Merge main into doctor and retain both changelog entries

Merge pull request #126 from saphid/t3/fix-smoke-process-cleanup
fix: stop test-owned smoke processes before removing titles
2026-10-24 08:01:36 +02:00 · 2026-10-24 02:58:52 +00:00 · 2026-10-24 02:58:08 +00:00 · 2026-10-24 02:55:40 +00:00 · 2026-10-24 02:51:38 +00:00 · 2026-10-24 02:49:35 +00:00
426 changed files with 115339 additions and 1060 deletions
@@ -22,6 +22,7 @@ desktop or panels.
 | See the Frame from the Mac, or the Mac inside the Frame | `docs/streaming.md` | `scripts/run-on-frame.sh mac-screen` |
 | Files and clipboard | `docs/file-transfer.md` | `scripts/push.sh`, `scripts/paste-to-frame.sh` |
 | Android apps (Lepton) | `docs/apks.md` | `scripts/install-apk.sh` |
+| Browse and install Android apps from inside the headset (APK Store panel) | `docs/apks.md` | `scripts/store-on-frame.sh` |
 | Reach the Frame off the home LAN (Tailscale) | `docs/tailscale.md` | `scripts/tailscale-on-frame.sh` |
 | Install or buy Steam games, Frame ratings | `docs/steam-games.md` | `ui/frame_steam.py` |
 | Flatpaks | `docs/streaming.md` | `scripts/install-apps.sh` |
@@ -48,3 +49,9 @@ running `--help`: `paste-to-frame.sh`, `serve-bootstrap.sh` and
 - `sudo` on the Frame asks for the user's Developer Mode password. Hand those
  steps to the user (open Terminal) and keep automation to non-sudo commands.
 - The Mac uses BSD userland and zsh (no `timeout`, use `head -n`).
+- **First-party first.** For any new capability, investigate the first-party
+  way before anything else: Valve (SteamOS, Steam, Steam Link), Apple (the Mac
+  and iPhone), and KDE (the Frame's desktop is Plasma). It's usually the best
+  answer. If it isn't, write down why not. If it is, find what Frame Control
+  can do to make it easier to set up (install over SSH, pre-seed settings,
+  pair automatically, tell the user the one setting to turn on).
@@ -6,3 +6,8 @@
 *.json text eol=lf
 *.md text eol=lf
 *.bat text eol=crlf
+# Test fixtures are byte-exact (hashes, signatures): never convert line endings.
+tests/fixtures/** -text
+*.apk binary
+*.jar binary
+*.obb binary
@@ -5,6 +5,11 @@ on:
    branches: [main]
  pull_request:

+# A newer push to the same pull request supersedes a run still in progress.
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
+
 jobs:
  checks:
    runs-on: ubuntu-latest
@@ -17,7 +22,9 @@ jobs:
        with:
          node-version: "24"
      - name: Install zsh
-        run: sudo apt-get update -qq && sudo apt-get install -y -qq zsh
+        # The runner image's third-party apt sources (packages.microsoft.com) sometimes
+        # fail to refresh; Ubuntu's own lists still update, and the install fails if not.
+        run: (sudo apt-get update -qq || true) && sudo apt-get install -y -qq zsh
      - name: Script syntax
        run: |
          sh -n ui/local-bin/ssh
@@ -35,7 +42,9 @@ jobs:
      - name: Server tests
        run: python -m unittest discover -s tests -v
      - name: App syntax
-        run: node --check app/main.js && node --check app/build/make-icon.js && node --check app/build/fetch-deps.js && node --check app/preload.js && node --check app/install-link.js
+        run: node --check app/main.js && node --check app/build/make-icon.js && node --check app/build/fetch-deps.js && node --check app/preload.js && node --check app/install-link.js && node --check app/updater.js
+      - name: App tests
+        run: node --test app/test/*.test.js
      - name: Website
        run: node --test site/test/*.test.mjs && node --check site/public/js/site.js && node --check site/public/js/feedback.js

@@ -82,6 +91,8 @@ jobs:
    steps:
      - uses: actions/checkout@v4
      - name: Install zsh
-        run: sudo apt-get update -qq && sudo apt-get install -y -qq zsh
+        # The runner image's third-party apt sources (packages.microsoft.com) sometimes
+        # fail to refresh; Ubuntu's own lists still update, and the install fails if not.
+        run: (sudo apt-get update -qq || true) && sudo apt-get install -y -qq zsh
      - name: End-to-end tests
        run: scripts/e2e.sh
@@ -0,0 +1,53 @@
+name: PC host libraries
+# PC in the headset (Experimental). The Windows job builds the native adapter
+# and runs the real-encoder protocol tests; the installer's own bundle is
+# tested in release.yml. The Linux adapter from PR #38 isn't shipped or offered
+# in the app, but shares capture.c and controller.c, so it's still built here.
+# These builds add the GPL x264 encoder for headless tests only: they're never
+# uploaded or shipped.
+on:
+  pull_request:
+    paths: ["desktop/**", "ui/**", "tests/**", "scripts/macview-bench.py", ".github/workflows/pc-host.yml"]
+  workflow_dispatch:
+jobs:
+  linux:
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, ubuntu-24.04-arm]
+    runs-on: ${{ matrix.os }}
+    steps:
+      - uses: actions/checkout@v4
+      - name: Native libraries
+        run: |
+          sudo apt-get update -qq
+          sudo apt-get install -y libgstreamer1.0-dev libgstreamer-plugins-base1.0-dev libglib2.0-dev gstreamer1.0-plugins-base gstreamer1.0-plugins-good gstreamer1.0-plugins-bad gstreamer1.0-plugins-ugly gstreamer1.0-pipewire libpipewire-0.3-modules libspa-0.2-modules
+          python3 desktop/build.py --with-x264
+      - name: Tests (real x264, fake capture consent/input)
+        env:
+          FRAME_PC_REQUIRE_NATIVE: "1"
+          FRAME_PC_REQUIRE_X264: "1"
+        run: python3 -m unittest discover -s tests -p 'test_pc*.py' -v
+  windows:
+    runs-on: windows-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: msys2/setup-msys2@v2
+        with:
+          msystem: UCRT64
+          update: true
+      # A mirror's HTTP 429 once failed this install (2026-10-24): retry, so a
+      # flaky mirror can't break a release build.
+      - name: MSYS2 packages for the PC host
+        shell: msys2 {0}
+        run: |
+          for attempt in 1 2 3; do
+            pacman -S --needed --noconfirm mingw-w64-ucrt-x86_64-gcc mingw-w64-ucrt-x86_64-pkgconf mingw-w64-ucrt-x86_64-python mingw-w64-ucrt-x86_64-gstreamer mingw-w64-ucrt-x86_64-gst-plugins-base mingw-w64-ucrt-x86_64-gst-plugins-good mingw-w64-ucrt-x86_64-gst-plugins-bad mingw-w64-ucrt-x86_64-gst-plugins-ugly && exit 0
+            sleep $((attempt * 30))
+          done
+          exit 1
+      - name: Native build and tests (x264; WGC and MF need a desktop/GPU)
+        shell: msys2 {0}
+        run: |
+          python desktop/build.py --with-x264
+          FRAME_PC_REQUIRE_NATIVE=1 FRAME_PC_REQUIRE_X264=1 python -m unittest discover -s tests -p 'test_pc*.py' -v
@@ -2,19 +2,60 @@ name: release

 # Pushing a v* tag builds Frame Control for macOS, Windows and Linux and attaches
 # the installers to that tag's GitHub release (created as a draft if missing).
-# Pull requests that touch the app build the same installers as artifacts.
+# Pull requests that change how the installers are built (the Electron app, the
+# Windows PC helper libraries, the macOS helper, or this workflow) build the same installers as
+# artifacts. ui/, scripts/, frame/ and apk-catalog/ are copied in as they are and
+# are tested by the checks workflow; run this one by hand to package them early.
 on:
  push:
    tags: ["v*"]
  pull_request:
-    paths: ["app/**", "ui/**", "scripts/**", "frame/**", "apk-catalog/**", ".github/workflows/release.yml"]
+    paths: ["app/**", "desktop/**", "mac/frame-mac-view/**", ".github/workflows/release.yml"]
  workflow_dispatch:

+# A newer push to the same pull request supersedes a build still in progress.
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
+
 permissions:
  contents: write

 jobs:
+  # One job makes the draft, before the builds: three matrix jobs each running
+  # "view || create" could race and make duplicate drafts. The draft is tied to
+  # the pushed tag (the tag must already exist on GitHub, then tag_name is set
+  # explicitly), so scripts/publish-release.sh and `gh release upload` find it
+  # by tag. It is created with the REST API so the id comes straight back: the
+  # release list can lag a fresh draft (v0.4.2's first run found nothing and
+  # PATCHed releases/ with an empty id). tests/test_release_workflow.py runs
+  # this step against tests/fakegh.
+  draft:
+    if: startsWith(github.ref, 'refs/tags/')
+    runs-on: ubuntu-latest
+    steps:
+      - name: Create the draft release
+        env:
+          GH_TOKEN: ${{ github.token }}
+          GH_REPO: ${{ github.repository }}
+        run: |
+          tag="${GITHUB_REF_NAME}"
+          # A rerun finds the draft an earlier attempt made.
+          id=$(gh api --paginate "repos/$GH_REPO/releases?per_page=100" --jq ".[] | select(.tag_name == \"$tag\") | .id" | head -n 1)
+          if [ -z "$id" ]; then
+            # Given a missing tag, GitHub would create it on the default branch.
+            gh api "repos/$GH_REPO/git/ref/tags/$tag" >/dev/null \
+              || { echo "tag $tag isn't on GitHub" >&2; exit 1; }
+            id=$(gh api -X POST "repos/$GH_REPO/releases" -f tag_name="$tag" -f name="Frame Control ${tag#v}" \
+              -F draft=true -f body="" --jq .id)
+          fi
+          [ -n "$id" ] || { echo "no release id for $tag; not PATCHing releases/" >&2; exit 1; }
+          gh api -X PATCH "repos/$GH_REPO/releases/$id" -f tag_name="$tag" --jq '"release " + (.id|tostring) + " tag_name " + .tag_name'
+
  build:
+    needs: draft
+    # Still runs for pull requests and manual runs, where the draft job is skipped.
+    if: ${{ !failure() && !cancelled() }}
    strategy:
      fail-fast: false
      matrix:
@@ -34,22 +75,71 @@ jobs:
      - uses: actions/setup-node@v4
        with:
          node-version: "24"
+      # PC in the headset (Experimental, Windows only): our capture adapter and
+      # the GStreamer libraries it uses, bundled into resources/desktop/bundle.
+      - uses: msys2/setup-msys2@v2
+        if: runner.os == 'Windows'
+        with:
+          msystem: UCRT64
+          update: true
+      # A mirror's HTTP 429 once failed this install (2026-10-24): retry, so a
+      # flaky mirror can't break a release build.
+      - name: MSYS2 packages for the PC host
+        if: runner.os == 'Windows'
+        shell: msys2 {0}
+        run: |
+          for attempt in 1 2 3; do
+            pacman -S --needed --noconfirm mingw-w64-ucrt-x86_64-gcc mingw-w64-ucrt-x86_64-pkgconf mingw-w64-ucrt-x86_64-python mingw-w64-ucrt-x86_64-gstreamer mingw-w64-ucrt-x86_64-gst-plugins-base mingw-w64-ucrt-x86_64-gst-plugins-good mingw-w64-ucrt-x86_64-gst-plugins-bad mingw-w64-ucrt-x86_64-gst-plugins-ugly && exit 0
+            sleep $((attempt * 30))
+          done
+          exit 1
+      - name: Windows PC streaming libraries
+        if: runner.os == 'Windows'
+        shell: msys2 {0}
+        run: python desktop/build.py --out app/build/desktop/win-x64
      - name: Build
        working-directory: app
        shell: bash
        run: npm ci && npm run ${{ matrix.script }}
        env:
          CSC_IDENTITY_AUTO_DISCOVERY: "false"
+      # The installer, not just the unpacked build: its 7-Zip payload once dropped
+      # the OpenXR layer's arm64 library without any error. Install it silently
+      # and compare every installed file with the unpacked build.
+      - name: Check the Windows installer installs every file
+        if: matrix.os == 'windows-latest'
+        shell: pwsh
+        run: |
+          $setup = Get-ChildItem app/dist/Frame-Control-Setup-*.exe | Select-Object -First 1
+          $p = Start-Process -FilePath $setup.FullName -ArgumentList '/S' -Wait -PassThru
+          if ($p.ExitCode -ne 0) { throw "installer exited $($p.ExitCode)" }
+          $installed = Join-Path $env:LOCALAPPDATA 'Programs\Frame Control'
+          for ($i = 0; $i -lt 60 -and -not (Test-Path (Join-Path $installed 'Uninstall Frame Control.exe')); $i++) { Start-Sleep 2 }
+          node app/build/check-resources.js (Join-Path $installed 'resources') app/dist/win-unpacked/resources
+          if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE }
+      # From the installed app, so this also proves the installer carries the
+      # libraries. After the check above, and without writing __pycache__,
+      # so it can't change what that check compares.
+      - name: Test the installed app's Python with the Windows PC libraries
+        if: runner.os == 'Windows'
+        shell: bash
+        env:
+          PYTHONDONTWRITEBYTECODE: "1"
+        run: |
+          resources="$(cygpath -u "$LOCALAPPDATA")/Programs/Frame Control/resources"
+          python="$resources/python/python.exe"
+          native="$(cygpath -w "$resources/desktop/bundle")"
+          FRAME_PC_NATIVE="$native" FRAME_PC_REQUIRE_NATIVE=1 FRAME_PC_INSTALLER_BUNDLE=1 "$python" -B -m unittest discover -s tests -p 'test_pc*.py' -v
      - name: Upload to the release
        if: startsWith(github.ref, 'refs/tags/')
        shell: bash
        env:
          GH_TOKEN: ${{ github.token }}
        run: |
-          tag="${GITHUB_REF_NAME}"
-          gh release view "$tag" >/dev/null 2>&1 || gh release create "$tag" --draft --title "Frame Control ${tag#v}" --notes ""
-          gh release upload "$tag" ${{ matrix.files }} --clobber
+          gh release upload "${GITHUB_REF_NAME}" ${{ matrix.files }} --clobber
+      # Tagged builds are attached to the release above; only other builds need artifacts.
      - uses: actions/upload-artifact@v4
+        if: ${{ !startsWith(github.ref, 'refs/tags/') }}
        with:
          name: frame-control-${{ matrix.os }}
          path: |
@@ -1,7 +1,21 @@
 .DS_Store
 __pycache__/
 apk-catalog/data/cache/
-apk-catalog/data/index-v2.json*
+apk-catalog/data/index-v2*.json*
 compat-db/.env.lakebed.server
 compat-db/.lakebed/
 tests/smoke/results/
+mac/bin/
+
+# Shared desktop streaming controller build products
+desktop/controller.dll
+desktop/controller.dylib
+desktop/controller.so
+desktop/*.obj
+desktop/*.lib
+desktop/*.exp
+desktop/bundle/
+app/build/desktop/
+
+# Downloaded at build time (frame/kdeconnect/fetch.py, app/build/fetch-deps.js)
+frame/kdeconnect/packages/
@@ -0,0 +1,16 @@
+# Changelog
+
+## Unreleased
+
+- Add `scripts/frame-doctor.sh` for read-only setup and pairing diagnostics,
+  with verified SSH identity, private summaries and actionable next steps
+  (issue #34). Connection steps links to the diagnostic guide. Repairs remain
+  manual; restart-loop evidence stops deeper diagnostics. The real Frame
+  verified the SSH connection and safety stop; deeper checks were skipped.
+- Fixed headset smoke cleanup reporting success while a smoke-title process
+  could remain running after its files were removed. Cleanup now stops and
+  rechecks processes tied to exact reserved test-title paths, uses PID handles
+  for safe TERM/KILL escalation, and records stopped PIDs. Unrelated titles and
+  shared Wine services are left alone. Verified with local Linux process
+  regressions and a 13/13 real Frame smoke run that stopped five test-owned
+  Proton process PIDs before removing the Windows test title.
@@ -35,8 +35,8 @@ See what the headset sees, install games and Android apps, move files and text a
 <tr>
 <td width="50%" valign="top">

-**👓 Headset view**<br>
-Live video of what the lenses show (about 30 fps), or a still of both eyes. Zoom, pan, full screen, save as PNG.
+**👓 Headset view and Desktop**<br>
+Live video of what the lenses show, or of the app panel in use, flat and still however the wearer looks around. Turn on Control and tap or click right on it to use the Frame from your phone or computer.

 </td>
 <td width="50%" valign="top">
@@ -50,7 +50,7 @@ Charge, charging watts and time left, storage, memory, temperature, Wi-Fi, and w
 <td valign="top">

 **🎮 Steam games**<br>
-Everything you own with its Steam Frame rating. Install onto the headset with live progress, and search the store.
+Everything you own with its Steam Frame rating. Install onto the headset with live progress, search the store, and force quit a game that stops responding.

 </td>
 <td valign="top">
@@ -63,14 +63,17 @@ About 4,500 F-Droid apps rated for the Frame. One click installs each as its own
 <tr>
 <td valign="top">

-**📁 Files, games and clipboard**<br>
-Drag files onto the window to send them. Drop a game's .zip, folder or .exe to add it to the Steam library, with Proton or the Linux runtime picked for you. Send text or your clipboard straight to the headset's desktop.
+**📁 Files and games**<br>
+Drag files onto the window to send them. Drop a game's .zip, folder or .exe to add it to the Steam library, with Proton or the Linux runtime picked for you.

 </td>
 <td valign="top">

-**📸 Screenshots**<br>
-Browse the shots you take in the headset and save them to your Pictures folder.
+**📸 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.

 </td>
 </tr>
@@ -86,13 +89,27 @@ Install desktop apps like Moonlight or VLC, and set each Android app's resolutio
 **⚡ One-click tools**<br>
 SSH, SFTP, Steam Link, remote desktop, volume, sleep, restart and shut down.

+</td>
+</tr>
+<tr>
+<td colspan="2" valign="top">
+
+**💾 Backup and recovery**<br>
+Copy game saves, settings and Android app data to your computer, and restore them after a factory reset. Sign-in files and keys are never copied. A Return to Game Mode button helps when the headset is stuck in Desktop Mode. Not tested on a headset yet. [How it works](docs/backup-and-recovery.md).
+
 </td>
 </tr>
 </table>

-Nothing is installed on the Frame for any of this: the app uses what SteamOS
+The optional [Family and comfort](docs/family-comfort.md) card adds session
+limits, breaks, local alerts and one-click casting. A session copies a small
+Frame Control worker into your headset user account.
+
+For the other features, nothing is installed on the Frame: the app uses what SteamOS
 already ships (sideloading a game copies Valve's own devkit scripts to
-`~/devkit-utils`, as Valve's Devkit Client does). [How each feature works](docs/frame-control.md).
+`~/devkit-utils`, as Valve's Devkit Client does). The optional
+[performance HUD](docs/vr-utilities.md) copies our own Python helpers into
+`~/.local/share/frame-control/vr/`. [How each feature works](docs/frame-control.md).

 ## Install

@@ -107,6 +124,10 @@ already ships (sideloading a game copies Valve's own devkit scripts to
 a computer. Build it from [`ios/`](ios) in Xcode; see [docs/iphone.md](docs/iphone.md).

 The app brings its own Python and `adb`; SSH is built into macOS and Windows.
+From 0.4 it updates itself: when a new version is published, a banner offers
+**Update and restart**. It sends anonymous usage statistics, which you can turn
+off. Sharing compatibility results and error details is opt-in. See
+[docs/privacy.md](docs/privacy.md).
 Google doesn't publish `adb` for arm64 Linux, so that build uses your
 distribution's. If you already have `adb`, the app uses yours.

