(ssh('/opt/steamvr/bin/linuxarm64/vrcmd --stats'))[0]['activity_level']
+
+def read_rows():
+    for raw in recorder.stdout:
+        rows.put(json.loads(raw))
+
+def collect(seconds):
+    end = time.monotonic() + seconds
+    while time.monotonic() < end:
+        try:
+            result['events'].append(rows.get(timeout=max(0.01, end-time.monotonic())))
+        except queue.Empty:
+            break
+
+try:
+    result['initial_focus'] = focus()
+    result['initial_activity'] = activity()
+    assert not result['initial_focus'].get('window'), 'existing focused panel; do not interrupt'
+    assert result['initial_activity'] == 3, 'headset is not in standby'
+    ssh('mkdir /tmp/frame-test.lock && printf %s ' + shlex.quote(owner) + ' > /tmp/frame-test.lock/owner')
+    locked = True
+    result['lock_acquired'] = True
+    assert not focus().get('window') and activity() == 3, 'session changed during lock acquisition'
+    code = base64.b64encode(Path(__file__).with_name('recorder.py').read_bytes()).decode()
+    command = 'python3 -u -c ' + shlex.quote("import base64;exec(compile(base64.b64decode('" + code + "'),'pr66-recorder','exec'))")
+    recorder = subprocess.Popen([*SSH, command], stdin=subprocess.PIPE, stdout=subprocess.PIPE,
+                                stderr=subprocess.PIPE, text=True)
+    threading.Thread(target=read_rows, daemon=True).start()
+    initial = rows.get(timeout=8)
+    result['recorder'] = initial
+    assert initial.get('ready'), initial
+    window = initial['window']
+    for _ in range(20):
+        current = focus()
+        if current.get('window') == window:
+            break
+        time.sleep(0.2)
+    result['focused_panel'] = current
+    assert current.get('window') == window and current.get('display') == ':0', 'owned panel did not obtain focus; no input sent'
+    server.remote_input({})
+    for _ in range(50):
+        if server._touch.status.get('state') in ('ready', 'error'):
+            break
+        time.sleep(0.1)
+    result['agent_status'] = dict(server._touch.status)
+    assert result['agent_status'].get('state') == 'ready', result['agent_status']
+    target = {'window': window, 'display': ':0'}
+    assert focus().get('window') == window and activity() == 3
+    result['position_response'] = server.remote_touch({'events': [dict(target, fx=0.5, fy=0.5)]})
+    time.sleep(0.2)
+    assert focus().get('window') == window and activity() == 3
+    result['pad_response'] = server.remote_input({'events': [{'singleclick': True}, {'key': 'Frame66'}, {'specialKey': 12}]})
+    collect(2)
+    received = result['events']
+    result['received_text'] = ''.join(e.get('text', '') for e in received if e.get('type') == 2)
+    result['click_received'] = any(e.get('type') == 4 and e.get('code') == 1 and e.get('window') == window for e in received)
+    result['release_received'] = any(e.get('type') == 5 and e.get('code') == 1 and e.get('window') == window for e in received)
+    result['passed'] = result['click_received'] and result['release_received'] and result['received_text'] == 'Frame66\r'
+    assert result['passed'], 'event recorder did not observe expected click and keys'
+except Exception as e:
+    result['error'] = str(e)
+finally:
+    proc = server._touch.proc
+    server.reset_pad_route()
+    if proc:
+        try:
+            proc.wait(timeout=8)
+            result['input_helper_exited'] = True
+        except subprocess.TimeoutExpired:
+            proc.terminate()
+            proc.wait(timeout=5)
+            result['input_helper_terminated'] = True
+    if recorder:
+        recorder.stdin.close()
+        recorder.wait(timeout=8)
+        collect(0.2)
+        result['recorder_exit'] = recorder.returncode
+        result['recorder_stderr'] = recorder.stderr.read()[-1000:]
+    if locked:
+        result['final_focus'] = focus()
+        result['final_activity'] = activity()
+        ssh('test "$(cat /tmp/frame-test.lock/owner)" = ' + shlex.quote(owner)
+            + ' && rm /tmp/frame-test.lock/owner && rmdir /tmp/frame-test.lock')
+        result['owned_lock_released'] = True
+    Path('/tmp/pr66-live.json').write_text(json.dumps(result, indent=2) + '\n')
+    print(json.dumps(result, indent=2))
+    data.cleanup()
+sys.exit(0 if result.get('passed') else 1)
@@ -0,0 +1,280 @@
+{
+  "time": "2026-10-24T11:20:37.392607+00:00",
+  "owner": "pr66-7c2a8e8851f449bcb0bb6a040be94d99",
+  "base": "0da1fe5",
+  "route": "production server.remote_input -> TouchAgent SSH -> frame_touch.py -> gamescope EIS",
+  "events": [
+    {
+      "type": 6,
+      "window": 62914561,
+      "x": 320,
+      "y": 200,
+      "synthetic": false
+    },
+    {
+      "type": 6,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "synthetic": false
+    },
+    {
+      "type": 4,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 1,
+      "synthetic": false
+    },
+    {
+      "type": 5,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 1,
+      "synthetic": false
+    },
+    {
+      "type": 2,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 50,
+      "synthetic": false,
+      "text": "",
+      "keysym": 65505
+    },
+    {
+      "type": 2,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 41,
+      "synthetic": false,
+      "text": "F",
+      "keysym": 70
+    },
+    {
+      "type": 3,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 41,
+      "synthetic": false
+    },
+    {
+      "type": 3,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 50,
+      "synthetic": false
+    },
+    {
+      "type": 2,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 27,
+      "synthetic": false,
+      "text": "r",
+      "keysym": 114
+    },
+    {
+      "type": 3,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 27,
+      "synthetic": false
+    },
+    {
+      "type": 2,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 38,
+      "synthetic": false,
+      "text": "a",
+      "keysym": 97
+    },
+    {
+      "type": 3,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 38,
+      "synthetic": false
+    },
+    {
+      "type": 2,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 58,
+      "synthetic": false,
+      "text": "m",
+      "keysym": 109
+    },
+    {
+      "type": 3,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 58,
+      "synthetic": false
+    },
+    {
+      "type": 2,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 26,
+      "synthetic": false,
+      "text": "e",
+      "keysym": 101
+    },
+    {
+      "type": 3,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 26,
+      "synthetic": false
+    },
+    {
+      "type": 2,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 15,
+      "synthetic": false,
+      "text": "6",
+      "keysym": 54
+    },
+    {
+      "type": 3,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 15,
+      "synthetic": false
+    },
+    {
+      "type": 2,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 15,
+      "synthetic": false,
+      "text": "6",
+      "keysym": 54
+    },
+    {
+      "type": 3,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 15,
+      "synthetic": false
+    },
+    {
+      "type": 2,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 36,
+      "synthetic": false,
+      "text": "\r",
+      "keysym": 65293
+    },
+    {
+      "type": 3,
+      "window": 62914561,
+      "x": 639,
+      "y": 399,
+      "code": 36,
+      "synthetic": false
+    },
+    {
+      "cleaned": true
+    }
+  ],
+  "initial_focus": {
+    "window": null
+  },
+  "initial_activity": 3,
+  "lock_acquired": true,
+  "recorder": {
+    "ready": true,
+    "window": 62914561,
+    "pid": 317210,
+    "display": ":0",
+    "appid": 2000999066
+  },
+  "focused_panel": {
+    "window": 62914561,
+    "display": ":0",
+    "name": "Frame Control PR66 disposable input probe",
+    "x": 0,
+    "y": 0,
+    "width": 640,
+    "height": 400,
+    "root": [
+      1920,
+      1080
+    ],
+    "app": 2000999066
+  },
+  "agent_status": {
+    "state": "ready",
+    "focus": 62914561
+  },
+  "position_response": {
+    "state": "ready",
+    "focus": 62914561,
+    "sent": true
+  },
+  "pad_response": {
+    "state": "ready",
+    "focus": 62914561,
+    "sent": true
+  },
+  "received_text": "Frame66\r",
+  "click_received": true,
+  "release_received": true,
+  "passed": true,
+  "input_helper_exited": true,
+  "recorder_exit": 0,
+  "recorder_stderr": "",
+  "final_focus": {
+    "window": null
+  },
+  "final_activity": 3,
+  "owned_lock_released": true,
+  "implementation_commit": "5641d3c316dfe225321f7233840cc83ff8c3f527",
+  "device": {
+    "steamos": "0.5.6",
+    "build_id": "20261015.6141207",
+    "steamvr": "2.18.2",
+    "battery_before": 98,
+    "battery_after": 98
+  },
+  "delayed_cleanup": {
+    "time": "2026-10-24T11:21:46Z",
+    "shared_lock_absent": true,
+    "owned_recorder_absent": true,
+    "game_mode": "active",
+    "full_plasma": "inactive",
+    "steam_pid": 297826,
+    "vrserver_pid": 297506,
+    "focusable_windows": [],
+    "focused_windows": [],
+    "activity_level": 3
+  },
+  "limits": [
+    "No wearer or packaged Electron/GUI interaction",
+    "ASCII only; Unicode and cross-service ordering not tested on device",
+    "The positioning setup reached the recorder edge (639,399); pixel-accurate absolute targeting is not established by this probe"
+  ]
+}
@@ -0,0 +1,82 @@
+"""Disposable X11 event recorder used for the PR #66 focused-input observation."""
+import ctypes as C
+import json
+import os
+import select
+import signal
+import time
+
+X = C.CDLL('libX11.so.6')
+P, U = C.c_void_p, C.c_ulong
+for name, restype, args in [
+    ('XOpenDisplay', P, [C.c_char_p]), ('XDefaultRootWindow', U, [P]),
+    ('XCreateSimpleWindow', U, [P, U, C.c_int, C.c_int, C.c_uint, C.c_uint, C.c_uint, U, U]),
+    ('XInternAtom', U, [P, C.c_char_p, C.c_int]),
+    ('XChangeProperty', C.c_int, [P, U, U, U, C.c_int, C.c_int, P, C.c_int]),
+    ('XStoreName', C.c_int, [P, U, C.c_char_p]),
+    ('XSelectInput', C.c_int, [P, U, C.c_long]), ('XMapWindow', C.c_int, [P, U]),
+    ('XFlush', C.c_int, [P]), ('XPending', C.c_int, [P]),
+    ('XConnectionNumber', C.c_int, [P]), ('XNextEvent', C.c_int, [P, P]),
+    ('XLookupString', C.c_int, [P, P, C.c_int, P, P]),
+    ('XDestroyWindow', C.c_int, [P, U]), ('XCloseDisplay', C.c_int, [P]),
+]:
+    fn = getattr(X, name)
+    fn.restype, fn.argtypes = restype, args
+
+class InputEvent(C.Structure):
+    _fields_ = [('type', C.c_int), ('serial', U), ('send_event', C.c_int),
+               ('display', P), ('window', U), ('root', U), ('subwindow', U),
+               ('time', U), ('x', C.c_int), ('y', C.c_int), ('x_root', C.c_int),
+               ('y_root', C.c_int), ('state', C.c_uint), ('code', C.c_uint),
+               ('same_screen', C.c_int)]
+
+class Event(C.Union):
+    _fields_ = [('type', C.c_int), ('input', InputEvent), ('pad', C.c_long * 24)]
+
+def say(**row):
+    print(json.dumps(row), flush=True)
+
+def end(*_):
+    raise TimeoutError('probe ended')
+
+signal.signal(signal.SIGTERM, end)
+signal.signal(signal.SIGALRM, end)
+signal.alarm(45)
+d = X.XOpenDisplay(b':0')
+assert d, 'X display unavailable'
+w = 0
+try:
+    w = X.XCreateSimpleWindow(d, X.XDefaultRootWindow(d), 0, 0, 640, 400, 0, 0, 0x224466)
+    for name, value in [(b'STEAM_GAME', 2000999066), (b'_NET_WM_PID', os.getpid())]:
+        val = U(value)
+        X.XChangeProperty(d, w, X.XInternAtom(d, name, 0), 6, 32, 0, C.byref(val), 1)
+    X.XStoreName(d, w, b'Frame Control PR66 disposable input probe')
+    X.XSelectInput(d, w, 1 | 2 | 4 | 8 | 64 | 32768 | 2097152)
+    X.XMapWindow(d, w)
+    X.XFlush(d)
+    say(ready=True, window=w, pid=os.getpid(), display=':0', appid=2000999066)
+    while True:
+        while X.XPending(d):
+            event = Event()
+            X.XNextEvent(d, C.byref(event))
+            if event.type in (2, 3, 4, 5, 6):
+                e = event.input
+                row = dict(type=event.type, window=e.window, x=e.x, y=e.y,
+                           synthetic=bool(e.send_event))
+                if event.type != 6:
+                    row['code'] = e.code
+                if event.type == 2:
+                    buf, keysym = C.create_string_buffer(32), U()
+                    n = X.XLookupString(C.byref(e), buf, 32, C.byref(keysym), None)
+                    row.update(text=buf.raw[:n].decode('utf-8', 'replace'), keysym=keysym.value)
+                say(**row)
+        ready, _, _ = select.select([0, X.XConnectionNumber(d)], [], [], 0.2)
+        if 0 in ready and not os.read(0, 1024):
+            break
+except TimeoutError:
+    pass
+finally:
+    if w:
+        X.XDestroyWindow(d, w)
+    X.XCloseDisplay(d)
+    say(cleaned=True)
@@ -0,0 +1,168 @@
+# Saved panel layouts: feasibility recheck, 2026-10-24
+
+Issue: [#31](https://github.com/saphid/frame-control/issues/31). Base: main
+`5b47cd3`. This investigation covers spatial save/restore for existing
+Steam/gamescope panels without another overlay application.
+
+## Decision and user behavior
+
+Keep saved layouts unavailable. The current Frame exposes some geometry,
+including a standing-space coordinate transform, but this investigation did
+not establish a complete dock-state snapshot or a controlled restore path for
+these panels. The existing companion/headset switcher still lists panels and
+requests focus; users arrange them with SteamVR's dashboard controls. No layout
+file, new control, overlay application or background service is added.
+
+This is a limit of the investigated route, not proof that every native
+implementation is impossible. In particular, failure of
+`GetOverlayTransformAbsolute` does **not** mean all pose readback is denied.
+
+## Device and read-only results
+
+SteamOS **0.5.6**, BUILD_ID **20261015.6141207**; `vrcmd --version`:
+**2.18.2**. `gamescope-session.service` was active and
+`plasma-workspace.target` inactive: Game Mode. gamescope used the OpenVR backend
+and `--virtual-connector-strategy PerAppId`. Battery was 99%, Not charging,
+before the probe and 100%, Full, afterwards.
+
+The [exact probe](panel-layout-2026-10-24/probe.py) loads only
+`FnTable:IVROverlay_028` from the shipped
+`/opt/steamvr/bin/linuxarm64/libopenvr_api.so`. It uses application type 3
+(`VRApplication_Background`, avoiding runtime startup) and shuts down its own
+OpenVR connection in `finally`. Initialization and interface acquisition both
+returned 0. It calls getters only, with no image capture or input injection.
+
+[Raw result](panel-layout-2026-10-24/readback.json):
+
+| Getter | `valve.steam.desktopgame.0` | `valve.steam.gamepadui.main` |
+|---|---|---|
+| FindOverlay | 0, Success | 0, Success |
+| Rendering process | `gamescope-wl` | `gamescope-wl` |
+| Width | 0; 2.6700000763 m | 0; 2.6670000553 m |
+| Curvature | 0; 0 | 0; 0 |
+| Transform type | 0; 5, DashboardTab | 0; 5, DashboardTab |
+| Absolute transform | 18, WrongTransformType | 18, WrongTransformType |
+| Tracked-device-relative transform | 18, WrongTransformType | 18, WrongTransformType |
+| Coordinate transform, center, standing origin | 0; identity matrix | 0; scaled/translated matrix |
+| Coordinate transform, center, seated origin | 23, RequestFailed | 23, RequestFailed |
+| Coordinate transform, center, raw origin | 23, RequestFailed | 23, RequestFailed |
+
+The dashboard and both panels were reported hidden; neither panel was active.