@@ -175,9 +196,11 @@ entry to `~/.ssh/config` and keys at `~/.ssh/id_ed25519_frame` and
 ## Feedback

 This is a first public test, so reports are really useful, especially from
-Windows and Linux. The quickest way is the
-[feedback form](https://frame-control.pages.dev/feedback/): no GitHub account
-needed, and it opens an issue here. Please include:
+Windows and Linux. The quickest way is **Report a problem** in the app (the
+speech-bubble button at the top, or **Help → Report a Problem…**). It adds
+diagnostics with personal details removed, shows you exactly what's included,
+and sends it privately to the maintainer; nothing is published. Without the app,
+use the [feedback form](https://frame-control.pages.dev/feedback/). Please include:

 - what you tried and what happened
 - your computer's OS and your SteamOS build (Steam Settings → System)
@@ -198,13 +221,19 @@ Frame's software fits together, all checked against a real headset and labelled
 | [Scripts and headset setup](docs/scripts.md) | The command-line helpers, minimum typing, streaming options, floating panels |
 | [How the Frame works](docs/how-the-frame-works.md) | SteamVR → gamescope → Plasma, verified facts, debugging |
 | [Android apps (Lepton)](docs/apks.md) | Sideloading, the rated F-Droid catalogue, per-app instances |
+| [Import from my Quest](docs/quest-import.md) | Copy games you own off your Quest by USB; what runs, and why licence-checked games may not |
+| [My Meta library](docs/meta-library.md) | Experimental: install and test Quest games your Meta account owns, with your own token |
 | [Sideloading Linux and Windows games](docs/sideloading.md) | A .zip, folder or .exe as a Steam Devkit Game, runtime detection |
 | [Install links for websites](docs/web-install.md) | `frame-control://install` links and manifests, the rules, a button to paste |
-| [Steam games](docs/steam-games.md) · [VR video](docs/vr-video.md) · [WebXR in Chromium](docs/webxr-chromium.md) | Installing and buying, watching VR180/360, the Chromium build |
+| [VR comfort and HUD](docs/vr-utilities.md) · [Steam games](docs/steam-games.md) · [VR video](docs/vr-video.md) · [WebXR in Chromium](docs/webxr-chromium.md) | Installing and buying, watching VR180/360, the Chromium build |
+| [PC in the headset](docs/pc-in-headset.md) | Experimental, Windows only, off until you turn it on: your PC's windows and screens as panels in the Frame. Not yet tried on a real Windows desktop or headset |
+| [Mac in the headset](docs/mac-in-headset.md) | Mac windows and screens as panels in the Frame, with laser and keyboard input |
+| [VR mods and custom songs](docs/mods.md) | UEVR for Unreal games from Frame Control (install, play in VR, remove), the per-game table, Beat Saber blockers |
 | [SSH](docs/ssh.md) · [Streaming](docs/streaming.md) · [Files](docs/file-transfer.md) · [Panels](docs/panels.md) · [Tailscale](docs/tailscale.md) | Topic notes |
 | [Frame Control for iPhone](docs/iphone.md) | The iPhone and iPad app, how it runs the server on the Frame, pairing |
 | [Recovery and OS images](docs/recovery-and-images.md) | Where to download the Frame's OS, what's inside, testing without the headset |
-| [Testing](docs/testing.md) | Unit tests, end-to-end tests against a fake Frame in Docker, and the headset smoke test |
+| [AI agents and assistant](docs/agents.md) | Key-free MCP tools, human approvals, and an opt-in assistant panel |
+| [Testing](docs/testing.md) | Unit tests, end-to-end tests against a fake Frame in Docker, the headset smoke test and a Windows test VM |
 | [Open questions](docs/open-questions.md) | What's still unchecked |

 <details>
@@ -237,9 +266,12 @@ cd app && npm install && npm start      # run the app from the checkout

 The server is Python stdlib only; the app is Electron. GitHub Actions runs the
 tests on macOS, Windows and Linux, and a `v*` tag builds all three installers
-into the release. See [building](docs/frame-control.md#building).
+into a draft release, which reaches users once published. See
+[building](docs/frame-control.md#building) and [releasing](docs/releasing.md).

 ## License

-[MIT](LICENSE). Steam, Steam Frame and SteamVR are trademarks of Valve
+[MIT](LICENSE). The apps also ship other people's software under its own
+licence, notably KDE Connect (GPL) for the keyboard and trackpad; see
+[THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md). Steam, Steam Frame and SteamVR are trademarks of Valve
 Corporation. This project isn't affiliated with or endorsed by Valve.
@@ -0,0 +1,17 @@
+# Third-party software in Frame Control
+
+Frame Control's own code is under the [MIT licence](LICENSE). The apps also
+ship other people's software, unchanged, each under its own licence:
+
+| What | Where | Licence | Details |
+|---|---|---|---|
+| KDE Connect 24.02.2 and five libraries, as Valve builds them for the Frame | Desktop apps and the iPhone app; copied to the Frame for the keyboard and trackpad | GPL and LGPL (per package) | [frame/kdeconnect/NOTICE.md](frame/kdeconnect/NOTICE.md), licence texts in [frame/kdeconnect/LICENSES](frame/kdeconnect/LICENSES), complete source in the [kdeconnect-frame-24.02.2-1 release](https://github.com/saphid/frame-control/releases/tag/kdeconnect-frame-24.02.2-1) |
+| Python 3.12 ([python-build-standalone](https://github.com/astral-sh/python-build-standalone)) | Desktop apps | PSF License and others | Included with it, in the app's `python` folder |
+| adb (Android SDK Platform-Tools) | Desktop apps (not Linux arm64) | Apache-2.0 and others | `NOTICE.txt` in the app's `tools` folder |
+| Mozilla's CA certificate list, as published by curl | Desktop apps | MPL-2.0 | https://curl.se/docs/caextract.html |
+| Electron | Desktop apps | MIT (Chromium: BSD-3-Clause and others) | `LICENSE` and `LICENSES.chromium.html` in the app |
+| GStreamer 1.x, a few of its plugins and the libraries they need, as [MSYS2](https://packages.msys2.org/) builds them | Windows app only, for the experimental [PC in the headset](docs/pc-in-headset.md) | LGPL-2.1-or-later and others (per package); the GPL x264 encoder is left out | `libraries.json` (exact package versions) and the `licenses` folder in the app's `desktop/bundle` folder |
+
+Frame Control starts KDE Connect and talks to it over its network protocol;
+it doesn't link to it or include its code. On Windows, Frame Control's own
+`pc-host.dll` links to GStreamer dynamically; the libraries are unmodified.
@@ -6,11 +6,25 @@ A report: package, version, result (runs | crashes | install_failed |
 instance_failed, from an automated test), rating (works | issues | broken, from
 a person), notes, via (harness | probe | user), date, steamos, lepton, runtime.
 Newest wins, and a person's rating beats an automated result.
+
+An install_failed report saying the file had no arm64-v8a build (or needed a
+newer Android) is about that one APK file, often the wrong per-ABI download of
+an app that has an arm64 build, so it says nothing about the app and is left out.
 """
+import re
+
+FILE_FAULT = re.compile(r'no arm64-v8a build|needs Android API')
+
+
+def about_app(r):
+    """False for reports about one wrong APK file rather than the app itself."""
+    return not (r.get('result') == 'install_failed' and not r.get('rating')
+                and FILE_FAULT.search(r.get('notes') or ''))


 def verdict(reports):
    """(verdict, summary lines) for one package's reports, or None."""
+    reports = [r for r in reports or () if about_app(r)]
    if not reports:
        return None
    rs = sorted(reports, key=lambda r: r.get('date') or '')
@@ -0,0 +1,79 @@
+// Checks that the app's resources hold what Frame Control can't work without.
+// The OpenXR compatibility layer is why this exists: VR APK installs need its
+// arm64-v8a library (ui/frame_android.py XR_COMPAT_FILES).
+//
+// As electron-builder's afterPack hook it checks the unpacked app. From the
+// command line it checks an installed copy, which is what caught the Windows
+// installer silently dropping the library (see .github/workflows/release.yml):
+//   node build/check-resources.js INSTALLED_RESOURCES [REFERENCE_RESOURCES]
+// With a reference (the unpacked build's resources), every file in it must also
+// be in the installed copy, at the same size.
+const fs = require("fs");
+const path = require("path");
+
+const REQUIRED = [
+  "ui/server.py",
+  "ui/frame_android.py",
+  "frame/android/lepton-app.sh",
+  "frame/openxr-compat/XrApiLayer_FRAME_compat.json",
+  "frame/openxr-compat/prebuilt/arm64-v8a/libXrApiLayer_FRAME_compat.so.gz",
+];
+
+function resourcesDir(context) {
+  if (context.electronPlatformName === "darwin" || context.electronPlatformName === "mas") {
+    const app = `${context.packager.appInfo.productFilename}.app`;
+    return path.join(context.appOutDir, app, "Contents", "Resources");
+  }
+  return path.join(context.appOutDir, "resources");
+}
+
+function size(file) {
+  try {
+    return fs.statSync(file).size;
+  } catch {
+    return -1;
+  }
+}
+
+// Required files that are missing or empty.
+function missing(dir) {
+  return REQUIRED.filter((rel) => size(path.join(dir, rel)) <= 0);
+}
+
+// Files under reference that aren't in dir at the same size.
+function differences(dir, reference) {
+  const out = [];
+  (function walk(rel) {
+    for (const e of fs.readdirSync(path.join(reference, rel), { withFileTypes: true })) {
+      const r = path.join(rel, e.name);
+      if (e.isDirectory()) walk(r);
+      else if (e.isFile() && size(path.join(dir, r)) !== size(path.join(reference, r))) out.push(r.split(path.sep).join("/"));
+    }
+  })("");
+  return out;
+}
+
+exports.default = async function afterPack(context) {
+  const dir = resourcesDir(context);
+  const gone = missing(dir);
+  if (gone.length) {
+    throw new Error(`packaged app is missing required resources in ${dir}: ${gone.join(", ")}`);
+  }
+};
+exports.missing = missing;
+exports.differences = differences;
+exports.REQUIRED = REQUIRED;
+
+if (require.main === module) {
+  const [dir, reference] = process.argv.slice(2);
+  if (!dir) {
+    console.error("usage: node build/check-resources.js INSTALLED_RESOURCES [REFERENCE_RESOURCES]");
+    process.exit(2);
+  }
+  const problems = [...missing(dir), ...(reference ? differences(dir, reference) : [])];
+  if (problems.length) {
+    console.error(`${dir} is missing or has the wrong size for:\n  ${[...new Set(problems)].join("\n  ")}`);
+    process.exit(1);
+  }
+  console.log(`${dir}: all required resources present${reference ? ", and every file of " + reference : ""}`);
+}
@@ -1,7 +1,9 @@
 // Downloads what the app bundles so users install nothing else: a standalone
 // Python (python-build-standalone), adb (Android platform-tools) and a CA
 // bundle. Each goes in build/deps/<os>-<arch>/{python,tools}, which package.json
-// copies into the app's resources. Everything is pinned by version and SHA-256.
+// copies into the app's resources. Also KDE Connect for the Frame (the same arm64
+// packages for every build), into ../frame/kdeconnect/packages as listed in
+// ../frame/kdeconnect/packages.json. Everything is pinned by version and SHA-256.
 //   node build/fetch-deps.js mac arm64 | win x64 | linux x64 arm64
 const crypto = require("crypto");
 const fs = require("fs");
@@ -127,8 +129,26 @@ async function fetch(os, arch) {
  console.log(`${key}: ${version} -> ${out}`);
 }

+// As frame/kdeconnect/fetch.py does for the iPhone app's bundle.
+async function fetchKdeConnect() {
+  const dir = path.join(__dirname, "..", "..", "frame", "kdeconnect");
+  const manifest = JSON.parse(fs.readFileSync(path.join(dir, "packages.json"), "utf8"));
+  const out = path.join(dir, "packages");
+  fs.mkdirSync(out, { recursive: true });
+  const sha = (file) => crypto.createHash("sha256").update(fs.readFileSync(file)).digest("hex");
+  for (const p of manifest.packages) {
+    const file = path.join(out, p.file);
+    if (fs.existsSync(file) && sha(file) === p.sha256) continue;
+    await download(manifest.release + p.file, p.sha256, file);
+  }
+  const wanted = new Set(manifest.packages.map((p) => p.file));
+  for (const f of fs.readdirSync(out)) if (!wanted.has(f)) fs.rmSync(path.join(out, f), { recursive: true, force: true });
+  console.log(`KDE Connect for the Frame: ${manifest.packages.length} packages -> ${out}`);
+}
+
 (async () => {
  const [os, ...archs] = process.argv.slice(2);
  if (!os || !archs.length) throw new Error("usage: node build/fetch-deps.js <mac|win|linux> <arch>...");
  for (const arch of archs) await fetch(os, arch);
+  await fetchKdeConnect();
 })().catch((e) => { console.error(e.message); process.exit(1); });
@@ -1,7 +1,7 @@
 // Frame Control as a desktop app (macOS, Windows, Linux): starts ui/server.py on
 // a free loopback port and shows it in a native window. The server does all the
 // work over the `frame` SSH alias; this file only hosts it.
-const { app, BrowserWindow, Menu, clipboard, dialog, ipcMain, shell } = require("electron");
+const { app, BrowserWindow, Menu, Notification, clipboard, dialog, ipcMain, nativeImage, session, shell } = require("electron");
 const { execFile, spawn } = require("child_process");
 const { promisify } = require("util");
 const fs = require("fs");
@@ -10,6 +10,9 @@ const net = require("net");
 const os = require("os");
 const path = require("path");
 const { SCHEME, parseInstallLink, linkFromArgv } = require("./install-link");
+const updater = require("./updater");
+const { preferenceStore } = require("./preferences");
+const { createMetaSignIn } = require("./meta-signin");

 const run = promisify(execFile);

@@ -26,6 +29,7 @@ const LOG_DIR = IS_MAC ? path.join(os.homedir(), "Library", "Logs", "Frame Contr
 const LOG = path.join(LOG_DIR, "server.log");
 const BG = "#0d1117";
 const FRAME = process.env.FRAME_ALIAS || "frame";
+const preferences = preferenceStore(path.join(app.getPath("userData"), "client-preferences.json"));

 let server = null;
 let url = null;
@@ -114,7 +118,12 @@ function ping(target) {
 }

 async function startServer() {
+  // The version, the time zone and whether this is a built app go to ui/frame_telemetry.py,
+  // which sends nothing from a source checkout.
  const env = { ...process.env, PATH: await loginPath(), FRAME_CONTROL_APP: "1",
+                FRAME_CONTROL_VERSION: app.getVersion(), FRAME_CONTROL_LOG: LOG,
+                FRAME_CONTROL_TIMEZONE: Intl.DateTimeFormat().resolvedOptions().timeZone || "",
+                ...(app.isPackaged ? { FRAME_CONTROL_PACKAGED: "1" } : {}),
                ...(fs.existsSync(TOOLS) ? { FRAME_CONTROL_TOOLS: TOOLS } : {}) };
  python = await findPython(env);
  if (!python) throw new Error(`Frame Control needs Python 3.8 or later. ${PYTHON_HELP}`);
@@ -144,7 +153,7 @@ async function startServer() {
  const target = `http://127.0.0.1:${port}/`;
  for (let i = 0; i < 100; i++) {
    if (exited !== null) throw new Error(`The server exited (${exited}). See ${LOG}.`);
-    if (await ping(target)) { url = target; return; }
+    if (await ping(target)) { url = target; serverStarted = Date.now(); return; }
    await new Promise((r) => setTimeout(r, 100));
  }
  if (server === child) server = null;
@@ -154,51 +163,76 @@ async function startServer() {

 // Closing stdin lets server.py close its SSH connections and exit (the only clean
 // way on Windows); SIGTERM does the same elsewhere.
+// Resolves once it has exited (or after 20 s), so a replacement can take the server
+// lock: server.py allows one per user.
 function endServer(child) {
+  const gone = child.exitCode !== null || child.signalCode !== null ? Promise.resolve()
+    : new Promise((resolve) => child.once("exit", resolve));
  try { child.stdin.end(); } catch {}
  if (!IS_WIN) child.kill("SIGTERM");
-  setTimeout(() => { if (child.exitCode === null && child.signalCode === null) child.kill(); }, 5000).unref();
+  // server.py ignores a second SIGTERM while it shuts down, so the fallback is a hard kill.
+  setTimeout(() => { if (child.exitCode === null && child.signalCode === null) child.kill("SIGKILL"); }, 12000).unref();
+  return Promise.race([gone, new Promise((resolve) => setTimeout(resolve, 20000).unref())]);
 }

 function stopServer() {
  if (server) endServer(server);
 }

-function errorPage(message) {
+function errorPage(message, title = "Frame Control couldn't start") {
  const esc = (s) => s.replace(/[&<>]/g, (c) => ({ "&": "&amp;", "<": "&lt;", ">": "&gt;" }[c]));
  const html = `<!doctype html><meta charset="utf-8"><body style="margin:0;height:100vh;display:grid;
    place-items:center;background:${BG};color:#e6edf3;font:14px -apple-system,sans-serif">
-    <div style="max-width:560px;padding:32px;line-height:1.5"><h2>Frame Control couldn't start</h2>
-    <p>${esc(message)}</p><p style="color:#8b98a8">Fix it, then choose Frame → Restart Server.</p></div>`;
+    <div style="max-width:560px;padding:32px;line-height:1.5"><h2>${esc(title)}</h2>
+    <p>${esc(message)}</p>
+    <p><button onclick="this.disabled = true; frameApp.restartServer()" style="font:inherit;padding:6px 16px;
+      border-radius:6px;border:1px solid #30363d;background:#21262d;color:inherit;cursor:pointer">Try Again</button></p>
+    <p style="color:#8b98a8">Frame → Restart Server does the same.</p></div>`;
  return "data:text/html;charset=utf-8," + encodeURIComponent(html);
 }

+// A server that had been running starts again by itself (something stopped it: a
+// signal, a crash). One that stops again within a minute shows the error instead,
+// so a server that can't stay up doesn't restart forever.
+let serverStarted = 0;
 function serverDied(why) {
  url = null;
-  if (win) win.loadURL(errorPage(`The server stopped unexpectedly (${why}). See ${LOG}.`));
+  if (!win) return;
+  if (Date.now() - serverStarted > 60000) restartServer();
+  else win.loadURL(errorPage(`Its server stopped unexpectedly (${why}). See ${LOG}.`, "Frame Control stopped"));
 }

-async function restartServer() {
-  const old = server;
-  server = null;
-  url = null;
-  if (old) endServer(old);
-  await load();
+// Restarts that overlap share one: two could each start a server, and the one
+// that lost the lock would leave the app pointing at nothing.
+let restarting = null;
+function restartServer() {
+  if (!restarting) {
+    restarting = (async () => {
+      const old = server;
+      server = null;
+      url = null;
+      if (old) await endServer(old);
+      if (starting) await starting.catch(() => {});  // a start it cut short: then start afresh
+      await load();
+    })().finally(() => { restarting = null; });
+  }
+  return restarting;
 }

 // On macOS the page's sticky header becomes the title bar, clear of the traffic lights.
 const CHROME_CSS = IS_MAC && `
  header { padding-left: 92px !important; -webkit-app-region: drag; user-select: none; }
-  header a, header button, header input, header .chip { -webkit-app-region: no-drag; }
+  header a, header button, header input, header select, header .chip { -webkit-app-region: no-drag; }
 `;

 // Restart Server can start a new load while an older one is still waiting for
 // its server; only the newest load may touch the window.
 let loadGen = 0;
+let starting = null;  // loads that overlap share one server start
 async function load() {
  const gen = ++loadGen;
  try {
-    if (!url) await startServer();
+    if (!url) await (starting ||= startServer().finally(() => { starting = null; }));
    if (gen === loadGen && win) { await win.loadURL(url); firstRunCheck(); }
  } catch (e) {
    if (gen === loadGen && win) await win.loadURL(errorPage(e.message));
@@ -242,8 +276,129 @@ function fromUi(e) {
  } catch { return false; }
 }

+// The error page's Try Again button. The error page is the only data: page the window
+// shows (`url` can still be set then: the server answered but the page failed to load).
+ipcMain.handle("server:restart", (e) => {
+  if (win && e.sender === win.webContents && e.senderFrame && e.senderFrame.url.startsWith("data:")) restartServer();
+});
 ipcMain.handle("clipboard:read", (e) => fromUi(e) ? clipboard.readText() : "");
+// Text from the headset's clipboard onto this computer's.
+ipcMain.handle("clipboard:write", (e, text) => {
+  if (!fromUi(e) || typeof text !== "string") return false;
+  clipboard.writeText(text);
+  return true;
+});
+ipcMain.handle("preferences:read", (e) => fromUi(e) ? preferences.read() : {});
+ipcMain.handle("preferences:write", (e, key, value) => fromUi(e) ? preferences.write(key, value) : false);
+// A PNG or JPEG (a screenshot) onto the clipboard as an image.
+ipcMain.handle("clipboard:writeImage", (e, bytes) => {
+  if (!fromUi(e) || !(bytes instanceof Uint8Array)) return false;
+  const img = nativeImage.createFromBuffer(Buffer.from(bytes));
+  if (img.isEmpty()) throw new Error("not an image");
+  clipboard.writeImage(img);
+  return true;
+});
 ipcMain.handle("connection:setup", (e) => { if (fromUi(e)) setUpConnection(); });
+ipcMain.on("keys:capture", (e, on) => { if (fromUi(e)) win.webContents.setIgnoreMenuShortcuts(on === true); });
+ipcMain.handle("update:get", (e) => fromUi(e) ? publicUpdate() : null);
+ipcMain.handle("update:check", (e) => fromUi(e) ? checkForUpdate({ manual: true }).then(publicUpdate) : null);
+ipcMain.handle("update:install", (e) => { if (fromUi(e)) installUpdate(); });
+
+// The page reports the headsets it knows (the server's Devices tab), so the Frame
+// menu can switch between them. Only plain names and ids go into the menu.
+const ALIAS_RE = /^[A-Za-z0-9][A-Za-z0-9._-]{0,63}$/;
+let devices = [];
+ipcMain.on("devices:changed", (e, list) => {
+  if (!fromUi(e) || !Array.isArray(list)) return;
+  const next = list.slice(0, 20).filter(d => d && typeof d.id === "string" && ALIAS_RE.test(d.alias || ""))
+    .map(d => ({ id: d.id.slice(0, 80), name: String(d.name || d.alias).slice(0, 60), alias: d.alias, active: !!d.active }));
+  if (JSON.stringify(next) === JSON.stringify(devices)) return;
+  devices = next;
+  buildMenu();
+});
+const activeAlias = () => (devices.find(d => d.active) || {}).alias || FRAME;
+
+// SSH through the server, so it goes to the headset and address the app is using
+// (with its own pinned identity), and refuses when there's none.
+function openSsh() {
+  if (!url) return dialog.showErrorBox("Couldn't open SSH", "Frame Control's server isn't running.");
+  const body = JSON.stringify({ what: "terminal" });
+  const req = http.request(new URL("/api/open", url), {
+    method: "POST", timeout: 15000,
+    headers: { "Content-Type": "application/json", "X-Frame-UI": "1", "Content-Length": Buffer.byteLength(body) },
+  }, (res) => {
+    let data = "";
+    res.on("data", (c) => { data += c; });
+    res.on("end", () => {
+      if (res.statusCode === 200) return;
+      let why = `HTTP ${res.statusCode}`;
+      try { why = JSON.parse(data).error || why; } catch {}
+      dialog.showErrorBox("Couldn't open SSH", why);
+    });
+  });
+  req.on("error", (e) => dialog.showErrorBox("Couldn't open SSH", e.message));
+  req.on("timeout", () => req.destroy(new Error("the server didn't answer")));
+  req.end(body);
+}
+function showDevices() {
+  if (win && url) win.webContents.executeJavaScript('location.hash = "devices"').catch(() => {});
+}
+
+ipcMain.handle("comfort:notify", (e, message) => {
+  if (!fromUi(e) || typeof message !== "string" || message.length > 500) throw new Error("Invalid notification");
+  if (!Notification.isSupported()) throw new Error("System notifications are unavailable");
+  return new Promise((resolve, reject) => {
+    const notification = new Notification({title: "Frame Control", body: message});
+    const timer = setTimeout(() => reject(new Error("Notification delivery was not confirmed. Check system notification settings.")), 5000);
+    notification.once("show", () => { clearTimeout(timer); resolve(true); });
+    notification.once("failed", (_event, error) => {
+      clearTimeout(timer);
+      reject(new Error("Notification delivery failed. Check system notification settings: " + error));
+    });
+    notification.show();
+  });
+});
+
+// ---- Sign in with Meta (app/meta-signin.js, docs/meta-library.md) ----
+// The token goes from the sign-in window's own session straight to the server's
+// connect path; the page asks for it and only hears the server's answer.
+function postApi(apiPath, body) {
+  return new Promise((resolve, reject) => {
+    if (!url) return reject(new Error("Frame Control's server isn't running."));
+    const data = JSON.stringify(body);
+    const req = http.request(new URL(apiPath, url), {
+      method: "POST", timeout: 60000,
+      headers: { "Content-Type": "application/json", "X-Frame-UI": process.env.FRAME_UI_KEY || "1",
+                 "Content-Length": Buffer.byteLength(data) },
+    }, (res) => {
+      let text = "";
+      res.setEncoding("utf8");
+      res.on("data", (c) => { text += c; });
+      res.on("end", () => {
+        let parsed = {};
+        try { parsed = JSON.parse(text); } catch {}
+        resolve({ status: res.statusCode, data: parsed });
+      });
+    });
+    req.on("error", reject);
+    req.on("timeout", () => req.destroy(new Error("the server didn't answer")));
+    req.end(data);
+  });
+}
+
+let metaSignIn = null;
+const meta = () => (metaSignIn ||= createMetaSignIn({
+  BrowserWindow, session, shell, parent: () => win,
+  // The page only offers sign-in after its "I understand" box is ticked (checked again below).
+  connect: (token, attempt) => postApi("/api/meta", { action: "connect", token, acknowledged: true, attempt }),
+  cancel: (attempt) => postApi("/api/meta", { action: "cancel-signin", attempt }),
+}));
+ipcMain.handle("meta:signin", (e, acknowledged) => {
+  if (!fromUi(e)) return null;
+  if (acknowledged !== true) return { ok: false, error: "My Meta library is experimental: tick that you understand the warning first." };
+  return meta().signIn();
+});
+ipcMain.handle("meta:signout", (e) => fromUi(e) ? meta().signOut() : false);

 // frame-control://install links from websites (docs/web-install.md). They can
 // arrive before the window or server exists (macOS open-url on a cold launch),
@@ -276,6 +431,81 @@ ipcMain.on("install-link:ready", (e) => {
  deliverLinks();
 });

+// ---- updates (app/updater.js, docs/releasing.md) ----
+// Checked shortly after launch and every few hours; the page shows a banner and
+// the Update button calls installUpdate.
+const UPDATE_EVERY = 6 * 3600 * 1000;
+const update = { status: "idle", current: app.getVersion(), latest: null, error: null, progress: 0, how: null };
+
+function publicUpdate() {
+  const r = update.latest;
+  return { status: update.status, current: update.current, error: update.error, progress: update.progress,
+           latest: r && { version: r.version, notes: r.notes, page: r.page },
+           canInstall: !!update.how && update.how.method !== "manual", why: update.how && update.how.why };
+}
+
+function setUpdate(fields) {
+  Object.assign(update, fields);
+  if (win && linkPage === win.webContents) win.webContents.send("update:state", publicUpdate());
+}
+
+async function checkForUpdate({ manual = false } = {}) {
+  if (["checking", "downloading", "ready"].includes(update.status)) return;
+  setUpdate({ status: "checking", error: null });
+  try {
+    const latest = await updater.latestRelease();
+    const how = updater.updateMethod({ platform: process.platform, isPackaged: app.isPackaged,
+                                       execPath: process.execPath, env: process.env,
+                                       exists: fs.existsSync, writable: updater.writable });
+    if (updater.isNewer(latest.version, update.current)) {
+      setUpdate({ status: "available", latest, how });
+      if (manual) offerUpdateDialog();
+    } else {
+      setUpdate({ status: "none", latest, how });
+      if (manual) dialog.showMessageBox(win, { type: "info", message: "Frame Control is up to date",
+                                               detail: `You have ${update.current}, the newest version.` });
+    }
+  } catch (e) {
+    // A failed check: nothing to install, and never an older release kept from before.
+    setUpdate({ status: "check-failed", error: e.message, latest: null });
+    if (manual) dialog.showMessageBox(win, { type: "warning", message: "Couldn't check for updates", detail: e.message });
+  }
+}
+
+async function offerUpdateDialog() {
+  const r = update.latest;
+  const { response } = await dialog.showMessageBox(win, {
+    type: "info", message: `Frame Control ${r.version} is available`,
+    detail: `You have ${update.current}.` + (update.how.method === "manual" ? ` Download it from the release page (${update.how.why}).` : ""),
+    buttons: [update.how.method === "manual" ? "Open Release Page" : "Update and Restart", "Later"], defaultId: 0, cancelId: 1,
+  });
+  if (response === 0) installUpdate();
+}
+
+async function installUpdate() {
+  // "error" here only ever means an install failed, so trying again is safe.
+  if (update.status !== "available" && update.status !== "error") return;
+  if (!update.latest || !updater.isNewer(update.latest.version, update.current)) return;
+  if (!update.how || update.how.method === "manual") { shell.openExternal(update.latest.page); return; }
+  setUpdate({ status: "downloading", progress: 0, error: null });
+  try {
+    const start = await updater.prepare(update.latest, update.how,
+      (done, total) => { if (total) setUpdate({ progress: done / total }); }, update.current);
+    setUpdate({ status: "ready", progress: 1 });
+    start();
+    quitting = true;
+    app.quit();
+  } catch (e) {
+    setUpdate({ status: "error", error: e.message });
+  }
+}
+
+function scheduleUpdateChecks() {
+  if (process.env.FRAME_CONTROL_NO_UPDATE_CHECK === "1") return;
+  setTimeout(checkForUpdate, 8000);
+  setInterval(checkForUpdate, UPDATE_EVERY).unref();
+}
+
 function registerScheme() {
  // A checkout runs as `electron .`, so the OS must be told the script too.
  // (macOS takes the scheme from Info.plist, which only the built app has.)
@@ -292,7 +522,7 @@ function createWindow() {
    title: "Frame Control", backgroundColor: BG, show: false,
    ...(IS_MAC ? { titleBarStyle: "hiddenInset", trafficLightPosition: { x: 18, y: 26 } }
               : { icon: path.join(__dirname, "build", "icon.png") }),
-    webPreferences: { contextIsolation: true, nodeIntegration: false, sandbox: true,
+    webPreferences: { contextIsolation: true, nodeIntegration: false, sandbox: true, backgroundThrottling: false,
                      preload: path.join(__dirname, "preload.js") },
  });
  win.once("ready-to-show", () => win.show());
@@ -326,25 +556,37 @@ async function runInTerminal(argv) {
  }
 }

-async function setUpConnection() {
-  const alias = `FRAME_ALIAS=${FRAME}`;
+// Set Up Connection for the headset in use (or another alias, from the Devices tab).
+async function setUpConnection(name = activeAlias()) {
+  if (!ALIAS_RE.test(name)) return;
+  const alias = `FRAME_ALIAS=${name}`;
  if (IS_MAC) return runInTerminal(["env", alias, "zsh", path.join(SCRIPTS, "connect.sh")]);
  const py = python || await findPython({ ...process.env, PATH: await loginPath() });
-  const setup = [py || "python3", ...PY_FLAGS, path.join(ROOT, "ui", "frame_connect.py")];
-  // A new console inherits our environment on Windows; Linux terminals may not.
-  runInTerminal(IS_WIN ? setup : ["env", alias, ...setup]);
+  // --alias, since a new console on Windows (and some Linux terminals) doesn't get our environment.
+  runInTerminal([py || "python3", ...PY_FLAGS, path.join(ROOT, "ui", "frame_connect.py"), "--alias", name]);
 }

 function buildMenu() {
  const template = [
-    ...(IS_MAC ? [{ role: "appMenu" }] : []),
+    ...(IS_MAC ? [{ label: app.name, submenu: [
+      { role: "about" }, { label: "Check for Updates…", click: () => checkForUpdate({ manual: true }) },
+      { type: "separator" }, { role: "services" }, { type: "separator" },
+      { role: "hide" }, { role: "hideOthers" }, { role: "unhide" }, { type: "separator" }, { role: "quit" },
+    ] }] : []),
    { role: "fileMenu" },
    { role: "editMenu" },
    {
      label: "Frame",
      submenu: [
-        { label: "Set Up Connection…", click: setUpConnection },
-        { label: IS_MAC ? "Open SSH in Terminal" : "Open SSH in a Terminal", click: () => runInTerminal(["ssh", FRAME]) },
+        { label: "Set Up Connection…", click: () => setUpConnection() },
+        { label: IS_MAC ? "Open SSH in Terminal" : "Open SSH in a Terminal", click: openSsh },
+        { type: "separator" },
+        ...(devices.length > 1 ? [{
+          label: "Headset",
+          submenu: devices.map(d => ({ label: d.name, type: "radio", checked: d.active,
+                                       click: () => { if (win) win.webContents.send("use-device", d.id); } })),
+        }] : []),
+        { label: "Devices…", accelerator: "CmdOrCtrl+5", click: showDevices },
        { type: "separator" },
        { label: "Open in Browser", click: () => url && shell.openExternal(url) },
        { label: "Restart Server", click: () => win ? restartServer() : createWindow() },
@@ -364,7 +606,15 @@ function buildMenu() {
    ...(IS_MAC ? [{ role: "windowMenu" }] : []),
    {
      role: "help",
-      submenu: [{ label: "Project on GitHub", click: () => shell.openExternal("https://github.com/saphid/steam-frame") }],
+      submenu: [
+        ...(IS_MAC ? [] : [{ label: "Check for Updates…", click: () => checkForUpdate({ manual: true }) }]),
+        { label: "Report a Problem…", click: () => {
+          if (win && url && linkPage === win.webContents) win.webContents.send("report:open");
+          else shell.openExternal("https://frame-control.pages.dev/feedback/");  // the page isn't up
+        } },
+        { label: "Release Notes", click: () => shell.openExternal(updater.RELEASES) },
+        { label: "Project on GitHub", click: () => shell.openExternal("https://github.com/saphid/steam-frame") },
+      ],
    },
  ];
  Menu.setApplicationMenu(Menu.buildFromTemplate(template));
@@ -387,6 +637,7 @@ if (!app.requestSingleInstanceLock()) {
    registerScheme();
    buildMenu();
    createWindow();
+    scheduleUpdateChecks();
  });
  app.on("activate", () => { if (!win) createWindow(); });
  app.on("window-all-closed", () => app.quit());
@@ -0,0 +1,252 @@
+// "Sign in with Meta" for My Meta library (docs/meta-library.md), the way BSManager's
+// "Log in with Meta" works: Meta's own login page in a Frame Control window, then the
+// access token cookie (oc_ac_at) that page sets is handed to the local server's
+// existing connect path (format check, "who am I" with Meta, 0600 file).
+//
+// The window has its own persistent session (PARTITION), so the cookie is read only
+// from there: never from Frame Control's main window, never from any browser. It has
+// no preload and no Node, and it only navigates within Meta's login sites; anything
+// else opens in the system browser. Nothing here logs the token, and the page only
+// ever gets back the server's answer (connected, display name), never the token.
+// Electron is passed in, so app/test/meta-signin.test.js can run it with fakes.
+const crypto = require("crypto");
+
+const PARTITION = "persist:meta-signin";
+// BSManager's loginUrl (src/main/services/bs-version-download/bs-oculus-downloader.service.ts).
+const LOGIN_URL = "https://secure.oculus.com";
+const COOKIE = "oc_ac_at";
+// Sites the login can go through: the Oculus store web, Meta accounts, Facebook login,
+// and Instagram (Meta's login passes through www.instagram.com: seen live, 2026-10-14).
+const ALLOWED = ["oculus.com", "meta.com", "facebook.com", "instagram.com"];
+const TIMEOUT = 10 * 60 * 1000;  // as BSManager
+
+const underDomain = (host, domains) => {
+  host = String(host || "").replace(/^\./, "").toLowerCase();
+  return domains.some((d) => host === d || host.endsWith("." + d));
+};
+
+// Plain HTTPS on the default port, to one of the login sites.
+function allowedUrl(raw) {
+  let u;
+  try { u = new URL(String(raw)); } catch { return false; }
+  if (u.protocol !== "https:" || u.username || u.password || (u.port && u.port !== "443")) return false;
+  return underDomain(u.hostname, ALLOWED);
+}
+
+// Frames inside the page: any HTTPS address (Meta's login can embed captcha and CDN
+// frames from other sites), or an empty frame; never data:, blob:, javascript:, http:,
+// file: or any other scheme. Frames can't reach the cookie read or IPC (main process only).
+function allowedSubframeUrl(raw) {
+  if (/^about:(?:blank|srcdoc)$/.test(String(raw))) return true;
+  let u;
+  try { u = new URL(String(raw)); } catch { return false; }
+  return u.protocol === "https:" && !u.username && !u.password;
+}
+
+const onOculus = (raw) => { try { return underDomain(new URL(raw).hostname, ["oculus.com"]); } catch { return false; } };
+