+The standing-space matrix is the transform for the overlay point `(0.5, 0.5)`;
+it does not identify World/Dashboard/Theater/controller docking or recreate a
+dashboard frame. The Steam panel's basis has scale about 0.369. Treat neither
+that scaled matrix nor the hidden Gamescope panel's identity as a validated
+physical placement to feed to an absolute-transform setter. Visible placement,
+matrix freshness and correspondence to dashboard size were not checked by a
+wearer.
+
+## API and dashboard boundary
+
+The ABI and result names were checked against Valve's
+[OpenVR v2.12.14 C header](https://github.com/ValveSoftware/openvr/blob/v2.12.14/headers/openvr_capi.h)
+and [C++ interface comments](https://github.com/ValveSoftware/openvr/blob/v2.12.14/headers/openvr.h).
+That header declares `IVROverlay_028`; its SHA-256 is
+`d58a2b149e434b2b8ee25cb4253ca6157feed8092f621e78ecfed6846d35d3e4`.
+The absolute getter is explicitly limited to absolute transforms. The
+coordinate getter returns the 3D transform for a 2D overlay point. Public
+`IVROverlay_028` has no getter/setter pair for dashboard dock location or frame
+scene-graph state. The probe fails closed if that exact interface is absent.
+
+Read-only inspection of the shipped dashboard
+`/opt/steamvr/resources/webinterface/dashboard/chunk~8012d0c89.js`
+(SHA-256 `937ec62c67a481f030a416099ceb98b0fe2a702aea5a0705601c6c40d4969f6d`)
+found:
+
+- `dockLocation` reads an internal `_getDockLocation(frameID)` store;
+  `SetDockLocation` updates that store or sends a VRLink request.
+- `m_mapLastRelativeTransformForDockLocation` keeps per-location transforms.
+  World initialization requests a dashboard scene-graph transform through
+  `requestSGTransform`, then updates the component's `xfTransform`.
+- The `vrcmd_dock_overlay` handler accepts `dashboard`, `lefthand`, `righthand`,
+  `theater` and `world`. It selects the frame associated with the supplied key,
+  **falls back to the active frame if no match exists**, sets dock location and
+  may switch frame. This is not an exact pose restore or a readback response.
+
+Those are observations of shipped internal code, not proof that Frame Control
+can safely access its state through a supported external API. No dashboard
+JavaScript was injected, private store edited or VRLink command sent. The
+active-frame fallback is an additional reason not to experiment with docking
+commands on a stale key.
+
+The [2026-09-28 probe](../panels.md#saved-spatial-layouts-blocked-on-the-current-panel-route)
+on SteamOS 0.4.1 / SteamVR 2.18.1 recorded an absolute setter returning
+12 (`PermissionDenied`) on a temporary, test-owned gamescope panel. That setter
+result is **historical**, not a new test on 2.18.2. The newer read-only results
+above neither overturn nor reverify it.
+
+## Commands, safety and cleanup
+
+Read-only inspection used `ssh frame` with `BatchMode=yes` and
+`ConnectTimeout=10`, `cat /etc/os-release`, battery sysfs reads,
+`systemctl --user is-active gamescope-session.service plasma-workspace.target`,
+process inspection, and `vrcmd --version`, `--help`, `--overlays`. An initial
+`--listoverlays` attempt returned `Unknown command`; `--overlays` is the actual
+enumeration command. Dashboard sources were read/copied to local `/tmp` for
+inspection. No files were installed on the Frame.
+
+Run the retained probe from the repository root, following
+[the shared-device rules](../testing.md#headset-smoke-test). Stop if the lock is
+occupied or battery is at/below 20%; never remove another task's lock:
+
+```sh
+ssh -o BatchMode=yes -o ConnectTimeout=10 frame \
+  'set -eu; mkdir /tmp/frame-test.lock || exit 3; trap "rmdir /tmp/frame-test.lock" EXIT; timeout 25 python3 -' \
+  < docs/evidence/panel-layout-2026-10-24/probe.py
+```
+
+This command acquired the lock, completed with exit 0 and released it. A
+subsequent check confirmed the lock absent, Game Mode active, and Steam,
+`vrserver` and `vrcompositor` running. The probe's only headset filesystem
+change was creating/removing that shared lock. Its Python source ran from
+stdin. There were no test panels, installs, profiles or services to remove.
+
+No mutation test was essential: without verified dock/relative-frame readback,
+there was no exact original layout and bounded rollback to justify changing a
+panel. Existing user panels were never focused, shown, hidden, resized, moved
+or docked. No global settings, security, passwords, runtime processes or modes
+were changed; no reboot or reset occurred.
+
+## Validation and remaining gaps
+
+Existing `tests/test_panels.py` covers main-panel enumeration, hidden panels,
+focus revalidation, unavailable-runtime errors and HTTP authentication/Host/
+Origin guards. `tests/test_page_scripts.py` checks the production page's shared
+JavaScript scope. These offline checks do not prove spatial placement. The
+retained probe is evidence tooling, not a new product API; no fake layout
+backend or layout tests claiming device support were added.
+
+Validation passed:
+
+```sh
+python3 -m unittest discover -s tests -p test_panels.py -v       # 8 passed
+python3 -m unittest discover -s tests -p test_page_scripts.py -v # 5 passed
+python3 -m py_compile docs/evidence/panel-layout-2026-10-24/probe.py
+python3 -m json.tool docs/evidence/panel-layout-2026-10-24/readback.json
+git diff --check
+```
+
+The initial panel-suite attempt hit sandbox socket restrictions in four HTTP
+tests; rerunning with loopback sockets permitted passed all eight. No real
+headset connection is used by those tests. The full suite and fake Frame were
+not run for this evidence-only change.
+
+Before shipping layouts, a native route still needs proof of dock state and
+relative-frame readback, authorized restore on an owned test panel with exact
+rollback, and correspondence between saved size/pose and visible placement.
+It also needs stable source identity and a fresh capture-permission/ticket
+flow when reopening viewers. Reconnect, panel recreation, reboot recovery,
+Steam-launched gameplay, controller docking and worn-headset input were not
+tested here. The existing Frame Control-owned media renderer is not proof that
+Steam/gamescope panels can be restored. No third-party overlay app was used.
+
+## Announcement
+
+We rechecked saved panel layouts on SteamOS 0.5.6. SteamVR exposes some panel
+geometry, but we have not verified safe restoration of Steam's panel positions
+and docking. Saved layouts remain unavailable; use SteamVR's dashboard to
+arrange panels. The panel switcher continues to list and show open panels.
@@ -0,0 +1,89 @@
+"""Read-only issue #31 probe; pipe to `ssh frame 'timeout 25 python3 -'`.
+
+Acquire /tmp/frame-test.lock first per docs/testing.md. No overlay is created,
+focused, docked or changed. Background init (3) does not start SteamVR.
+ABI: ValveSoftware/openvr v2.12.14 headers/openvr_capi.h, IVROverlay_028.
+"""
+import ctypes as C
+import json
+from pathlib import Path
+
+H = C.c_uint64
+Matrix = C.c_float * 12
+
+
+class Vector2(C.Structure):
+    _fields_ = [('v', C.c_float * 2)]
+
+
+def probe():
+    vr = C.CDLL('/opt/steamvr/bin/linuxarm64/libopenvr_api.so')
+    vr.VR_InitInternal2.argtypes = [C.POINTER(C.c_int), C.c_int, C.c_char_p]
+    vr.VR_InitInternal2.restype = C.c_uint32
+    vr.VR_GetGenericInterface.argtypes = [C.c_char_p, C.POINTER(C.c_int)]
+    vr.VR_GetGenericInterface.restype = C.c_void_p
+    vr.VR_ShutdownInternal.argtypes = []
+    vr.VR_ShutdownInternal.restype = None
+    err = C.c_int()
+    vr.VR_InitInternal2(C.byref(err), 3, None)
+    report = {'init_error': err.value, 'interface': 'IVROverlay_028', 'panels': []}
+    if err.value:
+        return report
+    try:
+        ptr = vr.VR_GetGenericInterface(b'FnTable:IVROverlay_028', C.byref(err))
+        report['interface_error'] = err.value
+        if err.value or not ptr:
+            return report
+        table = C.cast(ptr, C.POINTER(C.c_void_p))
+
+        def function(slot, result, *args):
+            return C.CFUNCTYPE(result, *args)(table[slot])
+
+        def read(slot, handle, output_type):
+            output = output_type()
+            rc = function(slot, C.c_int, H, C.POINTER(output_type))(handle, C.byref(output))
+            return {'rc': rc, 'value': output.value if rc == 0 else None}
+
+        report['dashboard_visible'] = function(68, C.c_bool)()
+        for key in ('valve.steam.desktopgame.0', 'valve.steam.gamepadui.main'):
+            row = {'key': key}
+            report['panels'].append(row)
+            handle = H()
+            row['find_rc'] = function(0, C.c_int, C.c_char_p, C.POINTER(H))(
+                key.encode(), C.byref(handle))
+            if row['find_rc']:
+                continue
+            pid = function(10, C.c_uint32, H)(handle)
+            row['rendering_pid'] = pid
+            try:
+                row['rendering_comm'] = Path('/proc/%d/comm' % pid).read_text().strip()
+            except OSError:
+                row['rendering_comm'] = None
+            row['visible'] = function(45, C.c_bool, H)(handle)
+            row['active_dashboard'] = function(69, C.c_bool, H)(handle)
+            row['width_m'] = read(23, handle, C.c_float)
+            row['curvature'] = read(25, handle, C.c_float)
+            row['transform_type'] = read(32, handle, C.c_int)
+            for name, slot, scalar in (('absolute', 34, C.c_int),
+                                       ('tracked_device_relative', 36, C.c_uint32)):
+                origin = scalar()
+                matrix = Matrix()
+                rc = function(slot, C.c_int, H, C.POINTER(scalar), C.POINTER(Matrix))(
+                    handle, C.byref(origin), C.byref(matrix))
+                row[name] = {'rc': rc, 'origin_or_device': origin.value if rc == 0 else None,
+                             'matrix': list(matrix) if rc == 0 else None}
+            row['coordinate_transforms'] = []
+            for origin in (0, 1, 2):  # seated, standing, raw/uncalibrated
+                matrix = Matrix()
+                rc = function(46, C.c_int, H, C.c_int, Vector2, C.POINTER(Matrix))(
+                    handle, origin, Vector2((.5, .5)), C.byref(matrix))
+                row['coordinate_transforms'].append({
+                    'tracking_origin': origin, 'coordinates': [.5, .5], 'rc': rc,
+                    'matrix': list(matrix) if rc == 0 else None})
+        return report
+    finally:
+        vr.VR_ShutdownInternal()
+
+
+if __name__ == '__main__':
+    print(json.dumps(probe(), indent=2, allow_nan=False))
@@ -0,0 +1,152 @@
+{
+  "init_error": 0,
+  "interface": "IVROverlay_028",
+  "panels": [
+    {
+      "key": "valve.steam.desktopgame.0",
+      "find_rc": 0,
+      "rendering_pid": 297037,
+      "rendering_comm": "gamescope-wl",
+      "visible": false,
+      "active_dashboard": false,
+      "width_m": {
+        "rc": 0,
+        "value": 2.6700000762939453
+      },
+      "curvature": {
+        "rc": 0,
+        "value": 0.0
+      },
+      "transform_type": {
+        "rc": 0,
+        "value": 5
+      },
+      "absolute": {
+        "rc": 18,
+        "origin_or_device": null,
+        "matrix": null
+      },
+      "tracked_device_relative": {
+        "rc": 18,
+        "origin_or_device": null,
+        "matrix": null
+      },
+      "coordinate_transforms": [
+        {
+          "tracking_origin": 0,
+          "coordinates": [
+            0.5,
+            0.5
+          ],
+          "rc": 23,
+          "matrix": null
+        },
+        {
+          "tracking_origin": 1,
+          "coordinates": [
+            0.5,
+            0.5
+          ],
+          "rc": 0,
+          "matrix": [
+            1.0,
+            0.0,
+            0.0,
+            0.0,
+            0.0,
+            1.0,
+            0.0,
+            0.0,
+            0.0,
+            0.0,
+            1.0,
+            0.0
+          ]
+        },
+        {
+          "tracking_origin": 2,
+          "coordinates": [
+            0.5,
+            0.5
+          ],
+          "rc": 23,
+          "matrix": null
+        }
+      ]
+    },
+    {
+      "key": "valve.steam.gamepadui.main",
+      "find_rc": 0,
+      "rendering_pid": 297037,
+      "rendering_comm": "gamescope-wl",
+      "visible": false,
+      "active_dashboard": false,
+      "width_m": {
+        "rc": 0,
+        "value": 2.6670000553131104
+      },
+      "curvature": {
+        "rc": 0,
+        "value": 0.0
+      },
+      "transform_type": {
+        "rc": 0,
+        "value": 5
+      },
+      "absolute": {
+        "rc": 18,
+        "origin_or_device": null,
+        "matrix": null
+      },
+      "tracked_device_relative": {
+        "rc": 18,
+        "origin_or_device": null,
+        "matrix": null
+      },
+      "coordinate_transforms": [
+        {
+          "tracking_origin": 0,
+          "coordinates": [
+            0.5,
+            0.5
+          ],
+          "rc": 23,
+          "matrix": null
+        },
+        {
+          "tracking_origin": 1,
+          "coordinates": [
+            0.5,
+            0.5
+          ],
+          "rc": 0,
+          "matrix": [
+            0.3689989447593689,
+            0.0,
+            0.0008894503116607666,
+            -0.5048462152481079,
+            0.0,
+            0.36899998784065247,
+            0.0,
+            1.2510287761688232,
+            -0.0008894503116607666,
+            0.0,
+            0.3689989447593689,
+            -1.1726053953170776
+          ]
+        },
+        {
+          "tracking_origin": 2,
+          "coordinates": [
+            0.5,
+            0.5
+          ],
+          "rc": 23,
+          "matrix": null
+        }
+      ]
+    }
+  ],
+  "interface_error": 0,
+  "dashboard_visible": false
+}
@@ -0,0 +1,226 @@
+{
+  "date_utc": "2026-10-24T09:27:25.354Z",
+  "result": "PASS: literal packaged desktop Return to Game Mode button switched the real Frame from full Plasma; a repeat visible click made no second switch",
+  "documentation_base_commit": "71288d8980040c14f777f0e417f41f7105eaaac4",
+  "host": "lxso2",
+  "steamos_version": "0.5.6",
+  "steamos_build": "20261015.6141207",
+  "prior_evidence": [
+    "https://github.com/saphid/frame-control/issues/142",
+    "https://github.com/saphid/frame-control/pull/143",
+    "https://github.com/saphid/frame-control/pull/156"
+  ],
+  "package": {
+    "tag": "v0.4.17",
+    "asset": "Frame-Control-linux-x86_64.AppImage",
+    "url": "https://github.com/saphid/frame-control/releases/download/v0.4.17/Frame-Control-linux-x86_64.AppImage",
+    "sha256": "b1b8eb82bf73a227b2cd36e8f0ef3a2ebdf7967c5853efb08a9634f38227015d",
+    "verification": "sha256sum matched the live release asset digest. Independently extracted the retained asset into a new private test directory. cmp confirmed bundled frame_backup.py and frame_backup_remote.py equal the checkout versions. No source or payload was changed.",
+    "launch": "--appimage-extract, then squashfs-root/AppRun --no-sandbox --disable-gpu --remote-debugging-port=49372 --remote-debugging-address=127.0.0.1 --user-data-dir=/tmp/frame-return-button-20261024/electron-profile --window-size=1400,950",
+    "display": "Xvfb :138 -screen 0 1440x1000x24 -nolisten tcp; Electron content screenshot 1400x923",
+    "isolation": "Private XDG_CONFIG_HOME, XDG_CACHE_HOME, FRAME_CONTROL_DATA_DIR and Electron profile; FRAME_CONTROL_TELEMETRY=0 and FRAME_CONTROL_NO_UPDATE_CHECK=1. Existing authorized ssh frame identity retained; bundled server and Python used."