+// A cheap first look, so a signed-out or app value is never sent anywhere; the server's
+// check_token (ui/frame_meta.py) has the final word.
+function looksLikeUserToken(v) {
+  return typeof v === "string" && /^(?:OC|FRL)[A-Za-z0-9+/=_-]{18,4093}$/.test(v)
+    && !/^OC\d{15}$/.test(v) && /[a-z]/.test(v);
+}
+
+const fingerprint = (v) => crypto.createHash("sha256").update(v).digest("hex");
+
+// Anything token-shaped out of a message, in case a server error ever carried one.
+function scrub(text, token) {
+  let s = String(text == null ? "" : text);
+  if (token) s = s.split(token).join("<redacted>");
+  return s.replace(/(?:OC|FRL)[A-Za-z0-9+/=_%|:-]{8,}/g, "<redacted>").slice(0, 600);
+}
+
+// deps: { BrowserWindow, session, shell, parent() -> BrowserWindow|null, timeout,
+//         connect(token, attempt) -> Promise<{status, data}>  (the server's connect action),
+//         cancel(attempt) -> Promise  (the server's cancel-signin: never save that attempt) }
+function createMetaSignIn({ BrowserWindow, session, shell, connect, cancel, parent = () => null, timeout = TIMEOUT }) {
+  let active = null;  // { win, promise, finish }
+  const pending = new Set();  // cancels still to reach the server, from flows already finished
+  let signingOut = null;      // the sign-out under way; no sign-in starts while it runs
+  const SIGNING_OUT = "Frame Control is signing out of Meta; try again in a moment.";
+
+  const openOutside = (target) => {
+    if (/^https?:\/\//i.test(String(target))) shell.openExternal(String(target));
+  };
+
+  function signIn() {
+    if (signingOut) return Promise.resolve({ ok: false, error: SIGNING_OUT });
+    if (active) { active.win.focus(); return active.promise; }
+    const ses = session.fromPartition(PARTITION);
+    ses.setPermissionRequestHandler((_wc, _permission, callback) => callback(false));
+    ses.setPermissionCheckHandler(() => false);
+    // Electron adds its own name and the app's to the user agent; send a plain Chrome one.
+    if (ses.getUserAgent && ses.setUserAgent) {
+      ses.setUserAgent(ses.getUserAgent().replace(/\s(?:Electron|frame-control|Frame Control)\/\S+/g, ""));
+    }
+    const p = parent();
+    const win = new BrowserWindow({
+      width: 520, height: 760, title: "Sign in with Meta – Frame Control", autoHideMenuBar: true,
+      ...(p ? { parent: p } : {}),
+      webPreferences: { partition: PARTITION, contextIsolation: true, nodeIntegration: false, sandbox: true,
+                        nodeIntegrationInSubFrames: false, webviewTag: false, safeDialogs: true, devTools: false },
+    });
+    const wc = win.webContents;
+
+    let resolve;
+    const promise = new Promise((r) => { resolve = r; });
+    let done = false, busy = false, again = false, lastError = null, blocked = null;
+    const tried = new Set();  // fingerprints of values already sent, so each goes once
+    // The server refuses to save a connect from an attempt this window gave up on.
+    const attemptId = crypto.randomUUID();
+
+    const onCookie = (_e, cookie, _cause, removed) => {
+      if (!removed && cookie && cookie.name === COOKIE && underDomain(cookie.domain, ["oculus.com"])) attempt();
+    };
+    const timer = setTimeout(() => finish({ ok: false, error: "Sign-in timed out after 10 minutes. Try again." }), timeout);
+    if (timer.unref) timer.unref();
+
+    // Stops cookie intake first, then the window (destroy: a page's beforeunload can't
+    // keep it open), and answers the page. Unless it connected, and if it sent a token, it
+    // then has the server drop this attempt: only once the connect under way has settled,
+    // so the server always sees the cancel after the connect (and deletes the token if
+    // that connect saved it). The returned promise settles when that is done.
+    let finished = null;
+    let inflight = null;  // the last connect sent
+    function finish(result) {
+      if (done) return finished || Promise.resolve();  // destroy() below re-enters through "closed"
+      done = true;
+      ses.cookies.removeListener("changed", onCookie);
+      clearTimeout(timer);
+      if (active && active.win === win) active = null;
+      if (!win.isDestroyed()) win.destroy();
+      resolve(result);
+      if (result.ok || !inflight) return (finished = Promise.resolve());  // nothing sent: nothing to cancel
+      finished = Promise.resolve(inflight).catch(() => {}).then(() => cancel(attemptId)).catch(() => {});
+      pending.add(finished);
+      finished.then(() => pending.delete(finished));
+      return finished;
+    }
+
+    async function attempt() {
+      if (done) return;
+      if (busy) { again = true; return; }
+      busy = true;
+      try {
+        do {
+          again = false;
+          // Only this window's own session, and only Meta's Oculus cookie.
+          const cookies = await ses.cookies.get({ name: COOKIE });
+          const c = cookies.find((k) => underDomain(k.domain, ["oculus.com"]) && looksLikeUserToken(k.value));
+          if (!c || done) continue;
+          const key = fingerprint(c.value);
+          if (tried.has(key)) continue;
+          tried.add(key);
+          let r;
+          try {
+            inflight = connect(c.value, attemptId);
+            r = await inflight;
+          } catch {
+            lastError = "Frame Control's local server didn't answer.";
+            continue;
+          }
+          if (done) return;
+          const d = (r && r.data) || {};
+          if (r && r.status === 200 && d.connected) {
+            finish({ ok: true, connected: true, user: d.user && d.user.name ? { name: scrub(d.user.name, c.value).slice(0, 80) } : null,
+                     message: scrub(d.message || "Connected your Meta account (Experimental)", c.value) });
+            return;
+          }
+          lastError = scrub(d.error || `HTTP ${r && r.status}`, c.value);
+        } while (again && !done);
+      } catch {
+        // The session went away (the window closed); nothing to do.
+      } finally {
+        busy = false;
+      }
+    }
+
+    // Navigation stays on Meta's login sites. A link elsewhere opens in the system
+    // browser; a redirect elsewhere is just stopped (it may carry sign-in data).
+    // Every frame: the window itself stays on the list; frames inside it may load any
+    // HTTPS page (captchas), and one going anywhere else is just stopped.
+    wc.on("will-navigate", (e, target) => { if (!allowedUrl(target)) { e.preventDefault(); openOutside(target); } });
+    // A stopped page change in the window itself says where it went (the host only), so
+    // closing the stuck window explains why instead of "closed before you signed in".
+    const stopped = (to) => {
+      let host = "";
+      try { host = new URL(String(to)).hostname; } catch { /* not an address */ }
+      blocked = `Meta's sign-in went to ${scrub(host || "an address", null).slice(0, 80)}, which Frame Control doesn't allow.`;
+    };
+    wc.on("will-frame-navigate", (e) => {
+      if (e.isMainFrame ? allowedUrl(e.url) : allowedSubframeUrl(e.url)) return;
+      e.preventDefault();
+      if (e.isMainFrame) stopped(e.url);
+    });
+    // Redirects fire for frames too: the same rule per frame (details on the event in
+    // current Electron, positional arguments in older ones; unknown counts as the window).
+    wc.on("will-redirect", (e, target, _inPlace, isMainFrame) => {
+      const to = e && e.url !== undefined ? e.url : target;
+      const main = e && e.isMainFrame !== undefined ? e.isMainFrame : isMainFrame !== false;
+      if (main ? allowedUrl(to) : allowedSubframeUrl(to)) return;
+      e.preventDefault();
+      if (main) stopped(to);
+    });
+    // A page's beforeunload never keeps the window open.
+    wc.on("will-prevent-unload", (e) => e.preventDefault());
+    wc.setWindowOpenHandler(({ url: target }) => {
+      if (allowedUrl(target)) wc.loadURL(target).catch(() => {});
+      else openOutside(target);
+      return { action: "deny" };
+    });
+    // Signed in: the cookie changes, or the window lands back on Meta's store site.
+    ses.cookies.on("changed", onCookie);
+    wc.on("did-navigate", (_e, target) => { blocked = null; if (onOculus(target)) attempt(); });  // it recovered
+    wc.on("did-navigate-in-page", (_e, target) => { if (onOculus(target)) attempt(); });
+    win.on("closed", () => finish({ ok: false, cancelled: true,
+      error: lastError ? `The sign-in window closed before Frame Control could connect. ${lastError}`
+           : blocked ? `The sign-in window was closed before you signed in. ${blocked}`
+                     : "The sign-in window was closed before you signed in." }));
+
+    active = { win, promise, finish };
+    wc.loadURL(LOGIN_URL).catch(() => {});
+    return promise;
+  }
+
+  // Signs out of Meta in Frame Control: stops any sign-in under way (and has the server
+// refuse to save it), destroys the window, then clears everything
+  // the sign-in session stored (cookies, storage, cache), so the next sign-in starts fresh.
+  // One at a time (a second call joins the first), and no sign-in can start meanwhile.
+  // It goes round until no window is open and no connect or cancel is left, so nothing
+  // started before it can set a token after it (or after the Disconnect that follows).
+  function signOut() {
+    if (!signingOut) {
+      signingOut = (async () => {
+        for (;;) {
+          if (active) {
+            const { win, finish } = active;
+            await finish({ ok: false, cancelled: true, error: "Signed out of Meta in Frame Control." });
+            // Clear only once nothing can write to the session again.
+            if (!win.isDestroyed()) throw new Error("The Sign in with Meta window didn't close; try again.");
+          } else if (pending.size) {
+            // A connect under way settles and is cancelled before the caller (Disconnect)
+            // deletes the account, so it can't be saved afterwards.
+            await Promise.all([...pending]);
+          } else {
+            break;
+          }
+        }
+        const ses = session.fromPartition(PARTITION);
+        await ses.clearStorageData();
+        await ses.clearCache();
+        if (ses.clearAuthCache) await ses.clearAuthCache();
+        return true;
+      })().finally(() => { signingOut = null; });
+    }
+    return signingOut;
+  }
+
+  return { signIn, signOut, isOpen: () => !!active, isSigningOut: () => !!signingOut };
+}
+
+module.exports = { createMetaSignIn, allowedUrl, allowedSubframeUrl, looksLikeUserToken, scrub, PARTITION, LOGIN_URL, COOKIE, ALLOWED };
@@ -1,12 +1,12 @@
 {
  "name": "frame-control",
-  "version": "0.3.1",
+  "version": "0.4.10",
  "lockfileVersion": 3,
  "requires": true,
  "packages": {
    "": {
      "name": "frame-control",
-      "version": "0.3.1",
+      "version": "0.4.10",
      "license": "MIT",
      "devDependencies": {
        "electron": "^44.4.5",
@@ -1,7 +1,7 @@
 {
  "name": "frame-control",
  "productName": "Frame Control",
-  "version": "0.3.1",
+  "version": "0.4.10",
  "description": "Desktop app for managing a Valve Steam Frame over SSH",
  "private": true,
  "main": "main.js",
@@ -9,8 +9,8 @@
  "scripts": {
    "start": "env -u ELECTRON_RUN_AS_NODE electron .",
    "icon": "env -u ELECTRON_RUN_AS_NODE electron build/make-icon.js",
-    "dist": "node build/fetch-deps.js mac arm64 && electron-builder --mac --arm64 --publish never",
-    "dist:dir": "node build/fetch-deps.js mac arm64 && electron-builder --mac --arm64 --dir",
+    "dist": "sh ../mac/frame-mac-view/build.sh && node build/fetch-deps.js mac arm64 && electron-builder --mac --arm64 --publish never",
+    "dist:dir": "sh ../mac/frame-mac-view/build.sh && node build/fetch-deps.js mac arm64 && electron-builder --mac --arm64 --dir",
    "dist:linux": "node build/fetch-deps.js linux x64 arm64 && electron-builder --linux --x64 --arm64 --publish never",
    "dist:win": "node build/fetch-deps.js win x64 && electron-builder --win --x64 --publish never"
  },
@@ -29,6 +29,7 @@
        ]
      }
    ],
+    "afterPack": "build/check-resources.js",
    "directories": {
      "output": "dist",
      "buildResources": "build"
@@ -37,6 +38,9 @@
      "main.js",
      "preload.js",
      "install-link.js",
+      "updater.js",
+      "preferences.js",
+      "meta-signin.js",
      "package.json",
      "build/icon.png"
    ],
@@ -46,7 +50,18 @@
        "to": "ui",
        "filter": [
          "*.py",
-          "*.html"
+          "*.html",
+          "*.js",
+          "*.css",
+          "telemetry.json"
+        ]
+      },
+      {
+        "from": "../ui/apk_sources",
+        "to": "ui/apk_sources",
+        "filter": [
+          "*.py",
+          "*.json"
        ]
      },
      {
@@ -61,7 +76,16 @@
        "to": "frame/android",
        "filter": [
          "*.sh",
-          "*.py"
+          "*.py",
+          "*.js"
+        ]
+      },
+      {
+        "from": "../frame/openxr-compat",
+        "to": "frame/openxr-compat",
+        "filter": [
+          "XrApiLayer_FRAME_compat.json",
+          "prebuilt/**/*.so.gz"
        ]
      },
      {
@@ -72,6 +96,16 @@
          "!**/__pycache__/**"
        ]
      },
+      {
+        "from": "../frame/kdeconnect",
+        "to": "frame/kdeconnect",
+        "filter": [
+          "packages.json",
+          "NOTICE.md",
+          "LICENSES/**/*",
+          "packages/*.pkg.tar.zst"
+        ]
+      },
      {
        "from": "../apk-catalog",
        "to": "apk-catalog",
@@ -94,6 +128,14 @@
        "filter": [
          "**/*"
        ]
+      },
+      {
+        "from": "../THIRD_PARTY_NOTICES.md",
+        "to": "THIRD_PARTY_NOTICES.md"
+      },
+      {
+        "from": "../LICENSE",
+        "to": "LICENSE"
      }
    ],
    "mac": {
@@ -105,9 +147,19 @@
        "dmg",
        "zip"
      ],
+      "extraResources": [
+        {
+          "from": "../mac/bin",
+          "to": "mac/bin",
+          "filter": [
+            "frame-mac-view"
+          ]
+        }
+      ],
      "extendInfo": {
        "NSAppleEventsUsageDescription": "Frame Control opens Terminal for SSH sessions and for power actions that need the Developer Mode password.",
-        "NSLocalNetworkUsageDescription": "Frame Control connects to your Steam Frame over SSH on the local network."
+        "NSLocalNetworkUsageDescription": "Frame Control connects to your Steam Frame over SSH on the local network.",
+        "NSScreenCaptureUsageDescription": "Frame Control shows your Mac's windows and screens inside the Steam Frame when you ask it to."
      },
      "artifactName": "Frame-Control-mac-${arch}.${ext}"
    },
@@ -146,7 +198,16 @@
        "zip"
      ],
      "icon": "build/icon.png",
-      "artifactName": "Frame-Control-win-${arch}.${ext}"
+      "artifactName": "Frame-Control-win-${arch}.${ext}",
+      "extraResources": [
+        {
+          "from": "build/desktop/${os}-${arch}",
+          "to": "desktop/bundle",
+          "filter": [
+            "**/*"
+          ]
+        }
+      ]
    },
    "nsis": {
      "oneClick": false,
@@ -0,0 +1,27 @@
+// Client choices live outside the local server's changing URL/origin.
+const fs = require("node:fs");
+const path = require("node:path");
+const KEYS = ["mvSeparate", "padUsed", "pcView"];
+const valid = value => value === "0" || value === "1";
+
+function preferenceStore(file) {
+  function read() {
+    try {
+      const saved = JSON.parse(fs.readFileSync(file, "utf8"));
+      if (!saved || typeof saved !== "object" || Array.isArray(saved)) return {};
+      return Object.fromEntries(KEYS.filter(key => valid(saved[key])).map(key => [key, saved[key]]));
+    } catch { return {}; }
+  }
+  function write(key, value) {
+    if (!KEYS.includes(key) || !valid(value)) throw new Error("Invalid preference");
+    const saved = { ...read(), [key]: value };
+    fs.mkdirSync(path.dirname(file), { recursive: true });
+    const tmp = file + ".tmp";
+    fs.writeFileSync(tmp, JSON.stringify(saved), { mode: 0o600 });
+    fs.renameSync(tmp, file);
+    return true;
+  }
+  return { read, write };
+}
+
+module.exports = { preferenceStore };
@@ -1,16 +1,56 @@
-// Lets the page read this computer's clipboard through Electron, so sending it
-// to the Frame needs no pbpaste, PowerShell, xclip or wl-clipboard. Also tells
+// Lets the page read and write this computer's clipboard through Electron, so sending
+// it to the Frame, and copying the Frame's back, needs no pbpaste, PowerShell, xclip or wl-clipboard. Also tells
 // the page where a dropped file or folder lives, so a folder can be sideloaded
 // as a title without zipping it (the local server reads it from there).
-// It can open Set Up Connection when the headset can't be reached.
+// It can put a screenshot on the clipboard as an image, open Set Up Connection when
+// the headset can't be reached, and keeps the Frame menu's list of headsets up to date.
 // It also receives frame-control://install links (docs/web-install.md): only
 // what the link asked for, never an install; the page asks the user first.
+// And it passes update state both ways: see app/updater.js. It opens the Sign in
+// with Meta window and signs out of it (app/meta-signin.js).
 const { contextBridge, ipcRenderer, webUtils } = require("electron");

 contextBridge.exposeInMainWorld("frameApp", {
+  preferences: {
+    read: () => ipcRenderer.invoke("preferences:read"),
+    write: (key, value) => ipcRenderer.invoke("preferences:write", key, value),
+  },
+  notify: (message, request) => ipcRenderer.invoke("comfort:notify", message, request),
  readClipboard: () => ipcRenderer.invoke("clipboard:read"),
+  writeClipboard: (text) => ipcRenderer.invoke("clipboard:write", text),
+  writeImage: (bytes) => ipcRenderer.invoke("clipboard:writeImage", bytes),
  setUpConnection: () => ipcRenderer.invoke("connection:setup"),
+  restartServer: () => ipcRenderer.invoke("server:restart"),  // the "couldn't start" page's Try Again
+  // The Frame menu's headset switcher: the page tells it the headsets, and hears picks.
+  devicesChanged: (list) => ipcRenderer.send("devices:changed", list),
+  onUseDevice: (cb) => {
+    ipcRenderer.removeAllListeners("use-device");
+    ipcRenderer.on("use-device", (_e, id) => cb(String(id)));
+  },
+  // While the keyboard-and-trackpad panel holds the keyboard, ⌘W, ⌘R and the rest go to the Frame.
+  captureKeys: (on) => ipcRenderer.send("keys:capture", !!on),
  pathForFile: (file) => { try { return webUtils.getPathForFile(file) || ""; } catch { return ""; } },
+  // Updates (app/updater.js): the page shows a banner and an Update button.
+  update: {
+    get: () => ipcRenderer.invoke("update:get"),
+    check: () => ipcRenderer.invoke("update:check"),
+    install: () => ipcRenderer.invoke("update:install"),
+    onState: (cb) => {
+      ipcRenderer.removeAllListeners("update:state");
+      ipcRenderer.on("update:state", (_e, s) => cb(s));
+    },
+  },
+  // Help → Report a Problem… opens the page's report dialog (ui/frame_report.py).
+  onReportProblem: (cb) => {
+    ipcRenderer.removeAllListeners("report:open");
+    ipcRenderer.on("report:open", () => cb());
+  },
+  // Sign in with Meta (app/meta-signin.js): Meta's login page in its own window. The
+  // answer is the server's (connected, display name), never the token.
+  meta: {
+    signIn: (acknowledged) => ipcRenderer.invoke("meta:signin", acknowledged === true),
+    signOut: () => ipcRenderer.invoke("meta:signout"),
+  },
  onInstallLink: (cb) => {
    ipcRenderer.removeAllListeners("install-link");
    ipcRenderer.on("install-link", (_e, req) => cb({ kind: req.kind, target: req.target }));
@@ -0,0 +1,578 @@
+// Run: node --test app/test/
+// "Sign in with Meta" (app/meta-signin.js) with fake Electron windows, sessions and cookies.
+const nodeTest = require("node:test");
+// A broken flow can leave a sign-in waiting forever; fail it instead of hanging.
+const test = (name, fn) => nodeTest(name, { timeout: 5000 }, fn);
+const assert = require("node:assert");
+const { EventEmitter } = require("node:events");
+const { createMetaSignIn, allowedUrl, looksLikeUserToken, scrub, PARTITION, LOGIN_URL, COOKIE } = require("../meta-signin");
+
+const TOKEN = "OCAbcdefghijklmnopQRSTUVwxyz0123456789abcdEFGH";
+const OTHER = "OCzyxwvutsrqponmlkJIHGFEDcba9876543210zyxwVUTS";
+
+class FakeCookies extends EventEmitter {
+  constructor() { super(); this.store = []; this.reads = 0; }
+  async get(filter) {
+    this.reads++;
+    return this.store.filter((c) => !filter.name || c.name === filter.name);
+  }
+  set(cookie) {  // as the page would: store it, then tell listeners
+    this.store = this.store.filter((c) => !(c.name === cookie.name && c.domain === cookie.domain)).concat(cookie);
+    this.emit("changed", {}, cookie, "explicit", false);
+  }
+}
+
+class FakeSession {
+  constructor(name, log) { this.name = name; this.log = log; this.cookies = new FakeCookies(); this.cleared = []; this.permission = null; }
+  setPermissionRequestHandler(fn) { this.permission = fn; }
+  setPermissionCheckHandler(fn) { this.permissionCheck = fn; }
+  getUserAgent() { return "Mozilla/5.0 Chrome/140.0 frame-control/0.4.4 Electron/44.4.5 Safari/537.36"; }
+  setUserAgent(ua) { this.ua = ua; }
+  async clearStorageData(opts) { this.log.push("clear"); this.cleared.push(["storage", opts]); }
+  async clearCache() { this.cleared.push(["cache"]); }
+  async clearAuthCache() { this.cleared.push(["auth"]); }
+}
+
+function fakes({ timeout } = {}) {
+  const log = [];  // the order things happened in
+  const sessions = new Map();
+  const defaultSession = new FakeSession("default", log);
+  const session = {
+    defaultSession,
+    fromPartition(name) {
+      if (!sessions.has(name)) sessions.set(name, new FakeSession(name, log));
+      return sessions.get(name);
+    },
+  };
+  const windows = [];
+  class BrowserWindow extends EventEmitter {
+    constructor(opts) {
+      super();
+      this.opts = opts;
+      this.destroyed = false;
+      this.veto = false;  // a page whose beforeunload keeps the window open on close()
+      this.focused = 0;
+      const wc = new EventEmitter();
+      wc.loaded = [];
+      wc.loadURL = async (u) => { wc.loaded.push(u); };
+      wc.setWindowOpenHandler = (fn) => { wc.openHandler = fn; };
+      // Deliberately a different session: the code must read the partition's, not guess.
+      wc.session = defaultSession;
+      this.webContents = wc;
+      windows.push(this);
+    }
+    isDestroyed() { return this.destroyed; }
+    focus() { this.focused++; }
+    close() { if (this.destroyed || this.veto) return; this.destroy(); }
+    destroy() { if (this.destroyed) return; this.destroyed = true; log.push("destroyed"); this.emit("closed"); }
+  }
+  const opened = [];
+  const shell = { openExternal: (u) => { opened.push(u); } };
+  const calls = [], attempts = [], cancels = [];
+  let answer = () => ({ status: 200, data: { connected: true, user: { name: "Alex" }, message: "Connected your Meta account (Experimental)" } });
+  const connect = async (token, attempt) => { calls.push(token); attempts.push(attempt); log.push("connect"); return answer(token); };
+  const cancel = async (attempt) => {
+    cancels.push(attempt);
+    log.push(`cancel (cookie listeners: ${session.fromPartition(PARTITION).cookies.listenerCount("changed")})`);
+  };
+  const flow = createMetaSignIn({ BrowserWindow, session, shell, connect, cancel, parent: () => null, ...(timeout ? { timeout } : {}) });
+  return { flow, session, sessions, windows, opened, calls, attempts, cancels, log, setAnswer: (fn) => { answer = fn; },
+           meta: () => session.fromPartition(PARTITION) };
+}
+
+const tick = () => new Promise((r) => setImmediate(r));
+async function settle() { for (let i = 0; i < 10; i++) await tick(); }
+
+function navigate(wc, event, target) {
+  let prevented = false;
+  wc.emit(event, { preventDefault: () => { prevented = true; } }, target);
+  return prevented;
+}
+
+test("only plain HTTPS addresses on Meta's login sites are allowed", () => {
+  for (const ok of ["https://secure.oculus.com", "https://secure.oculus.com/my/profile", "https://auth.meta.com/login/",
+                    "https://www.meta.com/", "https://www.facebook.com/login.php", "https://oculus.com:443/",
+                    "https://www.instagram.com/accounts/login/"]) {
+    assert.ok(allowedUrl(ok), ok);
+  }
+  for (const bad of ["http://secure.oculus.com/", "https://oculus.com.evil.example/", "https://evil-oculus.com/",
+                     "https://notmeta.com/", "https://example.com/", "javascript:alert(1)", "file:///etc/passwd",
+                     "https://secure.oculus.com:8443/", "https://user:pw@secure.oculus.com/", "https://evil.example/?u=https://www.meta.com",
+                     "data:text/html,hi", "not a url"]) {
+    assert.ok(!allowedUrl(bad), bad);
+  }
+});
+
+test("the window is isolated: its own partition, no preload, no Node, sandboxed, on Meta's login page", async () => {
+  const f = fakes();
+  f.flow.signIn();
+  const [w] = f.windows;
+  const wp = w.opts.webPreferences;
+  assert.strictEqual(wp.partition, PARTITION);
+  assert.strictEqual(PARTITION, "persist:meta-signin");
+  assert.strictEqual(wp.contextIsolation, true);
+  assert.strictEqual(wp.nodeIntegration, false);
+  assert.strictEqual(wp.sandbox, true);
+  assert.strictEqual(wp.preload, undefined);
+  assert.strictEqual(wp.devTools, false);
+  assert.deepStrictEqual(w.webContents.loaded, [LOGIN_URL]);
+  assert.strictEqual(LOGIN_URL, "https://secure.oculus.com");
+  // Permission prompts (camera, notifications...) are refused.
+  let granted = null;
+  f.meta().permission(null, "media", (v) => { granted = v; });
+  assert.strictEqual(granted, false);
+  assert.strictEqual(f.meta().permissionCheck(null, "notifications", "https://secure.oculus.com"), false);
+  assert.ok(!/Electron|frame-control/.test(f.meta().ua));
+  w.close();
+});
+
+test("navigation stays on Meta's sites; other links open in the system browser", async () => {
+  const f = fakes();
+  f.flow.signIn();
+  const wc = f.windows[0].webContents;
+  assert.strictEqual(navigate(wc, "will-navigate", "https://auth.meta.com/login/"), false);
+  assert.strictEqual(navigate(wc, "will-navigate", "https://www.facebook.com/login.php"), false);
+  assert.strictEqual(navigate(wc, "will-navigate", "https://example.com/help"), true);
+  assert.deepStrictEqual(f.opened, ["https://example.com/help"]);
+  // A redirect elsewhere is stopped and not opened anywhere (it may carry sign-in data).
+  assert.strictEqual(navigate(wc, "will-redirect", "https://evil.example/cb?x=1"), true);
+  assert.strictEqual(navigate(wc, "will-redirect", "https://auth.meta.com/next"), false);
+  assert.deepStrictEqual(f.opened, ["https://example.com/help"]);
+  // New windows: never created. Meta's own load in this window; others go outside; non-web schemes nowhere.
+  assert.deepStrictEqual(wc.openHandler({ url: "https://www.meta.com/legal/terms/" }), { action: "deny" });
+  assert.deepStrictEqual(wc.openHandler({ url: "https://example.org/" }), { action: "deny" });
+  assert.deepStrictEqual(wc.openHandler({ url: "file:///etc/passwd" }), { action: "deny" });
+  await settle();
+  assert.deepStrictEqual(wc.loaded, [LOGIN_URL, "https://www.meta.com/legal/terms/"]);
+  assert.deepStrictEqual(f.opened, ["https://example.com/help", "https://example.org/"]);
+  f.windows[0].close();
+});
+
+test("the cookie is read only from the sign-in partition, only for Meta's Oculus domain", async () => {
+  const f = fakes();
+  // The same cookie elsewhere: the default session and another partition. Never read.
+  f.session.defaultSession.cookies.store.push({ name: COOKIE, domain: ".oculus.com", value: OTHER });
+  f.session.fromPartition("persist:other").cookies.store.push({ name: COOKIE, domain: ".oculus.com", value: OTHER });
+  const done = f.flow.signIn();
+  const w = f.windows[0];
+  // A cookie with the right name on another site, and a signed-out value: ignored.
+  f.meta().cookies.set({ name: COOKIE, domain: ".evil.example", value: OTHER });
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: "OC123456789012345" });
+  w.webContents.emit("did-navigate", {}, "https://secure.oculus.com/my/profile");
+  await settle();
+  assert.deepStrictEqual(f.calls, []);
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: TOKEN });
+  const r = await done;
+  assert.deepStrictEqual(f.calls, [TOKEN]);
+  assert.strictEqual(r.ok, true);
+  assert.strictEqual(f.session.defaultSession.cookies.reads, 0);
+  assert.strictEqual(f.session.fromPartition("persist:other").cookies.reads, 0);
+  assert.ok(f.sessions.get(PARTITION).cookies.reads > 0);
+});
+
+test("landing on the logged-in page picks up a cookie set earlier", async () => {
+  const f = fakes();
+  const done = f.flow.signIn();
+  f.meta().cookies.store.push({ name: COOKIE, domain: "secure.oculus.com", value: TOKEN });  // no change event
+  f.windows[0].webContents.emit("did-navigate", {}, "https://auth.meta.com/login/");  // not Meta's store site yet
+  await settle();
+  assert.deepStrictEqual(f.calls, []);
+  f.windows[0].webContents.emit("did-navigate", {}, "https://secure.oculus.com/my/profile");
+  assert.strictEqual((await done).ok, true);
+  assert.deepStrictEqual(f.calls, [TOKEN]);
+});
+
+test("after a connect the window closes, and the page gets the answer, never the token", async () => {
+  const f = fakes();
+  const done = f.flow.signIn();
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: TOKEN });
+  const r = await done;
+  assert.ok(f.windows[0].isDestroyed());
+  assert.ok(!f.flow.isOpen());
+  assert.deepStrictEqual(r, { ok: true, connected: true, user: { name: "Alex" }, message: "Connected your Meta account (Experimental)" });
+  assert.ok(!JSON.stringify(r).includes(TOKEN));
+  // The cookie listener is gone: a later change sends nothing.
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: OTHER });
+  await settle();
+  assert.deepStrictEqual(f.calls, [TOKEN]);
+});
+
+test("a token the server refuses keeps the window open, is sent once, and the next one connects", async () => {
+  const f = fakes();
+  f.setAnswer((t) => t === TOKEN ? { status: 400, data: { error: "Meta didn't accept this token." } }
+                                 : { status: 200, data: { connected: true, user: null } });
+  const done = f.flow.signIn();
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: TOKEN });
+  await settle();
+  f.windows[0].webContents.emit("did-navigate", {}, "https://secure.oculus.com/my/profile");  // same value again
+  await settle();
+  assert.deepStrictEqual(f.calls, [TOKEN]);
+  assert.ok(!f.windows[0].isDestroyed());
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: OTHER });
+  const r = await done;
+  assert.deepStrictEqual(f.calls, [TOKEN, OTHER]);
+  assert.strictEqual(r.ok, true);
+  assert.strictEqual(r.user, null);
+});
+
+test("the token is never logged, and an error that echoed it is scrubbed", async () => {
+  const seen = [];
+  const names = ["log", "info", "warn", "error", "debug", "trace"];
+  const saved = Object.fromEntries(names.map((n) => [n, console[n]]));
+  const writes = [process.stdout.write, process.stderr.write];
+  for (const n of names) console[n] = (...a) => { seen.push(a.map(String).join(" ")); };
+  process.stdout.write = (c, ...rest) => { seen.push(String(c)); return writes[0].call(process.stdout, c, ...rest); };
+  process.stderr.write = (c, ...rest) => { seen.push(String(c)); return writes[1].call(process.stderr, c, ...rest); };
+  let r;
+  try {
+    const f = fakes();
+    f.setAnswer((t) => ({ status: 500, data: { error: `RuntimeError: bad ${t} and ${t.slice(0, 12)}` } }));
+    const done = f.flow.signIn();
+    f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: TOKEN });
+    await settle();
+    f.windows[0].close();  // the user gives up
+    r = await done;
+  } finally {
+    for (const n of names) console[n] = saved[n];
+    [process.stdout.write, process.stderr.write] = writes;
+  }
+  assert.strictEqual(r.ok, false);
+  assert.strictEqual(r.cancelled, true);
+  assert.ok(!r.error.includes(TOKEN) && !r.error.includes(TOKEN.slice(0, 12)), r.error);
+  assert.match(r.error, /<redacted>/);
+  assert.ok(!seen.some((s) => s.includes(TOKEN.slice(0, 12))), "token in output");
+  assert.strictEqual(scrub(`x FRL${"a".repeat(30)} y`), "x <redacted> y");
+});
+
+test("closing the window before signing in reports it, and a second click reuses the open window", async () => {
+  const f = fakes();
+  const a = f.flow.signIn();
+  const b = f.flow.signIn();
+  assert.strictEqual(a, b);
+  assert.strictEqual(f.windows.length, 1);
+  assert.strictEqual(f.windows[0].focused, 1);
+  f.windows[0].close();
+  const r = await a;
+  assert.deepStrictEqual(r, { ok: false, cancelled: true, error: "The sign-in window was closed before you signed in." });
+  assert.deepStrictEqual(f.calls, []);
+});
+
+test("disconnect's sign-out closes the window and clears only the sign-in partition", async () => {
+  const f = fakes();
+  f.session.fromPartition("persist:other");
+  const pending = f.flow.signIn();
+  await f.flow.signOut();
+  const r = await pending;
+  assert.strictEqual(r.ok, false);
+  assert.ok(f.windows[0].isDestroyed());
+  assert.deepStrictEqual(f.meta().cleared, [["storage", undefined], ["cache"], ["auth"]]);
+  assert.deepStrictEqual(f.session.defaultSession.cleared, []);
+  assert.deepStrictEqual(f.session.fromPartition("persist:other").cleared, []);
+  // The next sign-in starts with a new window.
+  f.flow.signIn();
+  assert.strictEqual(f.windows.length, 2);
+  f.windows[1].close();
+});
+
+test("the first look at a cookie value matches the server's token rules", () => {
+  assert.ok(looksLikeUserToken(TOKEN));
+  assert.ok(looksLikeUserToken("FRL" + "aB3".repeat(10)));
+  for (const bad of [undefined, "", "OC123456789012345", "OC|123|abc", "OCabc%2Fdef" + "x".repeat(20), "OCshort",
+                     "XY" + "a".repeat(30), "OC" + "A".repeat(30)]) {
+    assert.ok(!looksLikeUserToken(bad), String(bad));
+  }
+});
+
+function frameNav(wc, url, isMainFrame) {
+  let prevented = false;
+  wc.emit("will-frame-navigate", { url, isMainFrame, preventDefault: () => { prevented = true; } });
+  return prevented;
+}
+
+test("the window stays on Meta's sites; frames inside it may load any HTTPS page, nothing else", async () => {
+  const f = fakes();
+  f.flow.signIn();
+  const wc = f.windows[0].webContents;
+  // Frames: captcha and CDN frames from other sites load; other schemes are stopped.
+  for (const ok of ["https://client-api.arkoselabs.com/fc/gc/", "https://newassets.hcaptcha.com/captcha/v1/x",
+                    "https://www.google.com/recaptcha/api2/anchor", "https://static.xx.fbcdn.net/x",
+                    "https://www.facebook.com/plugin", "about:blank", "about:srcdoc"]) {
+    assert.strictEqual(frameNav(wc, ok, false), false, ok);
+  }
+  for (const bad of ["http://www.facebook.com/plugin", "data:text/html,<script>x</script>", "blob:https://evil.example/1",
+                     "javascript:alert(1)", "file:///etc/passwd", "ftp://example.com/", "https://u:p@evil.example/",
+                     "chrome://settings", "not a url"]) {
+    assert.strictEqual(frameNav(wc, bad, false), true, bad);
+  }
+  // The window itself: Meta's sites only.
+  assert.strictEqual(frameNav(wc, "about:blank", true), true);
+  assert.strictEqual(frameNav(wc, "https://evil.example/", true), true);
+  assert.strictEqual(frameNav(wc, "https://client-api.arkoselabs.com/", true), true);
+  assert.strictEqual(frameNav(wc, "https://auth.meta.com/login/", true), false);
+  assert.deepStrictEqual(f.opened, []);  // stopped, never opened outside
+  f.windows[0].destroy();
+});
+
+function redirect(wc, url, isMainFrame, style) {
+  let prevented = false;
+  const e = { preventDefault: () => { prevented = true; } };
+  if (style === "details") wc.emit("will-redirect", Object.assign(e, { url, isMainFrame }));
+  else wc.emit("will-redirect", e, url, false, isMainFrame);
+  return prevented;
+}
+
+test("redirects follow the same rule per frame", async () => {
+  const f = fakes();
+  f.flow.signIn();
+  const wc = f.windows[0].webContents;
+  for (const style of ["details", "positional"]) {
+    // A captcha frame redirected within HTTPS loads; a frame redirected to data: or http: doesn't.
+    assert.strictEqual(redirect(wc, "https://client-api.arkoselabs.com/v2/next", false, style), false, style);
+    assert.strictEqual(redirect(wc, "https://newassets.hcaptcha.com/x", false, style), false, style);
+    assert.strictEqual(redirect(wc, "data:text/html,<b>x</b>", false, style), true, style);
+    assert.strictEqual(redirect(wc, "http://www.facebook.com/x", false, style), true, style);
+    assert.strictEqual(redirect(wc, "javascript:alert(1)", false, style), true, style);
+    // The window: Meta's sites only.
+    assert.strictEqual(redirect(wc, "https://client-api.arkoselabs.com/v2/next", true, style), true, style);
+    assert.strictEqual(redirect(wc, "https://auth.meta.com/next", true, style), false, style);
+  }
+  // Without frame information it is treated as the window.
+  assert.strictEqual(redirect(wc, "https://client-api.arkoselabs.com/", undefined, "positional"), true);
+  assert.deepStrictEqual(f.opened, []);
+  f.windows[0].destroy();
+});
+
+test("a redirect the window stops is named when the window is closed", async () => {
+  const f = fakes();
+  const p = f.flow.signIn();
+  const wc = f.windows[0].webContents;
+  assert.strictEqual(redirect(wc, "https://www.instagram.com/x?next=1", true, "details"), false);  // Meta's login goes here
+  assert.strictEqual(redirect(wc, "https://login.evil.example/cb?code=SECRET", true, "details"), true);
+  f.windows[0].close();
+  const r = await p;
+  assert.strictEqual(r.ok, false);
+  assert.match(r.error, /went to login\.evil\.example, which Frame Control doesn't allow/);
+  assert.doesNotMatch(r.error, /SECRET|code=/);
+});
+
+test("a stopped redirect is forgotten once the window loads a page again", async () => {
+  const f = fakes();
+  const p = f.flow.signIn();
+  const wc = f.windows[0].webContents;
+  assert.strictEqual(redirect(wc, "https://login.evil.example/cb", true, "details"), true);
+  wc.emit("did-navigate", {}, "https://auth.meta.com/login/");  // the user carried on
+  f.windows[0].close();
+  assert.deepStrictEqual(await p, { ok: false, cancelled: true, error: "The sign-in window was closed before you signed in." });
+});
+
+test("a page's beforeunload can't keep the window open", async () => {
+  const f = fakes();
+  const done = f.flow.signIn();
+  const w = f.windows[0];
+  let prevented = false;
+  w.webContents.emit("will-prevent-unload", { preventDefault: () => { prevented = true; } });
+  assert.ok(prevented);
+  // Even if close() would be vetoed, finishing destroys the window.
+  w.veto = true;
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: TOKEN });
+  assert.strictEqual((await done).ok, true);
+  assert.ok(w.isDestroyed());
+});
+
+test("closing during a connect: the cancel goes to the server only after that connect settles", async () => {
+  for (const outcome of ["saved", "refused", "failed"]) {
+    const f = fakes();
+    let release, fail;
+    f.setAnswer(() => new Promise((r, j) => { release = r; fail = j; }));
+    const done = f.flow.signIn();
+    f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: TOKEN });
+    await settle();
+    assert.deepStrictEqual(f.calls, [TOKEN]);