+  },
+  "preflight": {
+    "read_only_first": true,
+    "initial_mode": "gamescope-session.target=active; plasma-workspace-wayland.target=inactive",
+    "initial_battery_percent": 99,
+    "initial_battery_status": "Not charging; charger Full",
+    "steam_and_steamvr_running": true,
+    "running_steam_apps": [],
+    "default_login_mode": "game",
+    "lock": "mkdir /tmp/frame-test.lock succeeded atomically; no foreign lock was removed. Initial harmless probe attempt released its own lock, then a fresh atomic acquisition covered both full-session attempts."
+  },
+  "rollback": {
+    "unit": "fc-return-button-rollback-20261024",
+    "probe": "systemd-run --user --unit=fc-return-button-probe-20261024 --on-active=2s --timer-property=AccuracySec=1s --property=Type=oneshot --property=RemainAfterExit=yes /usr/bin/true",
+    "probe_execution": "Result=success\nExecMainStartTimestamp=Sat 2026-10-24 20:27:07 AEDT\nExecMainExitTimestamp=Sat 2026-10-24 20:27:07 AEDT\nExecMainCode=1\nExecMainStatus=0\n",
+    "arm": "systemd-run --user --unit=fc-return-button-rollback-20261024 --on-active=180s --timer-property=AccuracySec=1s /bin/sh -c <guarded command>",
+    "guarded_command": "if [ \"$(systemctl --user is-active gamescope-session.target)\" != active ]; then exec /usr/bin/steamosctl switch-to-game-mode; fi",
+    "verified_before_desktop": "TimersMonotonic={ OnActiveUSec=3min ; next_elapse=1d 21h 53min 49.130174s }\nNextElapseUSecMonotonic=1d 21h 53min 49.130174s\nBindsTo=\nPartOf=\nActiveState=active\nExecStart={ path=/bin/sh ; argv[]=/bin/sh -c if [ \"$(systemctl --user is-active gamescope-session.target)\" != active ]; then exec /usr/bin/steamosctl switch-to-game-mode; fi ; ignore_errors=no ; start_time=[n/a] ; stop_time=[n/a] ; pid=0 ; code=(null) ; status=0/0 }\nBindsTo=\nPartOf=\n",
+    "user_manager_linger": "yes",
+    "desktop_entry": "ssh frame steamosctl switch-to-desktop-mode",
+    "first_click_utc": "2026-10-24T09:27:27.037Z",
+    "cancelled_before_deadline": true,
+    "fired": false,
+    "limit": "The benign transient timer was observed executing. The actual rollback timer configuration was read back, then cancelled after the UI restored Game Mode; its recovery action was not deliberately allowed to fire."
+  },
+  "ui": {
+    "navigation": "Clicked nav Tools, Backup & recovery, and Check in the actual packaged Electron window. Initial Game Mode disabled Return. After full Plasma entry, Check enabled it.",
+    "automation": "Playwright connected to the packaged Electron instance over local CDP and used locator.click() on the visible button; accepted the native JavaScript confirm dialog. No Return API/CLI or production handler was directly invoked by the test. T3 preview was inspected separately and closed; it cannot control this Electron window.",
+    "clicks": [
+      {
+        "isTrusted": true,
+        "utc": "2026-10-24T09:27:27.037Z"
+      },
+      {
+        "isTrusted": true,
+        "utc": "2026-10-24T09:27:32.942Z"
+      }
+    ],
+    "button_bounds_css_pixels": {
+      "x": 47,
+      "y": 709.59375,
+      "width": 156.203125,
+      "height": 44
+    },
+    "confirmation_dialogs": [
+      {
+        "type": "confirm",
+        "message": "Return the headset to Game Mode?\n\nThis closes the desktop and any apps open in it, then starts Steam's Game Mode.",
+        "utc": "2026-10-24T09:27:27.042Z"
+      },
+      {
+        "type": "confirm",
+        "message": "Return the headset to Game Mode?\n\nThis closes the desktop and any apps open in it, then starts Steam's Game Mode.",
+        "utc": "2026-10-24T09:27:32.943Z"
+      }
+    ],
+    "desktop_check": {
+      "http": 200,
+      "body": {
+        "mode": "desktop",
+        "gamescope": true,
+        "plasma": true,
+        "nested_desktop": false,
+        "configured": "gamescope-wayland.desktop",
+        "config_file": "/etc/sddm.conf.d/steamos.conf",
+        "selector": "steamosctl",
+        "session_targets": {
+          "gamescope-session.target": "inactive",
+          "plasma-workspace-wayland.target": "active"
+        },
+        "mode_source": "session-targets"
+      }
+    },
+    "return_responses": [
+      {
+        "http": 200,
+        "body": {
+          "message": "Asked the headset to return to Game Mode; it restarts its session in a few seconds",
+          "state": {
+            "mode": "desktop",
+            "gamescope": true,
+            "plasma": true,
+            "nested_desktop": false,
+            "configured": "gamescope-wayland.desktop",
+            "config_file": "/etc/sddm.conf.d/steamos.conf",
+            "selector": "steamosctl",
+            "session_targets": {
+              "gamescope-session.target": "inactive",
+              "plasma-workspace-wayland.target": "active"
+            },
+            "mode_source": "session-targets"
+          },
+          "switched": true,
+          "command": "steamosctl switch-to-game-mode"
+        },
+        "utc": "2026-10-24T09:27:27.316Z"
+      },
+      {
+        "http": 200,
+        "body": {
+          "message": "The headset is already in Game Mode. The desktop you see is a panel: close it from the Steam dashboard.",
+          "state": {
+            "mode": "game",
+            "gamescope": true,
+            "plasma": false,
+            "nested_desktop": false,
+            "configured": "gamescope-wayland.desktop",
+            "config_file": "/etc/sddm.conf.d/steamos.conf",
+            "selector": "steamosctl",
+            "session_targets": {
+              "gamescope-session.target": "active",
+              "plasma-workspace-wayland.target": "inactive"
+            },
+            "mode_source": "session-targets"
+          },
+          "switched": false
+        },
+        "utc": "2026-10-24T09:27:33.126Z"
+      }
+    ],
+    "final_check": {
+      "http": 200,
+      "body": {
+        "mode": "game",
+        "gamescope": true,
+        "plasma": false,
+        "nested_desktop": false,
+        "configured": "gamescope-wayland.desktop",
+        "config_file": "/etc/sddm.conf.d/steamos.conf",
+        "selector": "steamosctl",
+        "session_targets": {
+          "gamescope-session.target": "active",
+          "plasma-workspace-wayland.target": "inactive"
+        },
+        "mode_source": "session-targets"
+      }
+    },
+    "final_return_disabled": true,
+    "final_text": "Game Mode: the headset is running normally. A desktop you see is a panel; close it from the Steam dashboard.",
+    "screenshots": [
+      "return-game-mode-packaged-2026-10-24-before.png",
+      "return-game-mode-packaged-2026-10-24-after.png"
+    ],
+    "screenshot_scope": "Cropped to the session controls and repeat-action notification; host identity, connection address and backup folder excluded."
+  },
+  "systemd_verification": {
+    "command": "ssh frame systemctl --user show gamescope-session.target plasma-workspace-wayland.target -p Id -p ActiveState -p SubState; ssh frame systemctl --user show gamescope-session.service -p InvocationID",
+    "before": "Id=gamescope-session.target\nActiveState=inactive\nSubState=dead\n\nId=plasma-workspace-wayland.target\nActiveState=active\nSubState=active\nInvocationID=892b843dbe1a4d2ba91b93057bd38463\n295589 /home/steamos/.local/share/Steam/steamrtarm64/steam -silent -vrdisable -deckard\n295066 /opt/steamvr/bin/linuxarm64/vrserver -waitformonitor -ipccontrolfilepath /tmp/SteamVR-IPCControlFile-294978\n",
+    "after_first_click": "Id=gamescope-session.target\nActiveState=active\nSubState=active\n\nId=plasma-workspace-wayland.target\nActiveState=inactive\nSubState=dead\nInvocationID=e8c1085b3bbd4e79a323b3e4a9bc6355\n297506 /opt/steamvr/bin/linuxarm64/vrserver -waitformonitor -ipccontrolfilepath /tmp/SteamVR-IPCControlFile-297420\n",
+    "after_repeat_click": "Id=gamescope-session.target\nActiveState=active\nSubState=active\n\nId=plasma-workspace-wayland.target\nActiveState=inactive\nSubState=dead\nInvocationID=e8c1085b3bbd4e79a323b3e4a9bc6355\n297826 /home/steamos/.local/share/Steam/steamrtarm64/steam -cef-enable-debugging -deckard -gamepadui -steamdeck -steamos3 -vrgamepadui\n297942 /home/steamos/.local/share/Steam/steamrtarm64/steam -child-update-ui -child-update-ui-socket 8 -cef-enable-debugging -deckard -gamepadui -steamdeck -steamos3 -vrgamepadui\n297506 /opt/steamvr/bin/linuxarm64/vrserver -waitformonitor -ipccontrolfilepath /tmp/SteamVR-IPCControlFile-297420\n",
+    "repeat_did_not_restart_game_mode": true,
+    "unchanged_game_service_invocation": "e8c1085b3bbd4e79a323b3e4a9bc6355",
+    "unchanged_steamvr_pid": 297506,
+    "delayed_final_readback_utc": "2026-10-24T09:28:05Z"
+  },
+  "cleanup": {
+    "complete": true,
+    "final_readback": "2026-10-24T09:28:05Z\nId=gamescope-session.target\nActiveState=active\n\nId=plasma-workspace-wayland.target\nActiveState=inactive\nInvocationID=e8c1085b3bbd4e79a323b3e4a9bc6355\nNEXT LEFT LAST PASSED UNIT ACTIVATES\n\n0 timers listed.\nId=fc-return-button-rollback-20261024.timer\nLoadState=not-found\nActiveState=inactive\n\nId=fc-return-button-rollback-20261024.service\nLoadState=not-found\nActiveState=inactive\n\nId=fc-return-button-probe-20261024.timer\nLoadState=not-found\nActiveState=inactive\n\nId=fc-return-button-probe-20261024.service\nLoadState=not-found\nActiveState=inactive\n297826 /home/steamos/.local/share/Steam/steamrtarm64/steam -cef-enable-debugging -deckard -gamepadui -steamdeck -steamos3 -vrgamepadui\n297506 /opt/steamvr/bin/linuxarm64/vrserver -waitformonitor -ipccontrolfilepath /tmp/SteamVR-IPCControlFile-297420\n99\nCharging\ngame\n",
+    "owned_lock_release_exit": 0,
+    "owned_lock_absent": true,
+    "host_processes": {
+      "xvfb": {
+        "pid": 1562020,
+        "exists": false,
+        "cmd": null,
+        "state": null
+      },
+      "electron": {
+        "pid": 1562023,
+        "exists": false,
+        "cmd": null,
+        "state": null
+      },
+      "x_socket_absent": true
+    },
+    "remaining_test_servers": [],
+    "battery_final_percent": 99,
+    "battery_final_status": "Charging",
+    "battery_range_observed_percent": [
+      99,
+      100
+    ],
+    "game_mode_steam_steamvr_running": true,
+    "default_login_mode_after": "game",
+    "headset_installs_or_global_settings_changed": false,
+    "rebooted": false,
+    "notes": "Stopping the timer succeeded. systemctl stop also named its never-started, unloaded service and returned 5; explicit readback confirms both rollback and probe timer/service units are not-found/inactive. No further switch was sent."
+  },
+  "attempts_and_harness_limits": [
+    "Initial harmless timer probe was garbage-collected before its exit metadata could be read. No Desktop Mode switch was attempted. Repeated with RemainAfterExit=yes and verified successful execution.",
+    "First full Plasma attempt clicked the visible packaged Return button and got HTTP 200 / switched=true. The harness then treated absent pgrep matches during the expected session restart as a fatal error. Independent readback confirmed Game Mode and SteamVR had returned. Rollback remained armed until that readback, then was cancelled. This incomplete attempt was not counted as the final pass.",
+    "Final full Plasma attempt tolerated transiently absent Steam/SteamVR while polling systemd targets, passed both trusted visible clicks, and completed the final Check. Stopping the unused rollback service returned unit-not-loaded; explicit subsequent readback verified all four owned probe/rollback units not-found/inactive and zero timers. Cleanup was completed before releasing our lock."
+  ],
+  "limits": [
+    "Unchanged AppImage payload ran after extraction on a virtual X desktop with Chromium sandbox and GPU disabled. FUSE mounting, desktop integration and hardware rendering were not tested.",
+    "Automated trusted pointer clicks in a visible Electron window; no physical-monitor human click or worn-headset visual/interaction assessment.",
+    "Verified only SteamOS 0.5.6 build 20261015.6141207 and the Linux x86_64 v0.4.17 package; older selectors/builds and other desktop platforms remain covered only by their existing tests.",
+    "No release, tag, application code, passwords or security/global settings changed."
+  ],
+  "validation": {
+    "focused_suite": "python3 -m unittest discover -s tests -p test_backup.py -v: 28 tests passed in 2.479s",
+    "json": "python3 -m json.tool: passed",
+    "diff": "git diff --check: passed"
+  }
+}
@@ -190,7 +190,24 @@ HTTP focus is not evidence of worn-headset laser input. Follow the

 ## Saved spatial layouts: blocked on the current panel route

-**Verified 2026-09-28**, same build, using a temporary xterm panel with
+**Rechecked 2026-10-24**, SteamOS 0.5.6, BUILD_ID `20261015.6141207`,
+SteamVR 2.18.2: a read-only probe of the existing Gamescope and Steam main
+panels still reports type 5 (`DashboardTab`), and both the absolute and
+tracked-device-relative getters return 18 (`WrongTransformType`). Width and
+curvature read successfully. **Some pose readback is available:**
+`GetTransformForOverlayCoordinates` at the panel center succeeds in standing
+space; seated and raw origins return 23 (`RequestFailed`). Both panels and the
+dashboard were hidden. That matrix snapshot does not identify docking, prove
+visible placement, or establish a restore operation. The shipped dashboard
+stores dock location and relative transforms internally; no supported external
+snapshot/restore pair was established. No setters or docking commands were run
+in this recheck. See the [exact probe, results, cleanup and remaining
+gaps](evidence/panel-layout-2026-10-24.md). Saved layouts remain unavailable.
+
+### Earlier mutation evidence (2026-09-28)
+
+**Verified 2026-09-28**, SteamOS 0.4.1, BUILD_ID `20260925.6191901`,
+SteamVR 2.18.1, using a temporary xterm panel with
 `STEAM_GAME=2000999031` and `FnTable:IVROverlay_028` from
 `/opt/steamvr/bin/linuxarm64/libopenvr_api.so`:

@@ -107,6 +107,12 @@ The Linux adapter from #38 (desktop portal, PipeWire, VA-API) is in the tree,
 because it shares `capture.c` and the controller, and CI still builds and tests
 it. The app never offers it and the Linux installers don't include it.