+    f.windows[0].close();  // the user gives up while the server and Meta are answering
+    const r = await done;  // the page hears at once
+    assert.strictEqual(r.ok, false);
+    await settle();
+    assert.deepStrictEqual(f.cancels, [], outcome);  // not before the connect has settled
+    if (outcome === "saved") release({ status: 200, data: { connected: true } });
+    else if (outcome === "refused") release({ status: 400, data: { error: "no" } });
+    else fail(new Error("socket hang up"));
+    await settle();
+    // The same attempt id the connect carried, after it settled and after cookie intake stopped.
+    assert.deepStrictEqual(f.cancels, [f.attempts[0]], outcome);
+    assert.match(f.cancels[0], /^[A-Za-z0-9-]{8,64}$/);
+    assert.deepStrictEqual(f.log, ["connect", "destroyed", "cancel (cookie listeners: 0)"], outcome);
+    assert.deepStrictEqual(f.calls, [TOKEN]);
+  }
+});
+
+test("closing before any connect needs no cancel", async () => {
+  const f = fakes();
+  const done = f.flow.signIn();
+  f.meta().cookies.set({ name: COOKIE, domain: ".evil.example", value: TOKEN });  // nothing usable
+  await settle();
+  f.windows[0].close();
+  assert.strictEqual((await done).ok, false);
+  await f.flow.signOut();
+  await settle();
+  assert.deepStrictEqual(f.calls, []);
+  assert.deepStrictEqual(f.cancels, []);
+});
+
+test("a timeout during a connect cancels after it settles", async () => {
+  const f = fakes({ timeout: 50 });
+  let release;
+  f.setAnswer(() => new Promise((r) => { release = r; }));  // Meta is slow
+  const done = f.flow.signIn();
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: TOKEN });
+  const r = await done;
+  assert.match(r.error, /timed out/);
+  assert.ok(f.windows[0].isDestroyed());
+  assert.deepStrictEqual(f.cancels, []);
+  release({ status: 400, data: { error: "no" } });
+  await settle();
+  assert.deepStrictEqual(f.cancels, [f.attempts[0]]);
+});
+
+test("disconnect during a connect waits for it, cancels it, then clears, before the page deletes the account", async () => {
+  const f = fakes();
+  let release;
+  f.setAnswer(() => new Promise((r) => { release = r; }));
+  const done = f.flow.signIn();
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: TOKEN });
+  await settle();
+  f.log.length = 0;
+  let signedOut = false;
+  const out = f.flow.signOut().then(() => { signedOut = true; });  // the page's Disconnect, step one
+  assert.strictEqual((await done).ok, false);
+  await settle();
+  assert.strictEqual(signedOut, false);  // still waiting for the connect
+  assert.deepStrictEqual(f.cancels, []);
+  release({ status: 200, data: { connected: true } });  // it saved, just now
+  await out;
+  // Only now would the page post its disconnect: the cancel (which deletes what that
+  // connect saved) has reached the server, and the session is cleared.
+  assert.deepStrictEqual(f.log, ["destroyed", "cancel (cookie listeners: 0)", "clear"]);
+  assert.deepStrictEqual(f.cancels, [f.attempts[0]]);
+});
+
+test("a later disconnect still waits for a closed window's connect", async () => {
+  const f = fakes();
+  let release;
+  f.setAnswer(() => new Promise((r) => { release = r; }));
+  const done = f.flow.signIn();
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: TOKEN });
+  await settle();
+  f.windows[0].close();
+  await done;
+  let signedOut = false;
+  const out = f.flow.signOut().then(() => { signedOut = true; });
+  await settle();
+  assert.strictEqual(signedOut, false);
+  release({ status: 400, data: { error: "no" } });
+  await out;
+  assert.deepStrictEqual(f.cancels, [f.attempts[0]]);
+});
+
+test("nothing sent means nothing to cancel; sign-out destroys the window, cancels, then clears", async () => {
+  const f = fakes();
+  // Nothing sent: closing cancels nothing.
+  const a = f.flow.signIn();
+  f.windows[0].close();
+  await a;
+  assert.deepStrictEqual(f.cancels, []);
+  // Something sent, then signed out: the attempt is cancelled before the session is cleared.
+  f.setAnswer(() => ({ status: 400, data: { error: "Meta didn't accept this token." } }));
+  const b = f.flow.signIn();
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: TOKEN });
+  await settle();
+  f.windows[1].veto = true;  // and its page tries to stay open
+  f.log.length = 0;
+  await f.flow.signOut();
+  assert.strictEqual((await b).ok, false);
+  assert.ok(f.windows[1].isDestroyed());
+  assert.deepStrictEqual(f.log, ["destroyed", "cancel (cookie listeners: 0)", "clear"]);
+  assert.deepStrictEqual(f.cancels, [f.attempts[0]]);
+  // After sign-out a new cookie in the partition sends nothing.
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: OTHER });
+  await settle();
+  assert.deepStrictEqual(f.calls, [TOKEN]);
+});
+
+test("each sign-in has its own attempt id", async () => {
+  const f = fakes();
+  f.setAnswer(() => ({ status: 400, data: { error: "no" } }));
+  const a = f.flow.signIn();
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: TOKEN });
+  await settle();
+  f.windows[0].close();
+  await a;
+  const b = f.flow.signIn();
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: OTHER });
+  await settle();
+  f.windows[1].close();
+  await b;
+  await settle();
+  assert.strictEqual(f.attempts.length, 2);
+  assert.notStrictEqual(f.attempts[0], f.attempts[1]);
+  assert.deepStrictEqual(f.cancels, f.attempts);
+});
+
+test("no sign-in can start while a sign-out waits on an earlier connect", async () => {
+  const f = fakes();
+  let release;
+  f.setAnswer(() => new Promise((r) => { release = r; }));
+  const first = f.flow.signIn();
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: TOKEN });
+  await settle();
+  const out = f.flow.signOut();  // Disconnect pressed while that connect is under way
+  assert.strictEqual((await first).ok, false);
+  assert.ok(f.flow.isSigningOut());
+  // The page answered, so the user tries again at once: refused, no window, nothing sent.
+  const second = await f.flow.signIn();
+  assert.deepStrictEqual(second, { ok: false, error: "Frame Control is signing out of Meta; try again in a moment." });
+  assert.strictEqual(f.windows.length, 1);
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: OTHER });
+  await settle();
+  assert.deepStrictEqual(f.calls, [TOKEN]);
+  // A second sign-out joins the first.
+  assert.strictEqual(f.flow.signOut(), out);
+  release({ status: 200, data: { connected: true } });
+  await out;
+  assert.deepStrictEqual(f.cancels, [f.attempts[0]]);
+  assert.strictEqual(f.meta().cleared.filter(([k]) => k === "storage").length, 1);
+  assert.ok(!f.flow.isSigningOut());
+  // Once it is done, signing in works again.
+  f.setAnswer(() => ({ status: 200, data: { connected: true } }));
+  const third = f.flow.signIn();
+  assert.strictEqual(f.windows.length, 2);
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: OTHER });
+  assert.strictEqual((await third).ok, true);
+});
+
+test("sign-out goes round until no window and no connect or cancel is left", async () => {
+  const f = fakes();
+  let releaseA;
+  f.setAnswer(() => new Promise((r) => { releaseA = r; }));
+  const a = f.flow.signIn();
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: TOKEN });
+  await settle();
+  f.windows[0].close();  // window A closed; its cancel waits on its connect
+  await a;
+  // Window B is open, with its own connect under way, when sign-out starts.
+  let releaseB;
+  f.setAnswer(() => new Promise((r) => { releaseB = r; }));
+  const b = f.flow.signIn();
+  f.meta().cookies.set({ name: COOKIE, domain: ".oculus.com", value: OTHER });
+  await settle();
+  let finished = false;
+  const out = f.flow.signOut().then(() => { finished = true; });
+  await b;
+  releaseB({ status: 400, data: { error: "no" } });
+  await settle();
+  assert.strictEqual(finished, false);  // A's connect is still under way
+  releaseA({ status: 200, data: { connected: true } });
+  await out;
+  assert.deepStrictEqual([...f.cancels].sort(), [...f.attempts].sort());
+  assert.strictEqual(f.attempts.length, 2);
+  assert.strictEqual(f.log[f.log.length - 1], "clear");
+});
@@ -0,0 +1,68 @@
+const test = require("node:test");
+const assert = require("node:assert/strict");
+const fs = require("node:fs");
+const os = require("node:os");
+const path = require("node:path");
+const { preferenceStore } = require("../preferences");
+
+function fixture(t) {
+  const dir = fs.mkdtempSync(path.join(os.tmpdir(), "frame-preferences-"));
+  t.after(() => fs.rmSync(dir, { recursive: true, force: true }));
+  return path.join(dir, "client-preferences.json");
+}
+
+test("desktop packaging includes every local application module dependency", () => {
+  const root = path.resolve(__dirname, "..");
+  const packaged = new Set(require("../package.json").build.files);
+  for (const file of ["main.js", "preload.js"]) {
+    const source = fs.readFileSync(path.join(root, file), "utf8");
+    for (const match of source.matchAll(/require\(["'](\.\/[^"']+)["']\)/g)) {
+      const target = match[1].slice(2) + ".js";
+      assert.ok(packaged.has(target), `${file} needs packaged ${target}`);
+    }
+  }
+});
+
+test("client choices survive a new native session and local server origin", t => {
+  const file = fixture(t);
+  const first = preferenceStore(file);
+  assert.deepEqual(first.read(), {});
+  first.write("mvSeparate", "0");
+  first.write("padUsed", "1");
+  // A new process/session reads the same client store, regardless of its URL.
+  const second = preferenceStore(file);
+  assert.deepEqual(second.read(), { mvSeparate: "0", padUsed: "1" });
+  second.write("mvSeparate", "1");
+  assert.deepEqual(first.read(), { mvSeparate: "1", padUsed: "1" });
+});
+
+test("only the non-sensitive binary choices can enter the native store", t => {
+  const file = fixture(t), store = preferenceStore(file);
+  for (const [key, value] of [["token", "secret"], ["__proto__", "1"], ["padUsed", true], ["mvSeparate", "yes"]]) {
+    assert.throws(() => store.write(key, value), /Invalid preference/);
+  }
+  assert.equal(fs.existsSync(file), false);
+  fs.writeFileSync(file, JSON.stringify({ token: "discard", mvSeparate: "0", padUsed: 1 }));
+  assert.deepEqual(store.read(), { mvSeparate: "0" });
+  store.write("padUsed", "1");
+  assert.deepEqual(JSON.parse(fs.readFileSync(file, "utf8")), { mvSeparate: "0", padUsed: "1" });
+});
+
+test("a missing or damaged preference file retains usable defaults", t => {
+  const file = fixture(t), store = preferenceStore(file);
+  assert.deepEqual(store.read(), {});
+  for (const bad of ["{broken", "null", "[]", '"text"']) {
+    fs.writeFileSync(file, bad);
+    assert.deepEqual(store.read(), {});
+  }
+  store.write("padUsed", "1");
+  assert.deepEqual(store.read(), { padUsed: "1" });
+});
+
+test("the PC in the headset opt-in is remembered and off unless saved", t => {
+  const file = fixture(t), store = preferenceStore(file);
+  assert.equal(store.read().pcView, undefined);
+  store.write("pcView", "1");
+  assert.deepEqual(preferenceStore(file).read(), { pcView: "1" });
+  assert.throws(() => store.write("pcView", "on"), /Invalid preference/);
+});
@@ -0,0 +1,80 @@
+// Run: node --test app/test/
+const test = require("node:test");
+const assert = require("node:assert");
+const { isNewer, assetName, updateMethod, macBundle } = require("../updater");
+
+test("versions compare numerically, and a release beats its pre-releases", () => {
+  assert.ok(isNewer("0.3.10", "0.3.9"));
+  assert.ok(isNewer("v1.0.0", "0.9.9"));
+  assert.ok(!isNewer("0.3.1", "0.3.1"));
+  assert.ok(!isNewer("0.3.0", "0.3.1"));
+  assert.ok(isNewer("1.0.0", "1.0.0-beta.1"));
+  assert.ok(!isNewer("1.0.0-beta.1", "1.0.0"));
+  assert.ok(!isNewer("garbage", "0.1.0"));
+});
+
+test("asset names match what electron-builder publishes", () => {
+  assert.strictEqual(assetName("darwin", "arm64", "mac-zip"), "Frame-Control-mac-arm64.zip");
+  assert.strictEqual(assetName("win32", "x64", "nsis"), "Frame-Control-Setup-x64.exe");
+  assert.strictEqual(assetName("linux", "x64", "appimage"), "Frame-Control-linux-x86_64.AppImage");
+  assert.strictEqual(assetName("linux", "arm64", "appimage"), "Frame-Control-linux-arm64.AppImage");
+});
+
+const base = { isPackaged: true, env: {}, exists: () => false, writable: () => true };
+
+test("macOS updates in place only from a writable, non-translocated location", () => {
+  const exe = "/Applications/Frame Control.app/Contents/MacOS/Frame Control";
+  assert.strictEqual(macBundle(exe), "/Applications/Frame Control.app");
+  assert.deepStrictEqual(updateMethod({ ...base, platform: "darwin", execPath: exe }),
+                         { method: "mac-zip", bundle: "/Applications/Frame Control.app" });
+  const dmg = "/Volumes/Frame Control 0.3.1/Frame Control.app/Contents/MacOS/Frame Control";
+  assert.strictEqual(updateMethod({ ...base, platform: "darwin", execPath: dmg }).method, "manual");
+  const trans = "/private/var/folders/x/AppTranslocation/ABC/d/Frame Control.app/Contents/MacOS/Frame Control";
+  assert.strictEqual(updateMethod({ ...base, platform: "darwin", execPath: trans }).method, "manual");
+  assert.strictEqual(updateMethod({ ...base, platform: "darwin", execPath: exe, writable: () => false }).method, "manual");
+});
+
+test("Windows needs the installer's copy; Linux needs an AppImage", () => {
+  const exe = "C:\\Users\\a\\AppData\\Local\\Programs\\Frame Control\\Frame Control.exe";
+  assert.strictEqual(updateMethod({ ...base, platform: "win32", execPath: exe, exists: () => true }).method, "nsis");
+  assert.strictEqual(updateMethod({ ...base, platform: "win32", execPath: exe }).method, "manual");
+  assert.strictEqual(updateMethod({ ...base, platform: "linux", execPath: "/opt/x", env: { APPIMAGE: "/home/a/F.AppImage" } }).method,
+                     "appimage");
+  assert.strictEqual(updateMethod({ ...base, platform: "linux", execPath: "/opt/Frame Control/frame-control" }).method, "manual");
+  assert.strictEqual(updateMethod({ ...base, isPackaged: false, platform: "darwin", execPath: "x" }).method, "manual");
+});
+
+test("update.json assets always download from this repository's release", () => {
+  const r = require("../updater").fromManifest({ version: "0.4.0", notes: "n", assets: [
+    { name: "Frame-Control-mac-arm64.zip", url: "https://evil.example/x.zip", digest: "sha256:" + "a".repeat(64) }] });
+  assert.strictEqual(r.assets[0].url, "https://github.com/saphid/frame-control/releases/download/v0.4.0/Frame-Control-mac-arm64.zip");
+  assert.throws(() => require("../updater").fromManifest({ version: "nope", assets: [] }));
+});
+
+test("the release page is this repository's tag page, never a draft's untagged-... link", () => {
+  const { fromManifest, fromApi } = require("../updater");
+  const tagPage = "https://github.com/saphid/frame-control/releases/tag/v0.4.0";
+  const page = (p) => fromManifest({ version: "0.4.0", assets: [], page: p }).page;
+  assert.strictEqual(page(tagPage), tagPage);
+  // 0.4.0's real update.json: the draft's address, a 404 once published.
+  assert.strictEqual(page("https://github.com/saphid/frame-control/releases/tag/untagged-c6ddfed7f75d67db2e99"), tagPage);
+  assert.strictEqual(page(undefined), tagPage);
+  assert.strictEqual(page("https://evil.example/releases/tag/v0.4.0"), tagPage);
+  assert.strictEqual(page("https://github.com/saphid/frame-control/releases/tag/v0.4.0?x=1"), tagPage);
+  // Paths that normalize to another repository.
+  assert.strictEqual(page("https://github.com/saphid/frame-control/releases/tag/..\\..\\..\\..\\other-owner\\other-repo"),
+                     tagPage);
+  assert.strictEqual(page("https://github.com/saphid/frame-control/releases/tag/%2e%2e/%2e%2e/%2e%2e/%2e%2e/other-owner/other-repo"),
+                     tagPage);
+  assert.strictEqual(page("https://github.com/saphid/frame-control/releases/tag/v0.3.9"), tagPage);  // only its own tag
+  assert.strictEqual(fromApi({ tag_name: "../../x", assets: [] }).page, "https://github.com/saphid/frame-control/releases");
+  assert.strictEqual(fromApi({ tag_name: "v0.4.0", assets: [],
+                               html_url: "https://github.com/saphid/frame-control/releases/tag/untagged-x" }).page, tagPage);
+});
+
+test("prepare refuses a release that isn't newer (no downgrades)", async () => {
+  const { prepare } = require("../updater");
+  const release = { version: "0.3.1", assets: [] };
+  await assert.rejects(prepare(release, { method: "appimage", appImage: "/nonexistent/x" }, null, "0.4.0"), /isn't newer/);
+  await assert.rejects(prepare(release, { method: "appimage", appImage: "/nonexistent/x" }, null, "0.3.1"), /isn't newer/);
+});
@@ -0,0 +1,263 @@
+// Update checks and self-update for the desktop app (docs/releasing.md).
+//
+// The newest version is GitHub's "latest" release of saphid/frame-control. Drafts
+// and pre-releases never count, so a build reaches people only when the
+// maintainer publishes it after testing (scripts/publish-release.sh). That
+// script attaches update.json (version, notes, each asset's SHA-256), read
+// through github.com's latest/download link: the REST API allows only 60
+// unauthenticated requests an hour per IP address, shared by everyone behind
+// the same router, so it's only the fallback.
+//
+// Every download is checked against the SHA-256 digest GitHub records for the
+// asset before anything is replaced. How the update is applied:
+//   macOS     the .zip: unpacked next to the running app, swapped in by a small
+//             script once the app has quit, then reopened.
+//   Windows   the NSIS installer, run silently over the current install; it
+//             reopens the app. A copy unpacked from the .zip is updated by hand.
+//   Linux     the AppImage replaces itself; .deb installs are updated by hand.
+// When the app can't update itself it opens the release page instead.
+const { execFile, spawn } = require("child_process");
+const crypto = require("crypto");
+const fs = require("fs");
+const https = require("https");
+const os = require("os");
+const path = require("path");
+
+const REPO = "saphid/frame-control";  // renamed from saphid/steam-frame; GitHub redirects the old name
+const LATEST = `https://api.github.com/repos/${REPO}/releases/latest`;
+const MANIFEST = `https://github.com/${REPO}/releases/latest/download/update.json`;
+const RELEASES = `https://github.com/${REPO}/releases`;
+
+// "0.3.1" or "v0.3.1" -> [0, 3, 1]; pre-release suffixes sort before the release.
+function parseVersion(v) {
+  const m = String(v || "").trim().replace(/^v/i, "").match(/^(\d+)\.(\d+)\.(\d+)(?:-([0-9A-Za-z.-]+))?$/);
+  return m ? { nums: [+m[1], +m[2], +m[3]], pre: m[4] || null } : null;
+}
+
+function isNewer(candidate, current) {
+  const a = parseVersion(candidate), b = parseVersion(current);
+  if (!a || !b) return false;
+  for (let i = 0; i < 3; i++) if (a.nums[i] !== b.nums[i]) return a.nums[i] > b.nums[i];
+  if (a.pre === b.pre) return false;
+  if (!a.pre) return true;   // 1.0.0 is newer than 1.0.0-beta
+  if (!b.pre) return false;
+  return a.pre > b.pre;
+}
+
+// The asset this copy of the app updates from, by the names electron-builder gives them.
+function assetName(platform, arch, method) {
+  if (method === "mac-zip") return `Frame-Control-mac-${arch}.zip`;
+  if (method === "nsis") return `Frame-Control-Setup-${arch}.exe`;
+  if (method === "appimage") return `Frame-Control-linux-${arch === "x64" ? "x86_64" : arch}.AppImage`;
+  return null;
+}
+
+// How this copy can update itself: mac-zip, nsis, appimage, or manual (with why).
+function updateMethod({ platform, isPackaged, execPath, env, exists, writable }) {
+  if (!isPackaged) return { method: "manual", why: "running from a source checkout" };
+  if (platform === "darwin") {
+    const bundle = macBundle(execPath);
+    if (!bundle) return { method: "manual", why: "can't find the app bundle" };
+    if (bundle.includes("/AppTranslocation/") || bundle.startsWith("/Volumes/")) {
+      return { method: "manual", why: "move Frame Control to Applications first" };
+    }
+    if (!writable(path.dirname(bundle))) return { method: "manual", why: `${path.dirname(bundle)} isn't writable` };
+    return { method: "mac-zip", bundle };
+  }
+  if (platform === "win32") {
+    // electron-builder's NSIS install puts its uninstaller next to the app.
+    const dir = path.dirname(execPath);
+    if (exists(path.join(dir, "Uninstall Frame Control.exe"))) return { method: "nsis" };
+    return { method: "manual", why: "not installed with the installer" };
+  }
+  if (platform === "linux" && env.APPIMAGE) {
+    if (!writable(path.dirname(env.APPIMAGE))) return { method: "manual", why: "the AppImage's folder isn't writable" };
+    return { method: "appimage", appImage: env.APPIMAGE };
+  }
+  return { method: "manual", why: "installed from a package" };
+}
+
+function macBundle(execPath) {
+  const i = execPath.indexOf(".app/Contents/MacOS/");
+  return i < 0 ? null : execPath.slice(0, i + 4);
+}
+
+function get(url, { headers = {}, timeout = 20000, redirects = 5 } = {}) {
+  return new Promise((resolve, reject) => {
+    const req = https.get(url, { headers: { "user-agent": "FrameControl-updater", ...headers }, timeout }, (res) => {
+      if ([301, 302, 303, 307, 308].includes(res.statusCode) && res.headers.location && redirects > 0) {
+        res.resume();
+        const next = new URL(res.headers.location, url);
+        if (next.protocol !== "https:") return reject(new Error("refusing a non-HTTPS redirect"));
+        return resolve(get(next.href, { headers, timeout, redirects: redirects - 1 }));
+      }
+      if (res.statusCode !== 200) { res.resume(); return reject(new Error(`HTTP ${res.statusCode} from ${new URL(url).host}`)); }
+      resolve(res);
+    });
+    req.on("timeout", () => req.destroy(new Error("timed out")));
+    req.on("error", reject);
+  });
+}
+
+async function getJson(url, headers) {
+  const res = await get(url, { headers });
+  let body = "";
+  for await (const chunk of res) body += chunk;
+  return JSON.parse(body);
+}
+
+// The release page the banner links to. update.json for 0.4.0 carried the draft's
+// address (releases/tag/untagged-...), which is dead once the release is published,
+// and a page from the manifest could also be steered elsewhere (e.g. tag/..\..\other/repo