+The [2026-10-24 lxso2 feasibility check](evidence/linux-pc-host-lxso2-2026-10-24.md)
+found no running or activatable desktop portal, no PipeWire socket and missing
+native build dependencies. Linux support remains disabled. The evidence records
+the build failure and prerequisites for a screen-only, view-only x264 slice;
+Linux capture, input and window modes have not been verified there.
+
 ## Measuring

 The Mac benchmark accepts the PC host:
@@ -7,6 +7,9 @@ This covers three directions, plus input:
 - **C. iPhone → Frame**: mirror the phone inside the headset.
 - **PC VR from Linux**: [feasibility and options](linux-vr-streaming.md),
  including Valve's streaming and USB support. No Linux host tested yet.
+- **Linux desktop → Frame panels**: [PC-host feasibility on lxso2](evidence/linux-pc-host-lxso2-2026-10-24.md).
+  Blocked on missing desktop portal/PipeWire and native build dependencies;
+  Linux desktop capture remains disabled and unverified.
 - **Input**: type and point in the Frame from the Mac or iPhone.
 - **Live view and Control**: watch a panel flat and tap on it to use it.

@@ -151,10 +154,31 @@ two fingers to scroll, two-finger tap to right-click) plus a text field that
 types on the Frame. On a computer, clicking the pad passes your mouse and
 keyboard through to the Frame until you press Esc (⌘ is sent as Ctrl on a Mac).

-It goes through **KDE Connect**, the first-party route (KDE makes the Frame's
-desktop): Frame Control's server runs [`ui/frame_input_agent.py`](../ui/frame_input_agent.py)
-on the Frame, which talks KDE Connect's own LAN protocol to the Frame's
-`kdeconnectd` as if it were a phone. KDE Connect does the typing and clicking.
+Since Control (below), the pad goes through **gamescope's own input socket**
+([`ui/frame_touch.py`](../ui/frame_touch.py)), so it reaches whichever panel has
+focus, Steam's own menus included, and needs nothing installed on the Frame.
+Only characters that aren't on a US keyboard (accents, emoji) go through
+**KDE Connect** as below; it starts the first time you type one. (Before
+Control, everything went through KDE Connect, which is why the notes below
+describe it in detail.)
+
+The server keyboard queue holds at most 200 events, including text being sent.
+A batch with an accent sends its whole keyboard part through KDE Connect, and
+later keys follow it there while typing is pending. If the batch will not fit,
+the server rejects it before sending any clicks or keys; the page keeps it and
+retries after polling. Accepted events are never evicted to make room. A timeout
+or helper failure retains unsent text and shows **Try again**; that button resumes
+the retained queue. The queue lives in memory and is lost if the server exits.
+Switching headsets cancels queued text and stops both input helpers before the
+new route is published. Both controls validate the intended headset before they
+start. Ordering between the two input services uses a typing-time estimate;
+neither service acknowledges when characters reach the application.
+
+[PR #66 verification on 2026-10-24](evidence/pad-gamescope-2026-10-24.md)
+records an owned focused X11 panel receiving a click and `Frame66` plus Enter
+on a real Frame with SteamOS 0.5.6, after the shared headset became available.
+The helper, window and owner-marked lock were cleaned up. This is ASCII input
+proof, not wearer-visible, Unicode or packaged Electron GUI verification.

 **Verified 2026-09-28** (SteamOS 0.4.1, build 20260925.6191901):

@@ -292,6 +316,10 @@ How (**verified 2026-09-29**, SteamOS 0.4.1, build 20260925.6191901):
  if it persists, report the exact message and SteamOS version. The confirmed
  empty-list advice is retained when the property is present but empty.
  [2026-10-24 investigation and verification limits](evidence/panel-discovery-2026-10-24.md).
+  A later [packaged v0.4.17 test on the real Frame](evidence/desktop-capture-packaged-2026-10-24.md)
+  captured a mapped, focused nested KDE compositor on SteamOS 0.5.6. SteamVR
+  still reported its overlay hidden, so that test does not establish the
+  wearer's visible-desktop condition in #73 or justify closing it.
 - Tested from the iPhone app (Simulator): a tap on the Desktop view focused a
  text box in the panel and the text field typed into it; a trackpad move went
  exactly (+40, +25).
@@ -12,6 +12,16 @@ A fourth, a SteamOS VM, may come later ([issue #6](https://github.com/saphid/ste

 ## Unit tests

+The gamescope keyboard/trackpad route has focused suites in `test_input.py`,
+`test_touch.py` and `test_service_regressions.py`. They cover pad translation,
+ASCII/modifier keys, Unicode queuing/retries, stale panel targets, release
+handling, route cancellation, and intended-headset start/status guards.
+`tests/page/pad_queue.mjs` and `tests/page/ctrl_gestures.mjs` execute the page's
+real queue and gesture functions in Node. These are offline tests; their
+acknowledgements do not prove that a real application received input.
+See [the 2026-10-24 PR #66 evidence](evidence/pad-gamescope-2026-10-24.md)
+for the real focused-panel click/key observation, cleanup and remaining limits.
+
 The read-only power/heat sampler has fixture and transport tests in
 `tests/test_power_sample.py`. Run
 `python3 -m unittest discover -s tests -p test_power_sample.py -v`.
@@ -165,6 +175,14 @@ Record the build, actual interaction results, cleanup and any unworn-headset
 limits alongside screenshots or logs. These are caller responsibilities; the
 smoke script below does not acquire this shared lock itself.

+[Packaged Desktop Capture evidence, 2026-10-24](evidence/desktop-capture-packaged-2026-10-24.md)
+records v0.4.17 GUI button captures of a real mapped nested desktop on SteamOS
+0.5.6, a bounded owned user unit, and cleanup back to the original Game Mode
+state. Check X window mapping/focus and SteamVR overlay visibility separately:
+successful capture of a mapped window does not prove the desktop is visible
+to a wearer. That run's overlay stayed `not_visible`; #73 remains unverified
+for its reported visible-desktop/0.4.5 condition. Only state/log metadata is public.
+
 ```sh
 scripts/frame-smoke.sh           # needs `ssh frame` to work without a password
 scripts/frame-smoke.sh --pair    # also pairs a throwaway key: approve it in the headset
@@ -374,7 +392,11 @@ trees on the fake Frame. None of it has run on a headset.

 `tests/test_media.py` covers layout evidence and overrides, OU eye ordering,
 hardware-decoder command construction, malformed splats, stereo parallax and
-fake-Frame library/process ownership. `tests/e2e/test_media_transfer.py` checks the real
+fake-Frame library/process ownership. It also covers optional-surround refusal
+at creation, setup, first upload and retry after standby, fatal screen refusal,
+and best-effort overlay teardown with shutdown.
+Run `python3 -m unittest discover -s tests -p test_media.py -v`.
+`tests/e2e/test_media_transfer.py` checks the real
 HTTP/SSH upload and library listing without pretending the fake renders VR.
 Real decode timings and captured stereo output are recorded in
 [vr-video.md](vr-video.md). Generated media only; no external player required.
@@ -101,6 +101,14 @@ Stills and the theatre surround are re-sent after waking. Five minutes
 without an accepted frame is reported as an error. Headset-view captures
 taken during standby show a flat dark frame, not our screen.

+**Optional surround:** if SteamVR refuses the theatre dark surround at startup
+or on a retry after standby, playback continues without it and the player log
+records `Theatre surround unavailable`. Screen errors still fail playback.
+On exit, the player attempts to destroy every owned overlay and calls OpenVR
+shutdown even if an overlay's teardown raises an exception. Refusal and
+teardown errors are covered by fake-OpenVR tests; a real surround refusal has
+not been induced safely. See the [current hardening checks](evidence/media-surround-2026-10-24.md).
+
 ![Media panel in Frame Control while a photo plays on the Frame](img/media-ui-panel.png)

 **Verified failed route:** GStreamer 1.24.2's `playbin` selected
@@ -31,7 +31,8 @@ function page(api) {

 // A request that fails, or reaches a link that had dropped, keeps its keys and releases.
 for (const [label, api] of [["request failed", async () => { throw new Error("offline"); }],
-                            ["not sent", async () => ({ state: "starting", sent: false })]]) {
+                            ["not sent", async () => ({ state: "starting", sent: false })],
+                            ["queue full", async () => ({ state: "waiting", sent: false })]]) {
  const { pad, padSend, shown } = page(api);
  padSend({ dx: 5 });  // sent and lost: a move, fine to drop
  await new Promise(r => setTimeout(r, 0));
@@ -45,4 +46,29 @@ for (const [label, api] of [["request failed", async () => { throw new Error("of
    fail(label + ": " + JSON.stringify(q));
  if (!shown.length || shown.at(-1).state === "ready") fail(label + ": state not shown");
 }
+// Acknowledged text is removed locally, but its pending status must reach padShow
+// so failures reported after acceptance are polled without another keypress.
+{
+  const { pad, padSend, shown } = page(async () => ({ state: "ready", sent: true, queued: 1 }));
+  padSend({ key: "é" });
+  await new Promise(r => setTimeout(r, 0));
+  if (pad.queue.length || shown.at(-1)?.queued !== 1) fail("accepted text is not polled");
+}
+
+// Exercise the real status/poll function: queued text leaves pointer controls
+// usable, still polls, shows an eventual error, and does not silently retry it.
+{
+  const pad = { state: "ready", queue: [] }, elements = {};
+  let poll = null;
+  const show = new Function("pad", "$", "PAD_STATES", "TOUCH", "padFlush", "devGen", "api",
+    "setTimeout", "clearTimeout", grab("padShow") + "; return padShow;")(
+    pad, id => elements[id] ||= { classList: { toggle() {} } }, {}, false, () => {}, 0,
+    async () => ({ state: "error", message: "Text is retained; try again." }),
+    fn => { poll = fn; return 1; }, () => { poll = null; });
+  show({ state: "ready", queued: 1 });
+  if (pad.state !== "ready" || !poll) fail("pending text disabled pointer or skipped poll");
+  await poll();
+  if (pad.state !== "error" || poll || !elements.padHint.textContent.includes("retained"))
+    fail("asynchronous error not surfaced");
+}
 console.log("queue rules ok");
@@ -18,6 +18,7 @@ import tempfile
 import threading
 import time
 import unittest
+from unittest import mock
 from pathlib import Path

 ROOT = Path(__file__).resolve().parent.parent
@@ -631,3 +632,428 @@ class PageQueue(unittest.TestCase):

 if __name__ == "__main__":
    unittest.main()
+
+
+class PadEvents(unittest.TestCase):
+    """The trackpad and key row go through gamescope; accents and emoji go the KDE Connect way."""
+
+    @classmethod
+    def setUpClass(cls):
+        import server
+        cls.pad = staticmethod(server.pad_events)
+
+    def test_moves_clicks_and_scroll(self):
+        out, extra = self.pad([{"dx": 3, "dy": -2}, {"singleclick": True}, {"singlehold": True},
+                               {"singlerelease": True}, {"scroll": True, "dy": 1}, {"rightclick": True}])
+        self.assertEqual(out[0], {"dx": 3, "dy": -2})
+        self.assertEqual(out[1:3], [{"button": "left", "down": True}, {"button": "left", "down": False}])
+        self.assertEqual(out[3:5], [{"button": "left", "down": True}, {"button": "left", "down": False}])
+        self.assertEqual(out[5], {"scroll": [0, -15]})  # up = content follows the finger
+        self.assertEqual(out[6:], [{"button": "right", "down": True}, {"button": "right", "down": False}])
+        self.assertEqual(extra, [])
+
+    def test_keys_and_modifiers(self):
+        out, _ = self.pad([{"specialKey": 12}, {"specialKey": 1, "ctrl": True}, {"key": "c", "ctrl": True}])
+        self.assertEqual(out[:2], [{"key": 28, "down": True}, {"key": 28, "down": False}])
+        self.assertEqual([e["key"] for e in out[2:6]], [29, 14, 14, 29])  # Ctrl+Backspace
+        self.assertEqual([(e["key"], e["down"]) for e in out[6:]],
+                         [(29, True), (46, True), (46, False), (29, False)])  # Ctrl+C
+
+    def test_text_stays_ascii_through_gamescope(self):
+        out, slow = self.pad([{"key": "Hi "}, {"key": "there"}])
+        self.assertEqual(out, [{"text": "Hi there"}])
+        self.assertEqual(slow, [])
+
+    def test_keyboard_goes_to_kde_whole_and_in_order(self):
+        events = [{"key": "a"}, {"key": "é"}, {"specialKey": 12}, {"key": "c", "ctrl": True}, {"dx": 2, "dy": 1}]
+        out, slow = self.pad(events, kde=True)
+        self.assertEqual(out, [{"dx": 2, "dy": 1}])  # the pointer still goes through gamescope
+        self.assertEqual(slow, events[:4])  # the keyboard as sent, in order
+
+    def test_shifted_key_with_modifier(self):
+        out, _ = self.pad([{"key": "A", "ctrl": True}])
+        self.assertEqual([e["key"] for e in out], [29, 42, 30, 30, 42, 29])
+
+
+class PadDelivery(unittest.TestCase):
+    """remote_input's promises: "sent" means the whole batch is taken; accents go once, in order."""
+
+    class Agent:
+        def __init__(self, sent=True, state="ready"):
+            self.sent, self.state, self.got = sent, state, []
+
+        def send(self, events):
+            if events and self.sent:
+                self.got.append(events)
+            return {"state": self.state, "sent": self.sent and bool(events)}
+
+    def setUp(self):
+        import server
+        self.s = server
+        self.saved = (server._touch, server._input, server.KDE_WAIT)
+        server._touch, server._input = self.Agent(), self.Agent()
+        self.reset()
+
+    def tearDown(self):
+        self.s._touch, self.s._input, self.s.KDE_WAIT = self.saved
+        self.settle()
+        self.reset()
+
+    def reset(self):
+        self.s._kde_queue.clear()
+        self.s._kde_error = None
+        self.s._kde_until = self.s._typed_until = 0.0
+        self.s._kde_worker[0] = None
+
+    def settle(self):
+        worker = self.s._kde_worker[0]
+        if worker:
+            worker.join(5)
+
+    def typed(self):
+        return [e for batch in self.s._input.got for e in batch]
+
+    def test_accent_only_batch_is_acknowledged_and_typed_once(self):
+        status = self.s.remote_input({"events": [{"key": "é"}]})
+        self.assertTrue(status["sent"])  # so the page doesn't send it again
+        self.settle()
+        self.assertEqual(self.typed(), [{"key": "é"}])
+
+    def test_mixed_text_stays_in_one_ordered_stream(self):
+        self.s.remote_input({"events": [{"key": "aéb"}, {"specialKey": 12}]})
+        self.settle()
+        self.assertEqual(self.typed(), [{"key": "aéb"}, {"specialKey": 12}])
+        self.assertEqual(self.s._touch.got, [])  # none of it split off through gamescope
+
+    def test_batch_gamescope_didnt_take_is_not_queued(self):
+        self.s._touch = self.Agent(sent=False, state="starting")
+        status = self.s.remote_input({"events": [{"dx": 1, "dy": 1}, {"key": "é"}]})
+        self.assertFalse(status["sent"])  # the page tries the whole batch again...