+// normalizes to another repository) before shell.openExternal. So the given page is
+// ignored: the link is always this repository's tag page, built from a valid version.
+function releasePage(version) {
+  const v = parseVersion(version);
+  if (!v) return RELEASES;
+  return `${RELEASES}/tag/v${v.nums.join(".")}${v.pre ? "-" + v.pre : ""}`;
+}
+
+// update.json and the API's release both become { version, notes, page, assets }.
+function fromManifest(m) {
+  if (!parseVersion(m.version) || !Array.isArray(m.assets)) throw new Error("update.json is malformed");
+  const version = String(m.version).replace(/^v/i, "");
+  const base = `https://github.com/${REPO}/releases/download/v${version}/`;
+  return { version, notes: String(m.notes || "").slice(0, 4000),
+           page: releasePage(version),
+           // Assets always come from this repository's release, whatever the manifest says.
+           assets: m.assets.map((a) => ({ name: String(a.name), url: base + encodeURIComponent(String(a.name)),
+                                          size: a.size, digest: a.digest || null })) };
+}
+
+function fromApi(r) {
+  if (r.draft || r.prerelease) throw new Error("GitHub returned an unpublished release");
+  const version = String(r.tag_name || "").replace(/^v/i, "");
+  return { version, notes: String(r.body || "").slice(0, 4000),
+           page: releasePage(version),
+           assets: (r.assets || []).map((a) => ({ name: a.name, url: a.browser_download_url, size: a.size,
+                                                 digest: a.digest || null })) };
+}
+
+async function latestRelease() {
+  try {
+    return fromManifest(await getJson(MANIFEST));
+  } catch (e) {
+    if (!/HTTP 404/.test(e.message)) throw e;  // releases before update.json existed
+  }
+  return fromApi(await getJson(LATEST, { accept: "application/vnd.github+json" }));
+}
+
+async function download(asset, dest, onProgress) {
+  const m = /^sha256:([0-9a-f]{64})$/.exec(asset.digest || "");
+  if (!m) throw new Error(`GitHub has no SHA-256 for ${asset.name}, so it can't be checked`);
+  const res = await get(asset.url, { timeout: 60000 });
+  const total = +res.headers["content-length"] || asset.size || 0;
+  const hash = crypto.createHash("sha256");
+  // "wx": a new file only, never through an existing file or symlink at that path.
+  const out = fs.createWriteStream(dest, { mode: 0o755, flags: "wx" });
+  let done = 0;
+  await new Promise((resolve, reject) => {
+    res.on("data", (chunk) => { hash.update(chunk); done += chunk.length; onProgress && onProgress(done, total); });
+    res.on("error", reject);
+    out.on("error", reject);
+    out.on("finish", resolve);
+    res.pipe(out);
+  });
+  if (hash.digest("hex") !== m[1]) {
+    fs.rmSync(dest, { force: true });
+    throw new Error(`${asset.name} didn't match its SHA-256; nothing was changed`);
+  }
+}
+
+const run = (cmd, args) => new Promise((resolve, reject) =>
+  execFile(cmd, args, { timeout: 120000 }, (err, stdout, stderr) => err ? reject(new Error((stderr || err.message).trim())) : resolve(stdout)));
+
+// Waits for this process to exit, swaps the new bundle in (putting the old one
+// back if that fails), and reopens the app.
+const MAC_SWAP = `set -u
+pid="$1"; app="$2"; new="$3"; stage="$4"
+while kill -0 "$pid" 2>/dev/null; do sleep 0.2; done
+old="$stage/old.app"
+if mv "$app" "$old"; then
+  if mv "$new" "$app"; then rm -rf "$old"; else mv "$old" "$app"; fi
+fi
+xattr -dr com.apple.quarantine "$app" 2>/dev/null
+rm -rf "$stage"
+open "$app"
+`;
+
+async function applyMac(release, bundle, onProgress) {
+  const asset = release.assets.find((a) => a.name === assetName("darwin", process.arch, "mac-zip"));
+  if (!asset) throw new Error(`the release has no ${assetName("darwin", process.arch, "mac-zip")}`);
+  // Staged beside the app, so the final move stays on one volume.
+  const stage = fs.mkdtempSync(path.join(path.dirname(bundle), ".frame-control-update-"));
+  try {
+    const zip = path.join(stage, asset.name);
+    await download(asset, zip, onProgress);
+    await run("/usr/bin/ditto", ["-x", "-k", zip, stage]);
+    fs.rmSync(zip, { force: true });
+    const name = fs.readdirSync(stage).find((n) => n.endsWith(".app"));
+    if (!name) throw new Error("the download has no app in it");
+    const fresh = path.join(stage, name);
+    const version = (await run("/usr/bin/plutil", ["-extract", "CFBundleShortVersionString", "raw",
+                                                   path.join(fresh, "Contents", "Info.plist")])).trim();
+    if (version !== release.version) throw new Error(`the download is version ${version}, not ${release.version}`);
+    const script = path.join(stage, "swap.sh");
+    fs.writeFileSync(script, MAC_SWAP);
+    return () => spawn("/bin/sh", [script, String(process.pid), bundle, fresh, stage],
+                       { detached: true, stdio: "ignore" }).unref();
+  } catch (e) {
+    fs.rmSync(stage, { recursive: true, force: true });
+    throw e;
+  }
+}
+
+async function applyNsis(release, onProgress) {
+  const asset = release.assets.find((a) => a.name === assetName("win32", process.arch, "nsis"));
+  if (!asset) throw new Error(`the release has no ${assetName("win32", process.arch, "nsis")}`);
+  const dir = fs.mkdtempSync(path.join(os.tmpdir(), "frame-control-update-"));
+  const exe = path.join(dir, asset.name);
+  await download(asset, exe, onProgress);
+  // /S: silent, into the existing install. --force-run: open the app afterwards.
+  return () => spawn(exe, ["--updated", "/S", "--force-run"], { detached: true, stdio: "ignore" }).unref();
+}
+
+async function applyAppImage(release, appImage, onProgress) {
+  const asset = release.assets.find((a) => a.name === assetName("linux", process.arch, "appimage"));
+  if (!asset) throw new Error(`the release has no ${assetName("linux", process.arch, "appimage")}`);
+  // A private folder beside the AppImage, so the final rename stays on one filesystem.
+  const stage = fs.mkdtempSync(path.join(path.dirname(appImage), ".frame-control-update-"));
+  try {
+    const next = path.join(stage, asset.name);
+    await download(asset, next, onProgress);
+    fs.chmodSync(next, 0o755);
+    fs.renameSync(next, appImage);  // the running copy keeps its open file
+  } finally {
+    fs.rmSync(stage, { recursive: true, force: true });
+  }
+  // Without FUSE the AppImage runs extracted (--appimage-extract-and-run, which isn't passed
+  // on to the app); a FUSE mount lives under /tmp/.mount_*. Keep the same mode on restart.
+  const extracted = process.env.APPIMAGE_EXTRACT_AND_RUN === "1" || !process.execPath.includes("/.mount_");
+  const env = { ...process.env, APPIMAGE: appImage, ...(extracted ? { APPIMAGE_EXTRACT_AND_RUN: "1" } : {}) };
+  // Started only once this process has exited, or the new copy would lose the single-instance lock.
+  return () => spawn("/bin/sh", ["-c", 'while kill -0 "$1" 2>/dev/null; do sleep 0.2; done; exec "$2"',
+                                 "sh", String(process.pid), appImage], { detached: true, stdio: "ignore", env }).unref();
+}
+
+// Downloads and prepares the update; returns a function that starts the swap,
+// to be called just before the app quits.
+async function prepare(release, how, onProgress, current) {
+  if (!isNewer(release.version, current)) throw new Error(`${release.version} isn't newer than ${current}`);
+  if (how.method === "mac-zip") return applyMac(release, how.bundle, onProgress);
+  if (how.method === "nsis") return applyNsis(release, onProgress);
+  if (how.method === "appimage") return applyAppImage(release, how.appImage, onProgress);
+  throw new Error(how.why || "this copy can't update itself");
+}
+
+function writable(dir) {
+  try { fs.accessSync(dir, fs.constants.W_OK); return true; } catch { return false; }
+}
+
+module.exports = { REPO, RELEASES, parseVersion, isNewer, assetName, updateMethod, macBundle, latestRelease,
+                   fromManifest, fromApi,
+                   download, prepare, writable };
@@ -0,0 +1,24 @@
+<!doctype html>
+<!-- Control experiment, run on the Frame itself: how often does Chromium on
+     gamescope put a canvas animation on screen, with no network or decoding
+     involved? Draws every animation frame for ?s= seconds and writes the
+     frame-gap histogram per second into the title, where xprop can read it. -->
+<html><head><meta charset="utf-8"><title>fmv-present</title></head>
+<body style="margin:0;background:#000"><canvas id="c" width="1280" height="720" style="width:100%;height:100%"></canvas>
+<script>
+const q = new URLSearchParams(location.search), secs = +q.get("s") || 15, tag = q.get("tag") || "";
+const ctx = document.getElementById("c").getContext("2d", { alpha: false, desynchronized: true });
+const perSec = []; let t0 = 0, last = 0, n = 0, gaps = [];
+function frame(t) {
+  if (!t0) t0 = last = t;
+  ctx.fillStyle = "#123"; ctx.fillRect(0, 0, 1280, 720);
+  ctx.fillStyle = "#fff"; ctx.fillRect((n * 21) % 1280, 0, 20, 720); n++;
+  const s = Math.floor((t - t0) / 1000);
+  (perSec[s] = perSec[s] || []).push(t - last); last = t;
+  if (t - t0 < secs * 1000) return requestAnimationFrame(frame);
+  const out = perSec.map(g => g.length).join(",");
+  document.title = `[${tag}] done fps/s=${out}`;
+}
+document.title = `[${tag}] running`;
+requestAnimationFrame(frame);
+</script></body></html>
@@ -0,0 +1,32 @@
+<!doctype html>
+<!-- Benchmark content: a long page of text that scrolls itself at a steady
+     speed, so every frame changes the way reading a real page does. -->
+<html lang="en"><head><meta charset="utf-8"><title>fmv-bench scroll</title>
+<style>
+  body { font: 17px/1.55 -apple-system, "Helvetica Neue", sans-serif; margin: 0 auto; max-width: 900px; padding: 24px; color: #1d1d1f; background: #fff; }
+  h2 { font-size: 22px; margin: 28px 0 8px; } code { background: #f2f2f5; padding: 1px 4px; border-radius: 3px; }
+</style></head><body><div id="doc"></div>
+<script>
+const words = "latency frame panel stream encoder decoder network display pixel window capture budget quality motion text sharp adaptive controller queue socket clock".split(" ");
+let seed = 7; const rnd = () => (seed = (seed * 16807) % 2147483647) / 2147483647;
+let html = "";
+for (let s = 0; s < 120; s++) {
+  html += `<h2>Section ${s + 1}: ${words[s % words.length]} and ${words[(s * 7) % words.length]}</h2>`;
+  for (let p = 0; p < 4; p++) {
+    let para = [];
+    for (let w = 0; w < 70; w++) para.push(rnd() < 0.05 ? `<code>${words[Math.floor(rnd() * words.length)]}()</code>` : words[Math.floor(rnd() * words.length)]);
+    html += `<p>${para.join(" ")}.</p>`;
+  }
+}
+document.getElementById("doc").innerHTML = html;
+// 240 points a second, whatever the display's refresh rate.
+// The title carries the page's own frame rate, so the source's rate is known.
+let last = performance.now(), frames = 0, since = last;
+function step(now) {
+  const dy = (now - last) * 0.24; last = now;
+  if (++frames && now - since >= 1000) { document.title = `fmv-bench scroll ${frames} fps`; frames = 0; since = now; }
+  if (window.scrollY + innerHeight >= document.body.scrollHeight - 2) window.scrollTo(0, 0); else window.scrollBy(0, dy);
+  requestAnimationFrame(step);
+}
+requestAnimationFrame(step);
+</script></body></html>
@@ -0,0 +1,9 @@
+<!doctype html>
+<!-- Benchmark content: a plain text editor, focused, for typing tests. -->
+<html lang="en"><head><meta charset="utf-8"><title>fmv-bench type</title>
+<style>
+  html, body { margin: 0; height: 100%; background: #fff; }
+  textarea { box-sizing: border-box; width: 100%; height: 100%; border: 0; padding: 20px; outline: none; resize: none;
+             font: 20px/1.5 ui-monospace, Menlo, monospace; color: #111; caret-color: transparent; }  /* a blinking caret would pass for a reply to a key */
+</style></head><body><textarea id="t" autofocus spellcheck="false"></textarea>
+<script>document.getElementById("t").focus();</script></body></html>
@@ -0,0 +1,677 @@
+{
+ "label": "usb1",
+ "date": "2026-09-28T21:16:03",
+ "commit": "1d05f57",
+ "config": {
+  "quality": "balanced",
+  "mode": "separate",
+  "net": "none",
+  "delay_ms": 0,
+  "buffer_ms": 250,
+  "host": "10.86.200.233",
+  "ssh_opts": [],
+  "encoder": "",
+  "duration_s": 15.0,
+  "browser_flags": []
+ },
+ "frame_build": "20260925.6191901",
+ "mac": "26.5.2",
+ "headset": "",
+ "scenarios": [
+  {
+   "scenario": "test",
+   "frames_sent": 911,
+   "frames_drawn": 911,
+   "frames_shown": 610,
+   "duration_s": 15.2,
+   "stages_ms": {
+    "capture": {
+     "p50": 0.0,
+     "p95": 0.0,
+     "p99": 0.0,
+     "n": 911
+    },
+    "queue": {
+     "p50": 0.0,
+     "p95": 0.1,
+     "p99": 0.1,
+     "n": 911
+    },
+    "encode": {
+     "p50": 4.4,
+     "p95": 5.0,
+     "p99": 5.2,
+     "n": 911
+    },
+    "socket": {
+     "p50": 0.1,
+     "p95": 0.2,
+     "p99": 0.2,
+     "n": 911
+    },
+    "network": {
+     "p50": 1.1,
+     "p95": 1.6,
+     "p99": 2.1,
+     "n": 911
+    },
+    "decode": {
+     "p50": 1.2,
+     "p95": 2.6,
+     "p99": 3.9,
+     "n": 911
+    },
+    "draw": {
+     "p50": 0.4,
+     "p95": 0.9,
+     "p99": 1.5,
+     "n": 911
+    },
+    "present": {
+     "p50": 5.8,
+     "p95": 13.6,
+     "p99": 16.7,
+     "n": 610
+    },
+    "content": {
+     "p50": 7.3,
+     "p95": 8.8,
+     "p99": 10.0,
+     "n": 911
+    },
+    "content_shown": {
+     "p50": 13.6,
+     "p95": 20.4,
+     "p99": 24.9,
+     "n": 610
+    }
+   },
+   "fps": 59.9,
+   "fps_shown": 40.1,
+   "late_pct": 0.22,
+   "stall_max": 32.3,
+   "stalls_over_100ms": 0,
+   "mbps": 0.48,
+   "keyframes": 1,
+   "size": "1280x720",
+   "captured": 912,
+   "viewer_never_drawn": 0,
+   "input_ms": {
+    "p50": 16.6,
+    "p95": 24.3,
+    "p99": 24.5,
+    "n": 30
+   },
+   "input_shown_ms": {
+    "p50": 20.8,
+    "p95": 31.8,
+    "p99": 33.8,
+    "n": 22
+   },
+   "input_parts_ms": {
+    "uplink": {
+     "p50": 0.7,
+     "p95": 1.0,
+     "p99": 1.0,
+     "n": 30
+    },
+    "mac": {
+     "p50": 9.1,
+     "p95": 15.0,
+     "p99": 15.2,
+     "n": 30
+    },
+    "back": {
+     "p50": 7.2,
+     "p95": 8.5,
+     "p99": 12.7,
+     "n": 30
+    }
+   },
+   "inputs": {
+    "asked": 30,
+    "sent": 30,
+    "seen": 30
+   },
+   "timeline": [
+    {
+     "t": 0,
+     "fps": 60,
+     "mbps": 0.5,
+     "content_p50": 7.3,
+     "content_p95": 8.6,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 1,
+     "fps": 60,
+     "mbps": 0.5,
+     "content_p50": 7.1,
+     "content_p95": 8.9,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 2,
+     "fps": 59,
+     "mbps": 0.48,
+     "content_p50": 7.2,
+     "content_p95": 8.6,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 3,
+     "fps": 60,
+     "mbps": 0.49,
+     "content_p50": 7.1,
+     "content_p95": 9.0,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 4,
+     "fps": 60,
+     "mbps": 0.49,
+     "content_p50": 6.9,
+     "content_p95": 8.2,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 5,
+     "fps": 60,
+     "mbps": 0.58,
+     "content_p50": 7.3,
+     "content_p95": 8.1,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 6,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 7.4,
+     "content_p95": 8.6,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 7,
+     "fps": 60,
+     "mbps": 0.46,
+     "content_p50": 7.4,
+     "content_p95": 8.8,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 8,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 7.3,
+     "content_p95": 8.9,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 9,
+     "fps": 60,
+     "mbps": 0.46,
+     "content_p50": 7.3,
+     "content_p95": 8.8,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 10,
+     "fps": 60,
+     "mbps": 0.45,
+     "content_p50": 7.3,
+     "content_p95": 8.7,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 11,
+     "fps": 60,
+     "mbps": 0.45,
+     "content_p50": 7.4,
+     "content_p95": 9.2,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 12,
+     "fps": 60,
+     "mbps": 0.46,
+     "content_p50": 7.4,
+     "content_p95": 8.4,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 13,
+     "fps": 60,
+     "mbps": 0.46,
+     "content_p50": 7.6,
+     "content_p95": 9.7,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 14,
+     "fps": 60,
+     "mbps": 0.46,
+     "content_p50": 7.6,
+     "content_p95": 8.5,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 15,
+     "fps": 12,
+     "mbps": 0.09,
+     "content_p50": 7.5,
+     "content_p95": 9.6,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    }
+   ],
+   "adapt": [],
+   "content_p50": 7.3,
+   "content_p95": 8.8,
+   "input_p50": 16.6,
+   "input_p95": 24.3,
+   "grades": {
+    "input_replies": "local",
+    "content_p50": "local",
+    "content_p95": "local",
+    "input_p50": "local",
+    "input_p95": "local",
+    "fps": "local",
+    "late_pct": "local",
+    "stall_max": "local"
+   },
+   "controller": {
+    "adapt": true,
+    "baseRtt": 0.917,
+    "ceiling": 5529600,
+    "fps": 60,
+    "inFlight": 1,
+    "scale": 1,
+    "slack": 41.666,
+    "target": 5529600,
+    "tier": 0
+   },
+   "events": [],
+   "source_fps": 60.0,
+   "cpu": {
+    "mac_agent_pct": 16.7,
+    "frame_viewer_pct": 48.5,
+    "frame_tailscaled_pct": 0.5,
+    "frame_sshd_pct": 1.1,
+    "frame_gamescope_pct": 12.0,
+    "frame_total_pct": 39.3
+   },
+   "viewer": "h264 software; ANGLE (Mesa, zink Vulkan 1.4(Turnip Adreno (TM) 750 (MESA_TURNIP)), OpenGL 4.6); raf 137 Hz",
+   "show_s": 1.15,
+   "panel": "valve.steam.desktopgame.2001639889",
+   "src": "test"
+  },
+  {
+   "scenario": "scroll",
+   "frames_sent": 848,
+   "frames_drawn": 848,
+   "frames_shown": 617,
+   "duration_s": 15.1,
+   "stages_ms": {
+    "capture": {
+     "p50": -3.9,
+     "p95": 4.2,
+     "p99": 7.3,
+     "n": 848
+    },
+    "queue": {
+     "p50": 0.0,
+     "p95": 0.1,
+     "p99": 0.1,
+     "n": 848
+    },
+    "encode": {
+     "p50": 6.7,
+     "p95": 9.7,
+     "p99": 17.7,
+     "n": 848
+    },
+    "socket": {
+     "p50": 0.1,
+     "p95": 0.2,
+     "p99": 0.4,
+     "n": 848
+    },
+    "network": {
+     "p50": 1.7,
+     "p95": 2.7,
+     "p99": 5.7,
+     "n": 848
+    },
+    "decode": {
+     "p50": 5.9,
+     "p95": 10.3,
+     "p99": 13.1,
+     "n": 848
+    },
+    "draw": {
+     "p50": 0.7,
+     "p95": 1.6,
+     "p99": 2.2,
+     "n": 848
+    },
+    "present": {
+     "p50": 4.4,
+     "p95": 15.8,
+     "p99": 20.6,
+     "n": 617
+    },
+    "content": {
+     "p50": 15.7,
+     "p95": 23.0,
+     "p99": 39.3,
+     "n": 848
+    },
+    "content_shown": {
+     "p50": 21.3,
+     "p95": 34.1,
+     "p99": 45.6,
+     "n": 617
+    }
+   },
+   "fps": 56.1,
+   "fps_shown": 40.7,
+   "late_pct": 3.3,
+   "stall_max": 238.5,
+   "stalls_over_100ms": 1,
+   "mbps": 9.01,
+   "keyframes": 1,
+   "size": "1920x1290",
+   "captured": 857,
+   "viewer_never_drawn": 0,
+   "input_ms": {
+    "n": 0
+   },
+   "input_shown_ms": {
+    "n": 0
+   },
+   "input_parts_ms": {
+    "uplink": {
+     "n": 0
+    },
+    "mac": {
+     "n": 0
+    },
+    "back": {
+     "n": 0
+    }
+   },
+   "inputs": {
+    "asked": 0,
+    "sent": 0,
+    "seen": 0
+   },
+   "timeline": [
+    {
+     "t": 0,
+     "fps": 44,
+     "mbps": 8.13,
+     "content_p50": 15.7,
+     "content_p95": 35.7,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 1,
+     "fps": 56,
+     "mbps": 7.04,
+     "content_p50": 16.3,
+     "content_p95": 47.3,
+     "bitrate": 7430400,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 2,
+     "fps": 57,
+     "mbps": 8.58,
+     "content_p50": 16.0,
+     "content_p95": 22.6,
+     "bitrate": 10055362,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 3,
+     "fps": 56,
+     "mbps": 9.29,
+     "content_p50": 16.4,
+     "content_p95": 21.1,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 4,
+     "fps": 58,
+     "mbps": 9.01,
+     "content_p50": 15.6,
+     "content_p95": 21.7,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 5,
+     "fps": 57,
+     "mbps": 9.13,
+     "content_p50": 15.9,
+     "content_p95": 21.7,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 6,
+     "fps": 56,
+     "mbps": 11.63,
+     "content_p50": 16.1,
+     "content_p95": 22.1,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 7,
+     "fps": 57,
+     "mbps": 8.7,
+     "content_p50": 16.3,
+     "content_p95": 22.2,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 8,
+     "fps": 57,
+     "mbps": 9.24,
+     "content_p50": 15.7,
+     "content_p95": 21.2,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 9,
+     "fps": 57,
+     "mbps": 9.21,
+     "content_p50": 15.0,
+     "content_p95": 20.5,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 10,
+     "fps": 56,
+     "mbps": 8.55,
+     "content_p50": 15.7,
+     "content_p95": 20.6,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 11,
+     "fps": 58,
+     "mbps": 9.51,
+     "content_p50": 14.7,
+     "content_p95": 19.9,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 12,
+     "fps": 57,
+     "mbps": 8.75,
+     "content_p50": 16.0,
+     "content_p95": 20.4,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 13,
+     "fps": 58,
+     "mbps": 9.25,
+     "content_p50": 15.0,
+     "content_p95": 20.7,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 14,
+     "fps": 57,
+     "mbps": 9.13,
+     "content_p50": 15.3,
+     "content_p95": 21.0,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 15,
+     "fps": 7,
+     "mbps": 1.17,
+     "content_p50": 14.2,
+     "content_p95": 15.3,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    }
+   ],
+   "adapt": [],
+   "content_p50": 15.7,
+   "content_p95": 23.0,
+   "input_p50": null,
+   "input_p95": null,
+   "grades": {
+    "content_p50": "local",
+    "content_p95": "local",
+    "fps": "acceptable",
+    "late_pct": "acceptable",
+    "stall_max": "acceptable"
+   },
+   "controller": {
+    "adapt": true,
+    "baseRtt": 1.25,
+    "ceiling": 14860800,
+    "fps": 60,
+    "inFlight": 0,
+    "scale": 1,
+    "slack": 41.666,
+    "target": 14860800,
+    "tier": 0
+   },
+   "events": [