+        self.assertEqual(self.s._kde_queue, [])  # ...so the accent isn't queued twice
+
+    def test_waits_for_kde_connect_then_sends_without_more_input(self):
+        self.s._input = self.Agent(sent=False, state="starting")
+        self.s.remote_input({"events": [{"key": "é"}]})
+        time.sleep(0.2)
+        self.assertEqual(self.s._kde_queue, [{"key": "é"}])
+        self.s._input.sent = True
+        self.settle()
+        self.assertEqual(self.typed(), [{"key": "é"}])
+        self.assertEqual(self.s._kde_queue, [])
+
+    def test_later_ascii_waits_behind_a_pending_accent(self):
+        self.s._input = self.Agent(sent=False, state="starting")
+        self.s.remote_input({"events": [{"key": "é"}]})
+        self.s.remote_input({"events": [{"key": "x"}]})
+        self.assertEqual(self.s._touch.got, [])  # "x" can't overtake the "é"
+        self.assertEqual(self.s._kde_queue, [{"key": "é"}, {"key": "x"}])
+        self.s._input.sent = True
+        self.settle()
+        self.assertEqual(self.typed(), [{"key": "é"}, {"key": "x"}])
+
+    def test_errors_and_timeout_retain_accepted_text_for_explicit_retry(self):
+        events = [{"key": "é"}] * 150
+        for state in ("error", "starting"):
+            with self.subTest(state=state):
+                self.reset()
+                self.s._input = self.Agent(sent=False, state=state)
+                self.s.KDE_WAIT = 0
+                self.assertTrue(self.s.remote_input({"events": events})["sent"])
+                self.settle()
+                self.assertEqual(self.s._kde_queue, events)
+                self.assertIsNone(self.s._kde_worker[0])
+                self.assertEqual(self.s._kde_until, 0)
+                status = self.s.pad_status({"state": "ready"})
+                self.assertEqual(status["state"], "error")
+                self.assertIn("retained", status["message"])
+                self.assertFalse(self.s.remote_input({"events": [{"key": "x"}]})["sent"])
+                self.s._input = self.Agent()
+                self.s.start_pad()
+                self.settle()
+                self.assertEqual(self.typed(), events)
+                self.assertEqual(self.s._kde_queue, [])
+                self.assertIsNone(self.s._kde_error)
+
+    def test_worker_exception_and_start_failure_are_visible_and_retryable(self):
+        for start_failure in (False, True):
+            with self.subTest(start_failure=start_failure):
+                self.reset()
+                events = [{"key": "é"}]
+                if start_failure:
+                    with mock.patch.object(threading.Thread, "start", side_effect=RuntimeError("no thread")):
+                        self.assertTrue(self.s.remote_input({"events": events})["sent"])
+                else:
+                    with mock.patch.object(self.s._input, "send", side_effect=OSError("send failed")):
+                        self.assertTrue(self.s.remote_input({"events": events})["sent"])
+                        self.settle()
+                self.assertEqual(self.s._kde_queue, events)
+                self.assertIsNone(self.s._kde_worker[0])
+                self.assertEqual(self.s._kde_until, 0)
+                self.assertEqual(self.s.pad_status({"state": "ready"})["state"], "error")
+                self.s._input = self.Agent()
+                self.s.start_pad()
+                self.settle()
+                self.assertEqual(self.typed(), events)
+
+    def test_ascii_just_after_an_accent_stays_behind_it(self):
+        self.s.remote_input({"events": [{"key": "é"}]})
+        self.settle()
+        self.s.remote_input({"events": [{"key": "x"}]})  # KDE Connect may still be typing the é
+        self.settle()
+        self.assertEqual(self.s._touch.got, [])
+        self.assertEqual(self.typed(), [{"key": "é"}, {"key": "x"}])
+
+    def test_accent_just_after_gamescope_text_waits_for_it(self):
+        self.s.remote_input({"events": [{"key": "x" * 50}]})
+        self.assertEqual(self.s._touch.got, [[{"text": "x" * 50}]])
+        start = time.time()
+        self.s.remote_input({"events": [{"key": "é"}]})
+        self.settle()
+        self.assertGreater(time.time() - start, 0.6)  # 50 keys at 16 ms, plus the margin
+        self.assertEqual(self.typed(), [{"key": "é"}])
+
+    def test_typing_time_counts_every_key_transition(self):
+        t = self.s.typing_seconds
+        self.assertAlmostEqual(t([{"text": "ab"}]), 4 * 0.008)
+        self.assertAlmostEqual(t([{"text": "A"}]), 4 * 0.008)  # shift down, key down, key up, shift up
+        self.assertAlmostEqual(t([{"text": "x" * 500}]), 8.0)  # a long paste
+        self.assertAlmostEqual(t([{"key": 29, "down": True}, {"dx": 3, "dy": 0}]), 0.008)
+
+    def test_the_margin_is_added_once_not_per_request(self):
+        for _ in range(10):
+            self.s.remote_input({"events": [{"key": "ab"}]})
+        # ten requests of 4 transitions each: the work queues up, with no margin added per request
+        self.assertLess(self.s._typed_until - time.time(), 10 * 4 * 0.008 + 0.05)
+
+    def hold_worker(self):
+        entered, release = threading.Event(), threading.Event()
+        class Gate:
+            def wait(self, timeout):
+                entered.set()
+                if not release.wait(3):
+                    raise RuntimeError("test did not release worker")
+                return False
+        patch = mock.patch.object(self.s, "_kde_cancel", Gate())
+        patch.start()
+        self.addCleanup(patch.stop)
+        self.addCleanup(release.set)
+        return entered, release
+
+    def test_saturation_rejects_whole_batch_and_retry_delivers_every_accepted_event_once(self):
+        entered, release = self.hold_worker()
+        self.s._typed_until = time.time() + 60
+        first = [{"key": "é"}] * 150
+        remainder = [{"key": "ü"}] * 50
+        overflow = [{"singleclick": True}, {"key": "ß"}] * 75
+        self.assertTrue(self.s.remote_input({"events": first})["sent"])
+        self.assertTrue(entered.wait(1))  # batch is in flight, waiting for prior ASCII
+        try:
+            self.assertTrue(self.s.remote_input({"events": remainder})["sent"])
+            status = self.s.remote_input({"events": overflow})
+            self.assertFalse(status["sent"])
+            self.assertEqual(status["state"], "waiting")
+            self.assertEqual(self.s._kde_queue, first + remainder)
+            self.assertEqual(len(self.s._kde_queue), self.s.KDE_QUEUE_LIMIT)
+            self.assertEqual(self.s._touch.got, [])  # rejected clicks cannot be duplicated
+        finally:
+            self.s._typed_until = 0
+            release.set()
+            self.settle()
+        self.assertEqual(self.typed(), first + remainder)
+        self.assertTrue(self.s.remote_input({"events": overflow})["sent"])
+        self.settle()
+        self.assertEqual(self.typed(), first + remainder + [{"key": "ß"}] * 75)
+        self.assertEqual(self.s._touch.got, [[e for _ in range(75) for e in (
+            {"button": "left", "down": True}, {"button": "left", "down": False})]])
+
+    def test_retrying_batch_keeps_its_capacity_and_order(self):
+        entered, release = self.hold_worker()
+        self.s._input = self.Agent(sent=False, state="starting")
+        first, second = [{"key": "é"}] * 150, [{"key": "ü"}] * 150
+        self.assertTrue(self.s.remote_input({"events": first})["sent"])
+        self.assertTrue(entered.wait(1))  # first send failed; worker is in retry delay
+        try:
+            self.assertFalse(self.s.remote_input({"events": second})["sent"])
+            self.assertEqual(self.s._kde_queue, first)
+        finally:
+            self.s._input.sent = True
+            release.set()
+            self.settle()
+        self.assertTrue(self.s.remote_input({"events": second})["sent"])
+        self.settle()
+        self.assertEqual(self.typed(), first + second)
+
+    def test_concurrent_requests_cannot_both_claim_the_last_capacity(self):
+        entered, release = self.hold_worker()
+        self.s._typed_until = time.time() + 60
+        first = [{"key": "é"}]
+        self.s.remote_input({"events": first})
+        self.assertTrue(entered.wait(1))
+        barrier, results = threading.Barrier(3), []
+        def request(key):
+            barrier.wait(2)
+            batch = [{"key": key}] * 150
+            results.append((batch, self.s.remote_input({"events": batch})["sent"]))
+        threads = [threading.Thread(target=request, args=(key,)) for key in ("ü", "ß")]
+        try:
+            for thread in threads:
+                thread.start()
+            barrier.wait(2)
+            for thread in threads:
+                thread.join(2)
+                self.assertFalse(thread.is_alive())
+            self.assertEqual(sorted(sent for _, sent in results), [False, True])
+        finally:
+            self.s._typed_until = 0
+            release.set()
+            self.settle()
+        accepted = next(batch for batch, sent in results if sent)
+        rejected = next(batch for batch, sent in results if not sent)
+        self.assertEqual(self.typed(), first + accepted)
+        self.assertTrue(self.s.remote_input({"events": rejected})["sent"])
+        self.settle()
+        self.assertEqual(self.typed(), first + accepted + rejected)
+
+    def test_input_queued_as_the_worker_finishes_is_still_sent(self):
+        for _ in range(40):
+            self.reset()
+            self.s._input.got.clear()
+            self.s.remote_input({"events": [{"key": "é"}]})
+            time.sleep(0.002)
+            self.s.remote_input({"events": [{"key": "ü"}]})
+            self.settle()
+            self.assertEqual(self.s._kde_queue, [])
+            self.assertEqual(self.typed(), [{"key": "é"}, {"key": "ü"}])
+
+
+class PadRoute(unittest.TestCase):
+    def test_unicode_start_and_headset_switch_do_not_deadlock(self):
+        # Real InputAgent.send/start takes the route lock. Race it with route(),
+        # which cancels the pad worker; mocks of send() hide this lock inversion.
+        # A child keeps a regression's blocked daemon threads out of this suite.
+        code = r'''
+import sandbox
+import server as s
+import threading
+from unittest import mock
+
+entered, release, switching = threading.Event(), threading.Event(), threading.Event()
+s._input = s.InputAgent(packages=[])
+s._input._launch = lambda *args: None  # exercise real start, without any SSH
+s._touch = mock.Mock()
+send = s._input.send
+def held(events):
+    entered.set()
+    assert release.wait(2)
+    return send(events)
+s._input.send = held
+s._power.stop = lambda *args: switching.set()
+s._queue_for_kde([{'key': 'é'}])
+worker = s._kde_worker[0]
+assert entered.wait(2), 'Unicode worker never reached send'
+switch = threading.Thread(target=s.route, args=('frame-pr66-new', []), daemon=True)
+switch.start()
+switching.wait(0.2)
+release.set()
+worker.join(2)
+switch.join(2)
+assert not worker.is_alive(), 'Unicode worker deadlocked starting input during headset switch'
+assert not switch.is_alive(), 'headset switch deadlocked cancelling Unicode input'
+assert not s._kde_queue
+'''
+        env = {**os.environ, "PYTHONPATH": os.pathsep.join([str(ROOT / 'tests'), str(ROOT / 'ui')])}
+        result = subprocess.run([sys.executable, '-c', code], env=env, capture_output=True,
+                                text=True, timeout=8)
+        self.assertEqual(result.returncode, 0, result.stdout + result.stderr)
+
+    def setUp(self):
+        import server
+        self.s = server
+        self.agents = mock.patch.multiple(server, _touch=mock.Mock(), _input=mock.Mock())
+        self.agents.start()
+        self.addCleanup(self.agents.stop)
+        self.s._touch.send.return_value = {"state": "ready", "sent": True}
+        self.s._input.send.return_value = {"state": "ready", "sent": True}
+        self.s.reset_pad_route()
+        self.s._touch.reset_mock(); self.s._input.reset_mock()
+        self.addCleanup(self.s.reset_pad_route)
+
+    def switch(self):
+        with mock.patch.multiple(self.s, FRAME="frame-a", HOST_OPTS=[], MUX=[], SSH=[], macview=None), \
+             mock.patch.multiple(self.s.frame_android, FRAME="frame-a", SSH_OPTS=[]):
+            self.s.route("frame-b", [])
+
+    def test_stale_worker_exception_cannot_damage_new_route(self):
+        entered, release = threading.Event(), threading.Event()
+        class BrokenWait:
+            def wait(self, timeout):
+                entered.set()
+                if not release.wait(2):
+                    raise RuntimeError("test did not release old worker")
+                raise OSError("old route wait failed")
+            def set(self):
+                pass  # hold its exception until the new route has been published
+        self.s._kde_cancel = BrokenWait()
+        self.s._typed_until = time.time() + 60
+        self.s.remote_input({"events": [{"key": "é"}]})
+        worker = self.s._kde_worker[0]
+        try:
+            self.assertTrue(entered.wait(1))
+            self.switch()
+            marker = object()
+            with self.s._kde_lock:
+                self.s._kde_queue.append({"key": "new"})
+                self.s._kde_worker[0] = marker
+                self.s._kde_until = 42
+        finally:
+            release.set()
+            worker.join(2)
+        self.assertFalse(worker.is_alive())
+        self.assertEqual(self.s._kde_queue, [{"key": "new"}])
+        self.assertIs(self.s._kde_worker[0], marker)
+        self.assertEqual(self.s._kde_until, 42)
+        self.assertIsNone(self.s._kde_error)
+
+    def test_switch_stops_both_input_agents_and_discards_old_queue(self):
+        self.s._kde_queue.append({"key": "old"})
+        self.s._kde_until = self.s._typed_until = time.time() + 60
+        cancel = self.s._kde_cancel
+        self.switch()
+        self.s._input.stop.assert_called_once()
+        self.s._touch.stop.assert_called_once()
+        self.assertEqual(self.s._kde_queue, [])
+        self.assertEqual((self.s._kde_until, self.s._typed_until), (0, 0))
+        self.assertTrue(cancel.is_set())
+
+    def test_switch_cancels_delayed_and_retrying_text_before_new_input(self):
+        for delaying in (True, False):
+            with self.subTest(delaying=delaying):
+                waiting, cancel = threading.Event(), threading.Event()
+                class ObservedCancel:
+                    def wait(self, timeout):
+                        waiting.set()
+                        return cancel.wait(timeout)
+                    def set(self):
+                        cancel.set()
+                self.s.reset_pad_route()
+                self.s._kde_cancel = ObservedCancel()
+                self.s._typed_until = time.time() + 60 if delaying else 0
+                self.s._input.reset_mock()
+                self.s._input.send.return_value = {"state": "starting", "sent": False}
+                self.s.remote_input({"events": [{"key": "é"}]})
+                worker = self.s._kde_worker[0]
+                self.assertTrue(waiting.wait(1), "worker never reached its cancellable delay")
+                self.switch()
+                worker.join(1)
+                self.assertFalse(worker.is_alive())
+                attempts = self.s._input.send.call_count
+                self.assertEqual(attempts, 0 if delaying else 1)
+                self.s._input.send.return_value = {"state": "ready", "sent": True}
+                self.s.remote_input({"events": [{"key": "ü"}]})
+                next_worker = self.s._kde_worker[0]
+                if next_worker:
+                    next_worker.join(1)
+                    self.assertFalse(next_worker.is_alive())
+                self.assertEqual(self.s._input.send.call_count, attempts + 1)
+                self.assertEqual(self.s._input.send.call_args.args[0], [{"key": "ü"}])
+                self.assertEqual(self.s._kde_queue, [])
@@ -68,7 +68,7 @@ class Media(unittest.TestCase):
        self.assertIn('-frames:v', cmd)

    def play_with(self, name, busy=0, on_pixels=None, sleeps=None, info=None,
-                  fail=None, layout='auto'):
+                  fail=None, layout='auto', on_create=None, on_call=None):
        """Run the player against a fake OpenVR; returns (status, pixels calls).

        Handle 0 is the theatre surround and 1 the screen. `fail(handle, n)` makes
@@ -82,11 +82,18 @@ class Media(unittest.TestCase):
            write_status(path, **values)

        class FakeOverlay:
+            created = 0
+
            def create(self, *a, **k):
-                return len(calls)
+                # Handles in creation order: surround 0, then screen 1 (theatre).
+                FakeOverlay.created += 1
+                if on_create:
+                    on_create(*a, **k)
+                return FakeOverlay.created - 1

            def call(self, *a):
-                pass
+                if on_call:
+                    on_call(*a)

            def pixels(self, handle, data, w, h):
                calls.append((handle, w, h))
@@ -132,6 +139,58 @@ class Media(unittest.TestCase):
        self.assertEqual(result['dropped'], 2)
        self.assertIn((0, 1, 1), calls[3:])

+    def test_theatre_surround_failure_does_not_stop_playback(self):
+        def fail_surround(n):
+            if n == 1:  # the surround's upload is the first pixels call
+                raise RuntimeError('OpenVR SetOverlayRaw failed: 11')
+        result, _ = self.play_with('clip_SBS.mp4', on_pixels=fail_surround)
+        self.assertEqual((result['state'], result['frames']), ('ended', 4))
+
+    def test_theatre_surround_creation_and_setup_failures_are_cosmetic(self):
+        def fail_create(key, *a, **k):
+            if key.endswith('.surround'):
+                raise RuntimeError('OpenVR CreateOverlay failed: 11')
+        def fail_alpha(name, *a):
+            if name == 'SetOverlayAlpha':
+                raise RuntimeError('OpenVR SetOverlayAlpha failed: 11')
+        for hooks in ({'on_create': fail_create}, {'on_call': fail_alpha}):
+            with self.subTest(hooks=hooks), patch('builtins.print') as log:
+                result, calls = self.play_with('clip_SBS.mp4', **hooks)
+            self.assertEqual((result['state'], result['frames']), ('ended', 4))
+            self.assertEqual(sum(h == 1 for h, w, hgt in calls), 4)
+            self.assertIn('Theatre surround unavailable:', log.call_args.args[0])
+
+    def test_theatre_surround_refusal_after_standby_is_not_retried(self):
+        def refuse_retry(n):
+            if n == 3:  # surround busy, screen accepted, surround retry refused
+                raise RuntimeError('OpenVR SetOverlayRaw failed: 11')
+        with patch('builtins.print') as log:
+            result, calls = self.play_with('clip_SBS.mp4', busy=1, on_pixels=refuse_retry)
+        self.assertEqual((result['state'], result['frames'], result['dropped']), ('ended', 4, 0))
+        self.assertEqual(sum(h == 0 for h, w, hgt in calls), 2)
+        self.assertIn('Theatre surround unavailable:', log.call_args.args[0])
+
+    def test_screen_refusal_still_fails_playback(self):
+        def refuse_screen(n):
+            if n == 2:
+                raise RuntimeError('OpenVR SetOverlayRaw failed: 11')
+        with self.assertRaisesRegex(RuntimeError, 'SetOverlayRaw failed: 11'):
+            self.play_with('clip_SBS.mp4', on_pixels=refuse_screen)
+
+    def test_close_attempts_every_overlay_and_shuts_down_after_teardown_errors(self):
+        for error in (RuntimeError('OpenVR error'), ValueError('bad handle'),
+                      InterruptedError('late Stop')):
+            with self.subTest(error=type(error).__name__):
+                overlay = player.Overlay.__new__(player.Overlay)
+                overlay.handles = [10, 20]
+                overlay.vr = unittest.mock.Mock()
+                overlay.call = unittest.mock.Mock(side_effect=[error, None])
+                overlay.close()
+                self.assertEqual(overlay.call.call_args_list, [
+                    unittest.mock.call('DestroyOverlay', 20),
+                    unittest.mock.call('DestroyOverlay', 10)])
+                overlay.vr.VR_ShutdownInternal.assert_called_once_with()
+
    def test_stop_mid_video_reports_stopped(self):
        result, _ = self.play_with('clip_SBS.mp4', on_pixels=lambda n: n == 3 and stop_now())
        self.assertEqual(result['state'], 'stopped')
@@ -296,7 +296,7 @@ class OneServer(unittest.TestCase):
        self.assertIn("Frame Control on", first.stdout.readline())
        second = subprocess.run([sys.executable, str(ROOT / "ui" / "server.py"), "--port", "0"], env=env,
                                capture_output=True, text=True, timeout=60)
-        self.assertEqual(second.returncode, 1)
+        self.assertEqual(second.returncode, 1, second.stderr)
        self.assertIn("already running", second.stderr)
        first.terminate()
        first.wait(10)
@@ -2,6 +2,7 @@
 import sandbox  # noqa: F401
 import contextlib
 import hashlib
+import io
 import json
 import os
 from pathlib import Path
@@ -19,6 +20,28 @@ import frame_agent as agent
 import frame_android as android


+class StartupTests(unittest.TestCase):
+    def test_rejected_server_starts_no_background_work(self):
+        # A losing process exits through Python teardown. Background TLS/index
+        # workers there can crash it, and it must not sweep the owner's files.
+        with mock.patch.object(server, 'LOCAL', False), \
+                mock.patch.object(server, 'PRIVATE', False), \
+                mock.patch.object(sys, 'argv', ['server.py', '--port', '0']), \
+                mock.patch.object(server, 'one_server', side_effect=SystemExit('already running')), \
+                mock.patch.object(server, 'LoopbackServer') as listener, \
+                mock.patch.object(server, 'sweep_tmp') as sweep, \
+                mock.patch.object(server.threading, 'Thread') as worker, \
+                mock.patch.object(server.frame_telemetry, 'start') as telemetry, \
+                mock.patch.object(server.frame_contact, 'start') as contact:
+            with self.assertRaisesRegex(SystemExit, 'already running'):
+                server.main()
+            listener.assert_not_called()
+            sweep.assert_not_called()
+            worker.assert_not_called()
+            telemetry.assert_not_called()
+            contact.assert_not_called()
+
+
 class TargetTests(unittest.TestCase):
    def setUp(self):
        self.active = {'id': 'a', 'name': 'Headset A'}
@@ -32,6 +55,7 @@ class TargetTests(unittest.TestCase):
        self.stack.enter_context(mock.patch.object(server, '_backfill', {'running': False, 'last': 0.0}))
        self.stack.enter_context(mock.patch.object(server.macview, 'retarget'))
        self.stack.enter_context(mock.patch.object(server._input, 'stop'))
+        self.real_devices_action = server.frame_link.devices_action
        self.stack.enter_context(mock.patch.object(server.frame_link, 'devices_action', self.switch))

    def switch(self, link, body, setup, busy):
@@ -90,6 +114,39 @@ class TargetTests(unittest.TestCase):
                server._live_tunnels.discard(proc)
                server.route(*before)

+    def test_input_post_reserves_headset_through_send_and_enqueue(self):
+        body = json.dumps({'events': [{'singleclick': True}, {'key': 'é'}]}).encode()
+        handler = server.Handler.__new__(server.Handler)
+        handler.path = '/api/input'
+        handler.local_request = lambda: True
+        handler.headers = {'Content-Length': str(len(body)), 'X-Frame-Device': 'a'}
+        handler.rfile = io.BytesIO(body)
+        handler.send_json, handler.send_error_json = mock.Mock(), mock.Mock()
+        stages = []
+        def check_switch(stage):
+            self.assertEqual(server._work[0], 1)
+            with self.assertRaises(server.Failure):
+                server.devices_post({'action': 'use', 'id': 'b'})
+            stages.append(stage)
+        def send(events):
+            check_switch('send')
+            return {'state': 'ready', 'sent': True}
+        def enqueue(events):
+            check_switch('enqueue')
+            self.assertEqual(events, [{'key': 'é'}])
+            return True
+        with mock.patch.object(server.frame_link, 'devices_action', self.real_devices_action), \
+                mock.patch.object(server._touch, 'send', side_effect=send), \
+                mock.patch.object(server, '_queue_for_kde', side_effect=enqueue):
+            handler.do_POST()
+        handler.send_error_json.assert_not_called()
+        self.assertTrue(handler.send_json.call_args.args[0]['sent'])
+        self.assertEqual(stages, ['send', 'enqueue'])
+        self.assertEqual(self.active['id'], 'a')
+        self.assertEqual(server._work[0], 0)
+        server.devices_post({'action': 'use', 'id': 'b'})
+        self.assertEqual(self.active['id'], 'b')
+
    def test_backfill_counts_work_until_error_cleanup(self):
        entered, release = threading.Event(), threading.Event()
        writes = []
@@ -380,9 +437,9 @@ class FinalGetStartTests(unittest.TestCase):
        self.active.update(id='b', name='Headset B')
        for name in ('input', 'touch'):
            with self.subTest(name=name):
-                fresh = server.InputAgent(packages=[]) if name == 'input' else server.TouchAgent()
+                fresh = server.TouchAgent()
                handler = self.handler('/api/%s?start=1' % name, 'a')
-                with mock.patch.object(server, '_' + name, fresh), mock.patch.object(fresh, 'start') as start:
+                with mock.patch.object(server, '_touch', fresh), mock.patch.object(fresh, 'start') as start:
                    handler.do_GET()
                start.assert_not_called()
                self.assertEqual(handler.send_error_json.call_args.args[1], 409)
@@ -391,14 +448,46 @@ class FinalGetStartTests(unittest.TestCase):
    def test_desktop_get_start_requires_an_intended_device(self):
        for name in ('input', 'touch'):
            with self.subTest(name=name):
-                fresh = server.InputAgent(packages=[]) if name == 'input' else server.TouchAgent()
+                fresh = server.TouchAgent()
                handler = self.handler('/api/%s?start=1' % name, None)
-                with mock.patch.object(server, '_' + name, fresh), mock.patch.object(fresh, 'start') as start:
+                with mock.patch.object(server, '_touch', fresh), mock.patch.object(fresh, 'start') as start:
                    handler.do_GET()
                start.assert_not_called()
                self.assertEqual(handler.send_error_json.call_args.args[1], 400)
                self.assertEqual(server._work[0], 0)

+    def test_input_get_starts_and_polls_gamescope_without_starting_kde_connect(self):
+        touch, kde = mock.Mock(), mock.Mock()
+        touch.status = {'state': 'ready', 'focus': 7}
+        touch.send.return_value = {'state': 'starting', 'sent': False}
+        with mock.patch.object(server, '_touch', touch), mock.patch.object(server, '_input', kde):
+            start = self.handler('/api/input?start=1', 'a')
+            start.do_GET()
+            touch.send.assert_called_once_with([])
+            start.send_json.assert_called_once_with(touch.send.return_value)
+            status = self.handler('/api/input', 'a')
+            status.do_GET()
+            status.send_json.assert_called_once_with(touch.status)
+        kde.send.assert_not_called()
+        kde.start.assert_not_called()
+
+    def test_input_poll_exposes_retained_text_error_and_start_retries(self):
+        touch = mock.Mock(status={'state': 'ready'})
+        touch.send.return_value = {'state': 'ready', 'sent': False}
+        with mock.patch.object(server, '_touch', touch), \
+                mock.patch.object(server, '_kde_queue', [{'key': 'é'}]), \
+                mock.patch.object(server, '_kde_error', 'Text is retained; try again.'), \
+                mock.patch.object(server, '_kde_worker', [None]), \
+                mock.patch.object(server, '_start_kde_locked') as retry:
+            status = self.handler('/api/input', 'a')
+            status.do_GET()
+            self.assertEqual(status.send_json.call_args.args[0]['state'], 'error')
+            retry.assert_not_called()  # polling never retries typing on its own
+            start = self.handler('/api/input?start=1', 'a')
+            start.do_GET()
+            retry.assert_called_once_with()
+            touch.send.assert_called_once_with([])
+
    def test_get_start_validation_and_thread_registration_are_one_reservation(self):
        attempts = []
        fresh = server.TouchAgent()
@@ -458,10 +547,10 @@ class FinalGetStartTests(unittest.TestCase):
                                    (None, None, '?start=1')):
            for name in ('input', 'touch'):
                with self.subTest(name=name, query=query, local=link is None):
-                    fresh = server.InputAgent(packages=[]) if name == 'input' else server.TouchAgent()
+                    fresh = server.TouchAgent()
                    handler = self.handler('/api/' + name + query, meant)
                    with mock.patch.object(server, 'LINK', link), mock.patch.object(server, 'LOCAL', link is None), \
-                            mock.patch.object(server, '_' + name, fresh), mock.patch.object(fresh, 'start') as start:
+                            mock.patch.object(server, '_touch', fresh), mock.patch.object(fresh, 'start') as start:
                        handler.do_GET()
                    self.assertEqual(start.call_count, 1 if query else 0)
                    handler.send_error_json.assert_not_called()
@@ -150,6 +150,14 @@ class Apply(unittest.TestCase):
        self.assertFalse(self.t.STALE[0])
        self.assertEqual(self.said[-1][1].get("stale"), None)

+    def test_stale_survives_a_bare_release(self):
+        gs = FakeGamescope()
+        panel = self.t.apply(gs, {"button": "left", "down": True, "window": 99, "display": ":1"}, None)
+        self.assertTrue(self.t.STALE[0])
+        self.t.apply(gs, {"button": "left", "down": False}, panel)  # the page's release names no panel
+        self.assertTrue(self.t.STALE[0])
+        self.assertEqual(self.said[-1][1].get("stale"), True)
+
    def test_same_window_id_on_the_other_display_is_another_panel(self):
        gs = FakeGamescope()
        self.t.apply(gs, {"fx": 0.5, "fy": 0.5, "window": 7, "display": ":0"}, None)
@@ -88,12 +88,14 @@ class Overlay:
    def close(self):
        # Best effort: SteamVR removes a disconnected client's overlays anyway,
        # and a teardown error must not overwrite a finished playback's status.
-        for h in reversed(self.handles):
-            try:
-                self.call('DestroyOverlay', h)
-            except RuntimeError:
-                pass
-        self.vr.VR_ShutdownInternal()
+        try:
+            for h in reversed(self.handles):
+                try:
+                    self.call('DestroyOverlay', h)
+                except Exception:  # signals are already ignored here, so Stop isn't lost
+                    pass
+        finally:
+            self.vr.VR_ShutdownInternal()


 def probe(path):
@@ -183,6 +185,9 @@ def play(args):
            except OverlayBusy:
                pending.insert(0, item)
                return
+            except RuntimeError as e:
+                # Only the cosmetic surround is queued here.
+                print('Theatre surround unavailable: %s' % e, flush=True)

    def hold(handle, data, w, h):
        """Keep a still (photo or splat) up until Stop, retrying through standby."""
@@ -203,10 +208,14 @@ def play(args):
    try:
        vr = Overlay()
        if args.theatre:
-            surround = vr.create('framecontrol.media.surround', 40, 4, order=0)
-            vr.call('SetOverlayAlpha', surround, .85)
-            if not show(surround, b'\x00\x00\x00\xff', 1, 1):
-                pending.append((surround, b'\x00\x00\x00\xff', 1, 1))
+            try:
+                surround = vr.create('framecontrol.media.surround', 40, 4, order=0)
+                vr.call('SetOverlayAlpha', surround, .85)
+                if not show(surround, b'\x00\x00\x00\xff', 1, 1):
+                    pending.append((surround, b'\x00\x00\x00\xff', 1, 1))
+            except RuntimeError as e:
+                # Cosmetic: play without the dark surround rather than not at all.
+                print('Theatre surround unavailable: %s' % e, flush=True)
        screen = vr.create('framecontrol.media.screen', 3 if args.theatre else 1.6, 2,
                           plan['layout'] != 'mono', aspect)
        if splat:
@@ -511,8 +511,10 @@ def apply(gs, event, panel):
        say("ready", focus=panel.get("window"), display=panel.get("display"), stale=True)  # the page re-syncs
        STALE[0] = True
        return panel
-    if STALE[0] and not stale:
+    release = event.get("down") is False and ("button" in event or "key" in event)
+    if STALE[0] and not stale and not (release and "window" not in event):
        # Anything that goes through (a trackpad move names no panel) means caught up: stop re-syncing.
+        # A bare release doesn't: the page leaves the panel off releases, so it says nothing about focus.
        STALE[0] = False
        say("ready", focus=(panel or {}).get("window"), display=(panel or {}).get("display"))
    if "fx" in event and panel and panel.get("window"):
@@ -1270,6 +1270,7 @@
    <div class="sub sync-note" id="clipsSyncNote" hidden></div>
    <div class="hint">Clips and background recordings from Steam's game recording in the headset. Export joins one into a single <code>.mp4</code> in <code>~/Videos/SteamFrame</code>, however long it is, without re-encoding it. Recordings you make with <b>Record</b> go there too.</div>
  </section>
+
  </div>

  <div class="page" data-page="games">
@@ -1498,8 +1499,7 @@
    <div class="shelf-head"><h2>Keyboard and trackpad</h2><span class="sub" id="padState"></span><span class="spacer"></span>
      <button class="small" id="padOn">Turn on</button>
    </div>
-    <p class="sub" id="padIntro">Use this computer's keyboard and mouse in apps on the Frame, such as Chromium or the
-      desktop's Linux apps.</p>
+    <p class="sub" id="padIntro">Use this computer's keyboard and mouse in the panel you're using in the headset.</p>
    <div id="padBody" hidden>
    <div class="pad-area" id="padArea" tabindex="0" role="application" aria-label="Trackpad for the Frame">
      <div class="pad-hint" id="padHint"></div>
@@ -1522,10 +1522,9 @@
        <button data-pad-key="12" title="Enter">Enter</button>
      </span>
    </div>
-    <div class="hint">Types and points in apps on the Frame, such as Chromium or the desktop's Linux apps; Steam's own VR menus
-      don't take it. It works through KDE Connect, which comes with Frame Control: the first time, it copies Valve's build
-      for the Frame (3.6 MB, 18 MB unpacked) into your home folder there and pairs with it. No internet needed, and nothing
-      else to install. <a href="#" data-about>Licences</a></div>
+    <div class="hint">Types and points in whichever panel has focus in the headset, through Valve's own input path: nothing to
+      install and no internet needed. Accents and emoji go through KDE Connect, which comes with Frame Control and starts
+      the first time you type one (it copies Valve's build for the Frame, 3.6 MB, into your home folder there). <a href="#" data-about>Licences</a></div>
    </div>
    <div class="shelf-head" style="margin-top:18px"><h2>Clipboard</h2></div>
    <textarea id="clipText" placeholder="Text to put on the Frame's clipboard…"></textarea>
@@ -1738,7 +1737,7 @@
        <button class="small" id="sessionCheck">Check</button></div>
      <div class="sub" id="sessionState">Check which mode the headset is in.</div>
      <div class="row" style="margin-top:12px"><button class="small" id="sessionGame" disabled>Return to Game Mode</button></div>
-      <div class="hint">Uses the same switch as Valve's developer tools (<code>steamos-session-select gamescope</code>).
+      <div class="hint">Uses Valve's supported Game Mode switch for this SteamOS build.
        It closes the desktop and anything open in it. If the headset can't be reached, hold Power for 10 seconds to
        restart it; <a href="https://github.com/saphid/frame-control/blob/main/docs/backup-and-recovery.md#stuck-in-desktop-mode" target="_blank">more ways back</a>.</div>
    </section>
@@ -6088,11 +6087,11 @@ const typesText = el => el?.matches?.('textarea, input:not([type=range], [type=c
 document.addEventListener("focusin", e => document.body.classList.toggle("typing", !!typesText(e.target)));
 document.addEventListener("focusout", () => document.body.classList.remove("typing"));

-// ---- keyboard and trackpad: events go to KDE Connect on the Frame (see ui/frame_input_agent.py) ----
+// ---- keyboard and trackpad: events go to the Frame through gamescope (ui/frame_touch.py); accents and emoji through KDE Connect ----
 const pad = { state: "off", queue: [], sending: false, poll: null, captured: false };
 const PAD_STATES = {
-  off: "", starting: "Connecting…", installing: "Setting up KDE Connect on the Frame (first time only)…",
-  pairing: "Pairing with KDE Connect…", ready: "On",
+  off: "", starting: "Connecting…", installing: "Setting up (first time only)…",
+  pairing: "Pairing with KDE Connect…", waiting: "Waiting for queued text…", ready: "On",
 };
 // KDE Connect's specialKey numbers (plugins/mousepad in its source).
 const PAD_SPECIAL = { Backspace: 1, Tab: 2, ArrowLeft: 4, ArrowUp: 5, ArrowRight: 6, ArrowDown: 7, PageUp: 8,
@@ -6111,14 +6110,14 @@ function padShow(status) {
  $("padArea").classList.toggle("off", !ready);
  $("padBody").hidden = pad.state === "off";
  $("padIntro").hidden = pad.state !== "off";
-  $("padHint").textContent = pad.state === "error" ? status.message || "Couldn't reach KDE Connect on the Frame."
+  $("padHint").textContent = pad.state === "error" ? status.message || "Couldn't reach the Frame."
    : !ready ? (busy ? PAD_STATES[pad.state] : "Turn on to use this as a trackpad for the Frame.")
    : TOUCH ? "Drag to move the pointer · tap to click · two fingers to scroll · two-finger tap to right-click"
    : pad.captured ? "Your mouse and keyboard now control the Frame. Press Esc to stop."
    : "Click here to use your mouse and keyboard on the Frame (Esc to stop), or drag like a trackpad.";
  clearTimeout(pad.poll);
  if (ready && pad.queue.length) padFlush();
-  if (busy) {
+  if (busy || status.queued) {
    const gen = devGen;
    pad.poll = setTimeout(async () => {
      const status = await api('/api/input').catch(e => ({state:'error', message:e.message}));
@@ -6141,7 +6140,7 @@ function padSend(event) {
  // While it reconnects, keys and clicks wait in the queue; pointer moves would be stale.
  // A button release is always kept, so a drag never leaves the button held on the Frame.
  const room = pad.queue.length < 200 || event.singlerelease;
-  const reconnecting = ["starting", "pairing", "installing"].includes(pad.state) && room;
+  const reconnecting = ["starting", "pairing", "installing", "waiting"].includes(pad.state) && room;
  // Even after an error: a button pressed on the Frame must be let go once it's back.
  const release = event.singlerelease && pad.state !== "off";
  if (pad.state !== "ready" && !(reconnecting && !isMove(event)) && !release) return;
@@ -6174,7 +6173,7 @@ async function padFlush() {
    const status = await api("/api/input", { events });
    if (gen !== devGen) return;
    if (!status.sent) padKeep(events);
-    if (status.state !== "ready") padShow(status);
+    if (status.state !== "ready" || status.queued) padShow(status);
  } catch (e) {
    if (gen !== devGen) return;
    padKeep(events);
@@ -69,6 +69,7 @@ import frame_report  # noqa: E402
 import frame_store  # noqa: E402
 import frame_telemetry  # noqa: E402
 import frame_titles  # noqa: E402
+import frame_touch  # noqa: E402
 import frame_webinstall  # noqa: E402
 import frame_vr  # noqa: E402
 import frame_utilities  # noqa: E402
@@ -111,7 +112,9 @@ frame_android.SSH_OPTS = SSH[1:]


 _power = frame_power.Observation()
-_route_lock = threading.Lock()
+# Pad delivery reserves the route before its own locks; InputAgent.start()
+# captures that same route synchronously, so this lock must be reentrant.
+_route_lock = threading.RLock()


 def route(alias, host_opts):
@@ -123,13 +126,10 @@ def route(alias, host_opts):
        moved = alias != FRAME
        if moved:
            _power.stop('Headset switched; received samples retained for the previous headset')
+            reset_pad = globals().get("reset_pad_route")
+            if reset_pad:
+                reset_pad()  # cancel queued input and both agents before publishing the new route
            frame_catalog._env.clear()  # the SteamOS and Lepton builds reports record are per headset
-            # Invalidate launches before publishing the new route. send/stop share
-            # a writer lock, so an old process cannot receive events after this.
-            for name in ("_input", "_touch"):
-                agent = globals().get(name)
-                if agent:
-                    agent.stop()
        FRAME = frame_android.FRAME = alias
        HOST_OPTS = list(host_opts)
        MUX[:] = [*MUX_BASE, *HOST_OPTS]
@@ -1499,12 +1499,244 @@ def licenses():
    return notices


+# KDE Connect's specialKey numbers -> Linux key codes (KEY_BACKSPACE, KEY_TAB, ...).
+PAD_KEYS = {1: 14, 2: 15, 4: 105, 5: 103, 6: 106, 7: 108, 8: 104, 9: 109, 10: 102, 11: 107, 12: 28, 13: 111, 14: 1,
+            **{21 + i: code for i, code in enumerate([59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 87, 88])}}
+PAD_MODS = (("ctrl", 29), ("alt", 56), ("shift", 42), ("super", 125))
+PAD_CLICKS = {"singleclick": "left", "rightclick": "right", "middleclick": "middle"}
+PAD_NOTCH = 15  # scroll units for one wheel notch
+
+
+def typed_in_kde(events):
+    """Whether these events must go the KDE Connect way: they hold an accent or emoji, or
+    earlier ones are queued, being sent or still being typed there (typing must stay in
+    order, so it can't be split between the two ways)."""
+    with _kde_lock:
+        waiting = bool(_kde_queue) or time.time() < _kde_until
+    return waiting or any(ch not in frame_touch.ASCII for e in events for ch in e.get("key", ""))
+
+
+def pad_events(events, kde=False):
+    """The trackpad's and key row's events as (gamescope events, KDE Connect events).
+
+    Gamescope's own input reaches every panel and needs nothing installed, but types only
+    US-keyboard characters. With kde, the keyboard goes to KDE Connect as it was sent, so
+    text stays in order; the pointer always goes through gamescope.
+    """
+    ascii_keys = frame_touch.ASCII
+    out, slow = [], []
+
+    def tap(code, mods):
+        down = [{"key": c, "down": True} for c in mods]
+        out.extend(down + [{"key": code, "down": True}, {"key": code, "down": False}]
+                   + [{"key": c, "down": False} for c in reversed(mods)])
+    for e in events:
+        mods = [code for name, code in PAD_MODS if e.get(name)]
+        if "dx" in e or "dy" in e:
+            if not e.get("scroll"):
+                out.append({k: e[k] for k in ("dx", "dy") if k in e})
+        if e.get("scroll"):
+            out.append({"scroll": [0, -(e.get("dy") or 0) * PAD_NOTCH]})
+        for flag, button in PAD_CLICKS.items():
+            if e.get(flag):
+                out += [{"button": button, "down": True}, {"button": button, "down": False}]
+        if e.get("doubleclick"):
+            out += [{"button": "left", "down": d} for d in (True, False, True, False)]
+        if e.get("singlehold"):
+            out.append({"button": "left", "down": True})
+        if e.get("singlerelease"):
+            out.append({"button": "left", "down": False})
+        if kde and ("key" in e or "specialKey" in e):
+            slow.append({k: e[k] for k in ("key", "specialKey", "ctrl", "alt", "shift", "super") if k in e})
+            continue
+        if "specialKey" in e and e["specialKey"] in PAD_KEYS:
+            tap(PAD_KEYS[e["specialKey"]], mods)
+        if "key" in e:
+            for ch in e["key"]:
+                if ch not in ascii_keys:
+                    continue  # only reached when kde is wrong; the caller checks typed_in_kde first
+                elif mods:
+                    code, shifted = ascii_keys[ch]
+                    tap(code, mods + ([42] if shifted and 42 not in mods else []))
+                else:
+                    if out and "text" in out[-1] and len(out[-1]["text"]) < 100:
+                        out[-1]["text"] += ch
+                    else:
+                        out.append({"text": ch})
+    return out, slow
+
+
 def remote_input(body):
-    """{"events": [...]} sends keyboard and pointer events; {} (or none yet) just starts the agent."""
+    """{"events": [...]} sends keyboard and pointer events; {} (or none yet) just starts the agent.
+
+    "sent" in the answer means the whole batch is taken (the page keeps it and tries again
+    if not), so the keyboard's share is queued only once that's so.
+    """
    events = body.get("events", [])
    if not isinstance(events, list) or len(events) > INPUT_BATCH_LIMIT:
        raise Failure(f"events must be a list of at most {INPUT_BATCH_LIMIT}", 400)
-    return _input.send([input_event(e) for e in events])
+    events = [input_event(e) for e in events]
+    # Serialize admission with other requests and the sender. Keep route -> pad
+    # order: either agent may synchronously capture its route during send().
+    with _route_lock, _pad_send_lock:
+        touch, slow = pad_events(events, typed_in_kde(events))
+        with _kde_lock:
+            blocked = slow and (_kde_error or len(_kde_queue) + len(slow) > KDE_QUEUE_LIMIT)
+        if blocked:
+            # Refuse the WHOLE batch before sending its clicks/moves. The page
+            # retains it and polls, so a retry cannot duplicate pointer events.
+            return pad_status({"state": "waiting", "sent": False})
+        status = _touch.send(touch)
+        if touch and not status.get("sent"):
+            return status
+        if touch:
+            _note_typed(touch)
+        if slow:
+            _queue_for_kde(slow)
+        # Poll at least once after accepting text, even if its worker finished
+        # already, so an asynchronous error is visible without another keypress.
+        return pad_status({**status, "sent": True}, pending=bool(slow))
+
+
+# Typing KDE Connect is to do: it starts the first time it's needed, so this waits for it.
+# Nothing on the Frame says when typed text has landed, so each way holds the other back
+# for as long as what it was given should take (a short margin on a time estimate).
+_kde_queue = []
+_kde_lock = threading.Lock()
+_pad_send_lock = threading.Lock()
+_kde_worker = [None]
+_kde_error = None
+_kde_until = 0.0     # keys typed through KDE Connect should have landed by then
+_typed_until = 0.0   # when gamescope's agent will have typed everything it's been given
+_kde_route = object()
+_kde_cancel = threading.Event()
+KDE_QUEUE_LIMIT = 200
+KDE_WAIT = 90  # seconds to wait before pausing queued text for an explicit retry
+TYPING_SETTLE = 0.6  # seconds past the estimate, once
+
+
+def reset_pad_route():
+    """Discard input for the headset being left, including batches waiting for an agent."""
+    global _kde_route, _kde_cancel, _kde_until, _typed_until, _kde_error
+    with _route_lock, _pad_send_lock, _kde_lock:
+        _kde_cancel.set()
+        _kde_cancel = threading.Event()
+        _kde_route = object()
+        _kde_queue.clear()
+        _kde_worker[0] = None
+        _kde_error = None
+        _kde_until = _typed_until = 0.0
+        _input.stop()
+        _touch.stop()
+
+
+def typing_seconds(touch):
+    """How long gamescope's agent takes over these: it sleeps 8 ms after every key transition."""
+    transitions = 0
+    for e in touch:
+        if "text" in e:
+            transitions += sum(2 + 2 * frame_touch.ASCII[ch][1] for ch in e["text"] if ch in frame_touch.ASCII)
+        elif "key" in e:
+            transitions += 1
+    return transitions * 0.008
+
+
+def _note_typed(touch):
+    global _typed_until
+    seconds = typing_seconds(touch)
+    if seconds:
+        with _kde_lock:  # the end of the work, which queues up; the margin is added once, where it's checked
+            _typed_until = max(_typed_until, time.time()) + seconds
+
+
+def pad_status(status, pending=False):
+    with _kde_lock:
+        if _kde_error:
+            return {**status, "state": "error", "message": _kde_error}
+        if (_kde_queue or pending) and status.get("state") != "error":
+            return {**status, "queued": len(_kde_queue) or 1}
+    return status
+
+
+def start_pad():
+    with _route_lock, _pad_send_lock:
+        status = _touch.send([])
+        with _kde_lock:
+            if _kde_queue and _kde_worker[0] is None:
+                _start_kde_locked()  # explicit Try again retries retained text
+        return pad_status(status)
+
+
+def _start_kde_locked():
+    global _kde_error
+    _kde_error = None
+    try:
+        worker = threading.Thread(target=_drain_kde, args=(_kde_route, _kde_cancel), daemon=True)
+        _kde_worker[0] = worker
+        worker.start()
+    except Exception:
+        _kde_worker[0] = None
+        _kde_error = "Couldn't start queued typing. Text is retained; try again."
+
+
+def _queue_for_kde(events):
+    with _kde_lock:
+        if len(_kde_queue) + len(events) > KDE_QUEUE_LIMIT:
+            return False
+        _kde_queue.extend(events)
+        if _kde_worker[0] is None and not _kde_error:
+            _start_kde_locked()
+        return True
+
+
+def _drain_kde(route, cancel):
+    """One sender, in order. In-flight text stays in the bounded queue until sent.
+    Errors retain it for an explicit retry; only a headset switch cancels it."""
+    global _kde_until, _kde_error
+    deadline = time.time() + KDE_WAIT
+    try:
+        while True:
+            with _kde_lock:
+                if route is not _kde_route:
+                    return
+                batch = _kde_queue[:INPUT_BATCH_LIMIT]
+                if not batch:
+                    _kde_worker[0] = None
+                    return
+                wait = _typed_until + TYPING_SETTLE - time.time()
+                _kde_until = time.time() + 3600  # held while in flight
+            if wait > 0 and cancel.wait(wait):
+                return
+            # Match route()/reset_pad_route()'s lock order. send() may start an
+            # agent and take _route_lock again to capture its immutable route.
+            with _route_lock, _pad_send_lock:
+                with _kde_lock:
+                    if route is not _kde_route:
+                        return
+                status = _input.send(batch)
+            with _kde_lock:
+                if route is not _kde_route:
+                    return
+                if status.get("sent"):
+                    del _kde_queue[:len(batch)]
+                    _kde_until = time.time() + len(batch) * 0.03 + TYPING_SETTLE
+                    deadline = time.time() + KDE_WAIT
+                    continue
+                _kde_until = 0.0
+                if status.get("state") == "error" or time.time() > deadline:
+                    _kde_error = "Queued typing didn't reach the input service. Text is retained; try again."
+                    _kde_worker[0] = None
+                    return
+            if cancel.wait(0.4):
+                return
+    except Exception:
+        # A stale worker must never retire the new route's worker or alter its
+        # queue/timing. Retain unsent text and let the page report/retry failure.
+        with _kde_lock:
+            if route is _kde_route:
+                _kde_until = 0.0
+                _kde_worker[0] = None
+                _kde_error = "Queued typing failed. Text is retained; try again."


 # ---- touch: the headset's panels, through gamescope's own input (frame_touch.py) ----
@@ -3302,7 +3534,7 @@ class Handler(BaseHTTPRequestHandler):
            elif path == "/api/panels":
                self.send_json(panels())
            elif path in ("/api/touch", "/api/input"):
-                agent = _touch if path == "/api/touch" else _input
+                agent = _touch  # both controls use gamescope; KDE Connect starts only for Unicode
                if parse_qs(url.query).get("start") == ["1"]:
                    # Validate and reserve before start() captures its immutable
                    # route and schedules a worker. Status-only polls stay free.
@@ -3310,9 +3542,11 @@ class Handler(BaseHTTPRequestHandler):
                    if LINK and not meant:
                        raise Failure("Control start needs the intended X-Frame-Device headset", 400)
                    with working(meant):
-                        result = agent.send([])
+                        result = start_pad() if path == "/api/input" else agent.send([])
                else:
                    result = dict(agent.status)
+                    if path == "/api/input":
+                        result = pad_status(result)
                self.send_json(result)
            elif path == "/api/job":
                self.send_json(job_status(url.query))
@@ -3637,6 +3871,11 @@ def main():
                    help="stop cleanly when stdin closes (the app closes it on quit; "
                         "Windows has no SIGTERM to catch)")
    args = ap.parse_args()
+    global LINK, _ONE_SERVER
+    if not LOCAL and not PRIVATE:
+        # Refuse before starting daemon workers: the losing process exits through
+        # Python teardown, which can race TLS work and crash instead of exit 1.
+        _ONE_SERVER = one_server()
    httpd = LoopbackServer(("127.0.0.1", args.port), Handler)
    sweep_tmp()
    if os.environ.get('FRAME_COMPAT_SEARCH_REPORTS') != '0':
@@ -3645,10 +3884,7 @@ def main():
    threading.Thread(target=apk_search.warm, daemon=True).start()  # big indexes download before the first search
    frame_telemetry.start()
    frame_contact.start()
-    global LINK, _ONE_SERVER
    if not LOCAL:
-        if not PRIVATE:  # a private server only uses the headsets (see one_server)
-            _ONE_SERVER = one_server()
        LINK = frame_link.Link(frame_devices.Registry(), env_alias=FRAME if FRAME_FROM_ENV else None,
                               mux_base=MUX_BASE, control=CONTROL, apply=route, explain=unreachable)
        LINK.work_lock, LINK.work = _work_lock, lambda: _work[0]
Author	SHA1	Message	Date
SAPHID	e7cb68f287	Retain accepted Unicode input under queue saturation and failure	2026-10-24 11:47:47 +00:00
SAPHID	215db7bfa8	Record real focused-panel click and ASCII key delivery	2026-10-24 11:23:11 +00:00
SAPHID	5641d3c316	Prevent Unicode input startup from deadlocking a headset switch	2026-10-24 11:19:36 +00:00
SAPHID	782a6e7d72	Document gamescope pad verification and shared-device limits	2026-10-24 11:13:47 +00:00
SAPHID	1737366171	Merge current main into pad-gamescope and retain input route guards	2026-10-24 11:10:58 +00:00
Alex SouthwellandSAPHID	0da1fe5077	Bump Frame Control to 0.4.18 (#161 ) Co-authored-by: SAPHID <saphid@users.noreply.github.com>	2026-10-24 10:53:48 +00:00
Alex Southwell Claude Opus 5.5 SAPHID	1f7615b312	Keep theatre playback running without optional surround (#51 ) * Never end playback over the cosmetic surround; always shut down OpenVR Review (SWE-2 Max): a non-standby error re-sending the theatre surround could end a healthy still, and a non-RuntimeError during DestroyOverlay skipped VR_ShutdownInternal. drain() now drops a surround that fails for other reasons, and close() shuts down in finally. Fake overlay handles no longer depend on call order. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> * Play without the theatre surround if SteamVR refuses it Review (SWE-2 Max): a non-standby error creating the cosmetic surround at startup still ended playback before the screen appeared. Theatre now logs it and plays without the surround. close() treats any per-overlay teardown error as best effort (signals are already ignored there). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> * Verify media surround fallback and guard server startup ownership * Record exact-head real Frame theatre playback smoke --------- Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Co-authored-by: SAPHID <saphid@users.noreply.github.com>	2026-10-24 10:45:48 +00:00
Alex SouthwellandSAPHID	2ded4c5a0d	Document saved panel layout feasibility on SteamVR 2.18.2 (#160 ) Co-authored-by: SAPHID <saphid@users.noreply.github.com>	2026-10-24 10:19:35 +00:00
Alex SouthwellandSAPHID	5b47cd3330	Document packaged Desktop Capture result and visibility limit (#159 ) Co-authored-by: SAPHID <saphid@users.noreply.github.com>	2026-10-24 10:03:36 +00:00
Alex SouthwellandSAPHID	aead570b5d	Clarify Return to Game Mode switch help (#158 ) Co-authored-by: SAPHID <saphid@users.noreply.github.com>	2026-10-24 09:47:24 +00:00
Alex SouthwellandSAPHID	de931bc56f	Document packaged Return to Game Mode button verification (#157 ) Co-authored-by: SAPHID <saphid@users.noreply.github.com>	2026-10-24 09:39:21 +00:00
Alex SouthwellandSAPHID	71288d8980	Document packaged desktop Steam clip Export verification (#156 ) Co-authored-by: SAPHID <saphid@users.noreply.github.com>	2026-10-24 09:19:46 +00:00
Alex SouthwellandSAPHID	64b9babe87	Document Linux PC-host feasibility blocker on lxso2 (#155 ) Co-authored-by: SAPHID <saphid@users.noreply.github.com>	2026-10-24 09:17:14 +00:00
SAPHID	cba23fb3af	fix(ui): reconcile PR 66 with current main	2026-10-18 00:37:13 +00:00
SAPHID	62b3cbf705	fix(input): cancel old headset input when switching routes	2026-10-17 23:55:09 +00:00
saphidandClaude Sonnet 5.5	024ec593f9	Merge origin/main into pad-gamescope Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-10-01 21:06:06 +10:00
saphidandClaude Sonnet 5.5	0373555519	Pad: count gamescope's real typing time, with the settle margin added once Review found the estimate used 12 ms a character where the agent takes 16 ms (32 shifted), and added the margin on every request so it piled up. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-10-01 21:01:53 +10:00
saphidandClaude Sonnet 5.5	3539215517	Pad: drain the KDE Connect queue without losing, stranding or reordering input Second review found the queue could be trimmed under an in-flight send (deleting unsent events), the worker could retire as a request enqueued, and ASCII could overtake an accent still being typed (or the reverse). The worker now takes its batch out while sending and puts it back on failure, retires under the lock, and each way waits out the other's estimated typing time. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-10-01 20:57:08 +10:00
saphidandClaude Sonnet 5.5	9f3fb89e30	Pad: keep typing in order and acknowledge it once, with accents through one KDE Connect queue Review found accent-only input repeated forever (sent was false), text split across two transports out of order, accents stuck until the next keypress, duplicated on retry, and racy. A batch with an accent now sends its whole keyboard through KDE Connect; later keys follow while any are queued; the server owns one ordered, bounded queue drained by a single thread; sent means the whole batch is taken. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-10-01 20:52:41 +10:00
saphidandClaude Sonnet 5.5	dc7854d10b	Touch: a bare release no longer clears the stale-focus notice before the page has seen it The page leaves the panel off releases, so a release says nothing about focus. Found in review of the stale-latch fix. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-10-01 20:52:41 +10:00
saphidandClaude Sonnet 5.5	5a6b925448	Trackpad and key row go through gamescope's input; KDE Connect only for accents and emoji The pad reaches whichever panel has focus (Steam's menus too), needs nothing installed, and no longer clamps to 1280x720. Characters that aren't on a US keyboard start KDE Connect the first time they're typed. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-09-29 20:03:37 +10:00
saphidandClaude Sonnet 5.5	60ade8c2e8	Merge origin/main into live-touch Kept both route tables: main's media, agent, assistant and Mac view routes plus /api/touch. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-09-29 11:04:47 +10:00
saphidandClaude Opus 5.5	aa1ac7368c	frame_touch: any event that goes through clears the stale flag A trackpad move names no panel, so waiting for an aimed event could leave the page re-reading panels for the rest of the session; a release still aimed at the old panel doesn't count. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>	2026-09-29 10:13:09 +10:00
saphidandClaude Opus 5.5	46b6966cf7	frame_touch: build the socket path on the Frame, so Windows can import it for tests Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>	2026-09-29 09:41:42 +10:00
saphidandClaude Opus 5.5	be2e770e66	Control: stale clears, pauses reconverge, pastes split per request - The Frame says it has caught up as soon as an aimed event lands after a stale one, so the page stops re-reading the panels. - After a device pause it lets go of everything it holds (releases that arrived while paused were dropped), and waits for the device once per batch, not once per event. - The quick focus check no longer freshens the panel geometry's age. - Each request carries at most about 100 characters of text. - Turning Control off while it connects doesn't report an error. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>	2026-09-29 09:40:00 +10:00
saphidandClaude Opus 5.5	597f98af6d	Control: fixes from the SWE-2 Max review - The Frame side tracks keys as well as buttons and lets go of both when the session ends. - A stale tap tells the page, which re-reads the panels at once. - A paused input device waits instead of ending the session; only a disconnect does. An OS error on one event skips it. - Presses check focus with two property reads and do the full lookup only when it changed. - Writes to an agent's stdin are serialized, so two devices sending at once can't tear a line (the keyboard agent too). - Connecting gives up with a message after 15 s instead of hanging on "Connecting…"; text goes in 100-character pieces so releases don't wait behind a long paste; a cancelled mouse gesture releases what's held. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>	2026-09-29 09:28:28 +10:00
saphidandClaude Opus 5.5	ac07efa8f1	Control: close the targeting gaps from the second review - The focused panel's display comes from GAMESCOPE_FOCUS_DISPLAY (gamescope packs ":1" into the first value), so a window id repeated across :0 and :1 can't be mistaken; the pid is only the fallback. - Presses, keys, text and scrolls read focus afresh on the Frame; only moves use a reading up to a second old. - A gesture remembers the panel it started on and does nothing more if that stops being the one in use; a press with no panel to aim at isn't sent. - Opening a screenshot clears the panel Control would act on; switching to another app releases held keys and buttons. - Trimming keeps a click with its position; the error backoff holds for new input too. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>	2026-09-29 08:52:00 +10:00
saphidandClaude Opus 5.5	0f4f0b863f	Control: fixes from review - Keys held on the Frame are released with buttons when Control stops or the view loses focus; keys for the Frame no longer trigger Frame Control's own shortcuts. - Taps only act when the picture on screen is the panel in use; positions, presses, keys, text and scrolls name their panel (display and window: ids repeat across :0 and :1, told apart by pid), and the Frame drops them if focus has moved on. Releases always go. - While connecting, a tap keeps its position; on an error only releases wait and retries back off; trimming a long queue never drops a release. - Lifting one of two scrolling fingers ends the scroll; a cancelled touch isn't a tap; clicks and holds on the bars around the picture do nothing. - A capture loop from before a Live restart can't stop the new video. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>	2026-09-29 08:40:46 +10:00
saphidandClaude Opus 5.5	e7c81733b1	Live view: a Desktop view that stays still, and Control to tap on the Frame The live view gets a second source and a way to use the Frame from it: - Desktop: the app panel in use in the headset, streamed from its own window (x11grab of gamescope's redirected window), so it doesn't move as the wearer looks around. A picker shows any other panel, view only. - Control: on the Desktop view a tap or click lands exactly where you put it; drag is a mouse drag, press and hold right-clicks, two fingers scroll, and on a computer the mouse, wheel and keyboard work directly. On the headset view the view is a trackpad. A text field and key row type from a phone. Input goes through gamescope's own EIS socket (the way Steam feeds Remote Play input) with the libei already on the image: ui/frame_touch.py, over the same long-lived ssh machinery as the keyboard agent, nothing to install. It reaches the panel that has focus on either X display, which the KDE Connect route can't. Verified on the Frame and from the iPhone app in the Simulator. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>	2026-09-29 08:23:55 +10:00