+    {
+     "t": 1.06,
+     "e": "down to 7430 kbit/s: queue 35 ms, held 4, oldest 32 ms, base 1 ms, sent 6282 got 4914 wants 11892"
+    }
+   ],
+   "source_fps": 56.8,
+   "cpu": {
+    "mac_agent_pct": 8.8,
+    "frame_viewer_pct": 74.8,
+    "frame_tailscaled_pct": 0.3,
+    "frame_sshd_pct": 1.2,
+    "frame_gamescope_pct": 10.5,
+    "frame_total_pct": 39.0
+   },
+   "viewer": "h264 software; ANGLE (Mesa, zink Vulkan 1.4(Turnip Adreno (TM) 750 (MESA_TURNIP)), OpenGL 4.6); raf 137 Hz",
+   "show_s": 1.16,
+   "panel": "valve.steam.desktopgame.2001968301",
+   "src": "separate:7914"
+  }
+ ]
+}
@@ -0,0 +1,690 @@
+{
+ "label": "usb2",
+ "date": "2026-09-28T21:17:48",
+ "commit": "1d05f57",
+ "config": {
+  "quality": "balanced",
+  "mode": "separate",
+  "net": "none",
+  "delay_ms": 0,
+  "buffer_ms": 250,
+  "host": "10.86.200.233",
+  "ssh_opts": [],
+  "encoder": "",
+  "duration_s": 15.0,
+  "browser_flags": []
+ },
+ "frame_build": "20260925.6191901",
+ "mac": "26.5.2",
+ "headset": "",
+ "scenarios": [
+  {
+   "scenario": "test",
+   "frames_sent": 913,
+   "frames_drawn": 913,
+   "frames_shown": 856,
+   "duration_s": 15.2,
+   "stages_ms": {
+    "capture": {
+     "p50": 0.0,
+     "p95": 0.0,
+     "p99": 0.0,
+     "n": 913
+    },
+    "queue": {
+     "p50": 0.0,
+     "p95": 0.1,
+     "p99": 0.1,
+     "n": 913
+    },
+    "encode": {
+     "p50": 4.3,
+     "p95": 4.9,
+     "p99": 5.2,
+     "n": 913
+    },
+    "socket": {
+     "p50": 0.1,
+     "p95": 0.2,
+     "p99": 0.2,
+     "n": 913
+    },
+    "network": {
+     "p50": 0.9,
+     "p95": 1.2,
+     "p99": 1.7,
+     "n": 913
+    },
+    "decode": {
+     "p50": 1.1,
+     "p95": 2.5,
+     "p99": 3.8,
+     "n": 913
+    },
+    "draw": {
+     "p50": 0.3,
+     "p95": 0.8,
+     "p99": 1.2,
+     "n": 913
+    },
+    "present": {
+     "p50": 2.6,
+     "p95": 8.6,
+     "p99": 15.9,
+     "n": 856
+    },
+    "content": {
+     "p50": 6.9,
+     "p95": 8.4,
+     "p99": 9.6,
+     "n": 913
+    },
+    "content_shown": {
+     "p50": 9.7,
+     "p95": 15.8,
+     "p99": 23.1,
+     "n": 856
+    }
+   },
+   "fps": 60.0,
+   "fps_shown": 56.3,
+   "late_pct": 0.0,
+   "stall_max": 22.0,
+   "stalls_over_100ms": 0,
+   "mbps": 0.57,
+   "keyframes": 1,
+   "size": "1280x720",
+   "captured": 913,
+   "viewer_never_drawn": 0,
+   "input_ms": {
+    "p50": 16.9,
+    "p95": 24.1,
+    "p99": 26.1,
+    "n": 33
+   },
+   "input_shown_ms": {
+    "p50": 19.7,
+    "p95": 26.6,
+    "p99": 29.2,
+    "n": 31
+   },
+   "input_parts_ms": {
+    "uplink": {
+     "p50": 0.7,
+     "p95": 2.5,
+     "p99": 5.7,
+     "n": 33
+    },
+    "mac": {
+     "p50": 8.1,
+     "p95": 15.4,
+     "p99": 17.2,
+     "n": 33
+    },
+    "back": {
+     "p50": 7.3,
+     "p95": 8.3,
+     "p99": 8.9,
+     "n": 33
+    }
+   },
+   "inputs": {
+    "asked": 30,
+    "sent": 33,
+    "seen": 33
+   },
+   "timeline": [
+    {
+     "t": 0,
+     "fps": 60,
+     "mbps": 0.49,
+     "content_p50": 7.1,
+     "content_p95": 8.7,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 1,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 7.0,
+     "content_p95": 8.3,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 2,
+     "fps": 60,
+     "mbps": 0.48,
+     "content_p50": 6.9,
+     "content_p95": 8.4,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 3,
+     "fps": 60,
+     "mbps": 0.48,
+     "content_p50": 7.0,
+     "content_p95": 9.1,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 4,
+     "fps": 60,
+     "mbps": 0.5,
+     "content_p50": 6.8,
+     "content_p95": 8.5,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 5,
+     "fps": 60,
+     "mbps": 0.67,
+     "content_p50": 7.0,
+     "content_p95": 8.2,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 6,
+     "fps": 60,
+     "mbps": 0.59,
+     "content_p50": 7.0,
+     "content_p95": 8.2,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 7,
+     "fps": 60,
+     "mbps": 0.63,
+     "content_p50": 6.7,
+     "content_p95": 7.5,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 8,
+     "fps": 60,
+     "mbps": 0.62,
+     "content_p50": 6.9,
+     "content_p95": 8.2,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 9,
+     "fps": 60,
+     "mbps": 0.64,
+     "content_p50": 6.8,
+     "content_p95": 8.0,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 10,
+     "fps": 60,
+     "mbps": 0.67,
+     "content_p50": 6.8,
+     "content_p95": 8.3,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 11,
+     "fps": 60,
+     "mbps": 0.62,
+     "content_p50": 6.7,
+     "content_p95": 7.9,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 12,
+     "fps": 60,
+     "mbps": 0.59,
+     "content_p50": 6.8,
+     "content_p95": 7.9,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 13,
+     "fps": 60,
+     "mbps": 0.57,
+     "content_p50": 6.7,
+     "content_p95": 8.3,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 14,
+     "fps": 60,
+     "mbps": 0.58,
+     "content_p50": 7.0,
+     "content_p95": 8.7,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 15,
+     "fps": 13,
+     "mbps": 0.13,
+     "content_p50": 6.4,
+     "content_p95": 7.3,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    }
+   ],
+   "adapt": [],
+   "content_p50": 6.9,
+   "content_p95": 8.4,
+   "input_p50": 16.9,
+   "input_p95": 24.1,
+   "grades": {
+    "input_replies": "local",
+    "content_p50": "local",
+    "content_p95": "local",
+    "input_p50": "local",
+    "input_p95": "local",
+    "fps": "local",
+    "late_pct": "local",
+    "stall_max": "local"
+   },
+   "controller": {
+    "adapt": true,
+    "baseRtt": 0.709,
+    "ceiling": 5529600,
+    "fps": 60,
+    "inFlight": 0,
+    "scale": 1,
+    "slack": 41.666,
+    "target": 5529600,
+    "tier": 0
+   },
+   "events": [],
+   "source_fps": 60.1,
+   "cpu": {
+    "mac_agent_pct": 17.9,
+    "frame_viewer_pct": 51.9,
+    "frame_tailscaled_pct": 0.2,
+    "frame_sshd_pct": 1.1,
+    "frame_gamescope_pct": 9.6,
+    "frame_total_pct": 27.9
+   },
+   "viewer": "h264 software; ANGLE (Mesa, zink Vulkan 1.4(Turnip Adreno (TM) 750 (MESA_TURNIP)), OpenGL 4.6); raf 134 Hz",
+   "show_s": 1.15,
+   "panel": "valve.steam.desktopgame.2001639889",
+   "src": "test"
+  },
+  {
+   "scenario": "scroll",
+   "frames_sent": 868,
+   "frames_drawn": 868,
+   "frames_shown": 868,
+   "duration_s": 15.2,
+   "stages_ms": {
+    "capture": {
+     "p50": -4.2,
+     "p95": 0.5,
+     "p99": 6.7,
+     "n": 868
+    },
+    "queue": {
+     "p50": 0.0,
+     "p95": 0.1,
+     "p99": 0.2,
+     "n": 868
+    },
+    "encode": {
+     "p50": 6.8,
+     "p95": 10.2,
+     "p99": 17.9,
+     "n": 868
+    },
+    "socket": {
+     "p50": 0.1,
+     "p95": 0.2,
+     "p99": 0.4,
+     "n": 868
+    },
+    "network": {
+     "p50": 1.5,
+     "p95": 2.3,
+     "p99": 7.6,
+     "n": 868
+    },
+    "decode": {
+     "p50": 5.9,
+     "p95": 10.2,
+     "p99": 20.3,
+     "n": 868
+    },
+    "draw": {
+     "p50": 0.7,
+     "p95": 1.4,
+     "p99": 2.1,
+     "n": 868
+    },
+    "present": {
+     "p50": 0.9,
+     "p95": 6.1,
+     "p99": 6.8,
+     "n": 868
+    },
+    "content": {
+     "p50": 15.3,
+     "p95": 23.5,
+     "p99": 53.1,
+     "n": 868
+    },
+    "content_shown": {
+     "p50": 17.7,
+     "p95": 25.6,
+     "p99": 57.5,
+     "n": 868
+    }
+   },
+   "fps": 57.2,
+   "fps_shown": 57.2,
+   "late_pct": 3.23,
+   "stall_max": 108.0,
+   "stalls_over_100ms": 1,
+   "mbps": 9.84,
+   "keyframes": 1,
+   "size": "1920x1290",
+   "captured": 870,
+   "viewer_never_drawn": 0,
+   "input_ms": {
+    "p50": 24.2,
+    "p95": 36.3,
+    "p99": 36.3,
+    "n": 6
+   },
+   "input_shown_ms": {
+    "p50": 24.8,
+    "p95": 38.5,
+    "p99": 38.5,
+    "n": 6
+   },
+   "input_parts_ms": {
+    "uplink": {
+     "p50": 2.0,
+     "p95": 5.4,
+     "p99": 5.4,
+     "n": 6
+    },
+    "mac": {
+     "p50": 7.5,
+     "p95": 15.8,
+     "p99": 15.8,
+     "n": 6
+    },
+    "back": {
+     "p50": 15.3,
+     "p95": 16.1,
+     "p99": 16.1,
+     "n": 6
+    }
+   },
+   "inputs": {
+    "asked": 0,
+    "sent": 6,
+    "seen": 6
+   },
+   "timeline": [
+    {
+     "t": 0,
+     "fps": 53,
+     "mbps": 9.21,
+     "content_p50": 16.1,
+     "content_p95": 71.2,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 1,
+     "fps": 57,
+     "mbps": 9.64,
+     "content_p50": 15.9,
+     "content_p95": 46.8,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 2,
+     "fps": 58,
+     "mbps": 8.99,
+     "content_p50": 14.8,
+     "content_p95": 19.2,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 3,
+     "fps": 57,
+     "mbps": 9.72,
+     "content_p50": 15.2,
+     "content_p95": 22.3,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 4,
+     "fps": 58,
+     "mbps": 9.64,
+     "content_p50": 16.1,
+     "content_p95": 23.5,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 5,
+     "fps": 57,
+     "mbps": 11.27,
+     "content_p50": 15.6,
+     "content_p95": 20.9,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 6,
+     "fps": 58,
+     "mbps": 10.44,
+     "content_p50": 15.1,
+     "content_p95": 19.1,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 7,
+     "fps": 57,
+     "mbps": 9.31,
+     "content_p50": 15.0,
+     "content_p95": 19.9,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 8,
+     "fps": 58,
+     "mbps": 11.69,
+     "content_p50": 15.9,
+     "content_p95": 22.0,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 9,
+     "fps": 58,
+     "mbps": 10.31,
+     "content_p50": 14.5,
+     "content_p95": 20.7,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 10,
+     "fps": 57,
+     "mbps": 10.47,
+     "content_p50": 15.2,
+     "content_p95": 20.6,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 11,
+     "fps": 58,
+     "mbps": 9.44,
+     "content_p50": 15.3,
+     "content_p95": 20.5,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 12,
+     "fps": 57,
+     "mbps": 9.17,
+     "content_p50": 15.9,
+     "content_p95": 18.7,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 13,
+     "fps": 58,
+     "mbps": 8.77,
+     "content_p50": 15.1,
+     "content_p95": 18.6,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 14,
+     "fps": 58,
+     "mbps": 9.72,
+     "content_p50": 15.2,
+     "content_p95": 19.8,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 15,
+     "fps": 9,
+     "mbps": 1.33,
+     "content_p50": 14.8,
+     "content_p95": 17.7,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    }
+   ],
+   "adapt": [],
+   "content_p50": 15.3,
+   "content_p95": 23.5,
+   "input_p50": 24.2,
+   "input_p95": 36.3,
+   "grades": {
+    "input_replies": "local",
+    "content_p50": "local",
+    "content_p95": "local",
+    "input_p50": "local",
+    "input_p95": "local",
+    "fps": "acceptable",
+    "late_pct": "acceptable",
+    "stall_max": "acceptable"
+   },
+   "controller": {
+    "adapt": true,
+    "baseRtt": 1.375,
+    "ceiling": 14860800,
+    "fps": 60,
+    "inFlight": 1,
+    "scale": 1,
+    "slack": 41.666,
+    "target": 14860800,
+    "tier": 0
+   },
+   "events": [],
+   "source_fps": 57.2,
+   "cpu": {
+    "mac_agent_pct": 11.4,
+    "frame_viewer_pct": 83.5,
+    "frame_tailscaled_pct": 0.3,
+    "frame_sshd_pct": 1.5,
+    "frame_gamescope_pct": 9.7,
+    "frame_total_pct": 32.5
+   },
+   "viewer": "h264 software; ANGLE (Mesa, zink Vulkan 1.4(Turnip Adreno (TM) 750 (MESA_TURNIP)), OpenGL 4.6); raf 139 Hz",
+   "show_s": 1.15,
+   "panel": "valve.steam.desktopgame.2001786096",
+   "src": "separate:8187"
+  }
+ ]
+}
@@ -0,0 +1,682 @@
+{
+ "label": "wifi1",
+ "date": "2026-09-28T21:16:56",
+ "commit": "1d05f57",
+ "config": {
+  "quality": "balanced",
+  "mode": "separate",
+  "net": "none",
+  "delay_ms": 0,
+  "buffer_ms": 250,
+  "host": "frame",
+  "ssh_opts": [],
+  "encoder": "",
+  "duration_s": 15.0,
+  "browser_flags": []
+ },
+ "frame_build": "20260925.6191901",
+ "mac": "26.5.2",
+ "headset": "",
+ "scenarios": [
+  {
+   "scenario": "test",
+   "frames_sent": 912,
+   "frames_drawn": 912,
+   "frames_shown": 611,
+   "duration_s": 15.2,
+   "stages_ms": {
+    "capture": {
+     "p50": 0.0,
+     "p95": 0.0,
+     "p99": 0.0,
+     "n": 912
+    },
+    "queue": {
+     "p50": 0.0,
+     "p95": 0.1,
+     "p99": 0.1,
+     "n": 912
+    },
+    "encode": {
+     "p50": 4.3,
+     "p95": 4.9,
+     "p99": 5.1,
+     "n": 912
+    },
+    "socket": {
+     "p50": 0.1,
+     "p95": 0.2,
+     "p99": 0.2,
+     "n": 912
+    },
+    "network": {
+     "p50": 4.0,
+     "p95": 5.3,
+     "p99": 7.8,
+     "n": 912
+    },
+    "decode": {
+     "p50": 1.1,
+     "p95": 2.8,
+     "p99": 4.0,
+     "n": 912
+    },
+    "draw": {
+     "p50": 0.4,
+     "p95": 0.8,
+     "p99": 1.2,
+     "n": 912
+    },
+    "present": {
+     "p50": 5.9,
+     "p95": 14.0,
+     "p99": 18.5,
+     "n": 611
+    },
+    "content": {
+     "p50": 10.1,
+     "p95": 12.3,
+     "p99": 13.9,
+     "n": 912
+    },
+    "content_shown": {
+     "p50": 16.7,
+     "p95": 23.6,
+     "p99": 28.3,
+     "n": 611
+    }
+   },
+   "fps": 60.0,
+   "fps_shown": 40.1,
+   "late_pct": 0.11,
+   "stall_max": 25.8,
+   "stalls_over_100ms": 0,
+   "mbps": 0.48,
+   "keyframes": 1,
+   "size": "1280x720",
+   "captured": 912,
+   "viewer_never_drawn": 0,
+   "input_ms": {
+    "p50": 26.4,
+    "p95": 34.2,
+    "p99": 53.5,
+    "n": 30
+   },
+   "input_shown_ms": {
+    "p50": 28.6,
+    "p95": 41.5,
+    "p99": 55.9,
+    "n": 18
+   },
+   "input_parts_ms": {
+    "uplink": {
+     "p50": 8.9,
+     "p95": 20.4,
+     "p99": 28.0,
+     "n": 30
+    },
+    "mac": {
+     "p50": 5.1,
+     "p95": 14.5,
+     "p99": 16.2,
+     "n": 30
+    },
+    "back": {
+     "p50": 10.1,
+     "p95": 13.7,
+     "p99": 13.8,
+     "n": 30
+    }
+   },
+   "inputs": {
+    "asked": 30,
+    "sent": 30,
+    "seen": 30
+   },
+   "timeline": [
+    {
+     "t": 0,
+     "fps": 60,
+     "mbps": 0.51,
+     "content_p50": 10.3,
+     "content_p95": 11.9,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 1,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 10.5,
+     "content_p95": 12.1,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 2,
+     "fps": 60,
+     "mbps": 0.49,
+     "content_p50": 10.1,
+     "content_p95": 12.8,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 3,
+     "fps": 60,
+     "mbps": 0.48,
+     "content_p50": 9.9,
+     "content_p95": 11.6,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 4,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 9.7,
+     "content_p95": 11.8,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 5,
+     "fps": 60,
+     "mbps": 0.59,
+     "content_p50": 10.2,
+     "content_p95": 11.7,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 6,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 10.2,
+     "content_p95": 12.3,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 7,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 9.7,
+     "content_p95": 12.1,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 8,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 10.4,
+     "content_p95": 13.1,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 9,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 9.9,
+     "content_p95": 12.7,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 10,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 9.8,
+     "content_p95": 12.4,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 11,
+     "fps": 60,
+     "mbps": 0.46,
+     "content_p50": 10.3,
+     "content_p95": 12.7,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 12,
+     "fps": 60,
+     "mbps": 0.49,
+     "content_p50": 10.2,
+     "content_p95": 12.0,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 13,
+     "fps": 60,
+     "mbps": 0.48,
+     "content_p50": 10.0,
+     "content_p95": 11.3,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 14,
+     "fps": 60,
+     "mbps": 0.46,
+     "content_p50": 10.2,
+     "content_p95": 12.3,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 15,
+     "fps": 12,
+     "mbps": 0.1,
+     "content_p50": 10.7,
+     "content_p95": 11.4,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    }
+   ],
+   "adapt": [],
+   "content_p50": 10.1,
+   "content_p95": 12.3,
+   "input_p50": 26.4,
+   "input_p95": 34.2,
+   "grades": {
+    "input_replies": "local",
+    "content_p50": "local",
+    "content_p95": "local",
+    "input_p50": "local",
+    "input_p95": "local",
+    "fps": "local",
+    "late_pct": "local",
+    "stall_max": "local"
+   },
+   "controller": {
+    "adapt": true,
+    "baseRtt": 4.958,
+    "ceiling": 5529600,
+    "fps": 60,
+    "inFlight": 1,
+    "scale": 1,
+    "slack": 47.417,
+    "target": 2764800,
+    "tier": 0
+   },
+   "events": [
+    {
+     "t": 16.58,
+     "e": "down to 2764 kbit/s: queue 19 ms, held 6, oldest 210 ms, base 4 ms, sent 322 got 265 wants 459"
+    }
+   ],
+   "source_fps": 60.0,
+   "cpu": {
+    "mac_agent_pct": 15.7,
+    "frame_viewer_pct": 42.5,
+    "frame_tailscaled_pct": 5.5,
+    "frame_sshd_pct": 0.7,
+    "frame_gamescope_pct": 8.7,
+    "frame_total_pct": 26.1
+   },
+   "viewer": "h264 software; ANGLE (Mesa, zink Vulkan 1.4(Turnip Adreno (TM) 750 (MESA_TURNIP)), OpenGL 4.6); raf 136 Hz",
+   "show_s": 1.26,
+   "panel": "valve.steam.desktopgame.2001639889",
+   "src": "test"
+  },
+  {
+   "scenario": "scroll",
+   "frames_sent": 852,
+   "frames_drawn": 852,
+   "frames_shown": 614,
+   "duration_s": 15.2,
+   "stages_ms": {
+    "capture": {
+     "p50": -4.4,
+     "p95": 2.6,
+     "p99": 4.2,
+     "n": 852
+    },
+    "queue": {
+     "p50": 0.0,
+     "p95": 0.1,
+     "p99": 0.1,
+     "n": 852
+    },
+    "encode": {
+     "p50": 6.7,
+     "p95": 9.8,
+     "p99": 16.7,
+     "n": 852
+    },
+    "socket": {
+     "p50": 0.1,
+     "p95": 0.2,
+     "p99": 0.2,
+     "n": 852
+    },
+    "network": {
+     "p50": 5.3,
+     "p95": 13.1,
+     "p99": 28.2,
+     "n": 852
+    },
+    "decode": {
+     "p50": 6.0,
+     "p95": 10.9,
+     "p99": 19.0,
+     "n": 852
+    },
+    "draw": {
+     "p50": 0.7,
+     "p95": 1.4,
+     "p99": 1.8,
+     "n": 852
+    },
+    "present": {
+     "p50": 4.1,
+     "p95": 16.2,
+     "p99": 20.6,
+     "n": 614
+    },
+    "content": {
+     "p50": 19.5,
+     "p95": 31.4,
+     "p99": 64.5,
+     "n": 852
+    },
+    "content_shown": {
+     "p50": 25.4,
+     "p95": 38.8,
+     "p99": 76.7,
+     "n": 614
+    }
+   },
+   "fps": 56.2,
+   "fps_shown": 40.5,
+   "late_pct": 4.69,
+   "stall_max": 253.7,
+   "stalls_over_100ms": 2,
+   "mbps": 9.02,
+   "keyframes": 1,
+   "size": "1920x1290",
+   "captured": 866,
+   "viewer_never_drawn": 0,
+   "input_ms": {
+    "n": 0
+   },
+   "input_shown_ms": {
+    "n": 0
+   },
+   "input_parts_ms": {
+    "uplink": {
+     "n": 0
+    },
+    "mac": {
+     "n": 0
+    },
+    "back": {
+     "n": 0
+    }
+   },
+   "inputs": {
+    "asked": 0,
+    "sent": 0,
+    "seen": 0
+   },
+   "timeline": [
+    {
+     "t": 0,
+     "fps": 45,
+     "mbps": 7.77,
+     "content_p50": 19.0,
+     "content_p95": 163.1,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 1,
+     "fps": 51,
+     "mbps": 6.54,
+     "content_p50": 19.4,
+     "content_p95": 107.0,
+     "bitrate": 7430400,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 2,
+     "fps": 57,
+     "mbps": 8.64,
+     "content_p50": 19.9,
+     "content_p95": 31.8,
+     "bitrate": 10055362,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 3,
+     "fps": 58,
+     "mbps": 9.46,
+     "content_p50": 19.1,
+     "content_p95": 24.9,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 4,
+     "fps": 58,
+     "mbps": 9.11,
+     "content_p50": 19.4,
+     "content_p95": 22.9,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 5,
+     "fps": 57,
+     "mbps": 9.18,
+     "content_p50": 19.9,
+     "content_p95": 25.3,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 6,
+     "fps": 58,
+     "mbps": 11.64,
+     "content_p50": 19.8,
+     "content_p95": 49.6,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 7,
+     "fps": 57,
+     "mbps": 9.12,
+     "content_p50": 20.5,
+     "content_p95": 31.1,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 8,
+     "fps": 58,
+     "mbps": 9.29,
+     "content_p50": 19.5,
+     "content_p95": 25.7,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 9,
+     "fps": 57,
+     "mbps": 9.43,
+     "content_p50": 19.8,
+     "content_p95": 49.6,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 10,
+     "fps": 58,
+     "mbps": 8.59,
+     "content_p50": 19.2,
+     "content_p95": 24.8,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 11,
+     "fps": 57,
+     "mbps": 9.42,
+     "content_p50": 19.9,
+     "content_p95": 27.1,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 12,
+     "fps": 58,
+     "mbps": 9.0,
+     "content_p50": 20.6,
+     "content_p95": 31.2,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 13,
+     "fps": 57,
+     "mbps": 8.84,
+     "content_p50": 19.6,
+     "content_p95": 25.3,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 14,
+     "fps": 57,
+     "mbps": 9.17,
+     "content_p50": 19.3,
+     "content_p95": 27.8,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 15,
+     "fps": 9,
+     "mbps": 1.52,
+     "content_p50": 19.3,
+     "content_p95": 30.0,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    }
+   ],
+   "adapt": [],
+   "content_p50": 19.5,
+   "content_p95": 31.4,
+   "input_p50": null,
+   "input_p95": null,
+   "grades": {
+    "content_p50": "local",
+    "content_p95": "local",
+    "fps": "acceptable",
+    "late_pct": "acceptable",
+    "stall_max": "bad"
+   },
+   "controller": {
+    "adapt": true,
+    "baseRtt": 6.959,
+    "ceiling": 14860800,
+    "fps": 60,
+    "inFlight": 2,
+    "scale": 1,
+    "slack": 43.54,
+    "target": 14860800,
+    "tier": 0
+   },
+   "events": [
+    {
+     "t": 1.09,
+     "e": "down to 7430 kbit/s: queue 131 ms, held 3, oldest 231 ms, base 6 ms, sent 4272 got 3724 wants 9321"
+    }
+   ],
+   "source_fps": 57.0,
+   "cpu": {
+    "mac_agent_pct": 8.5,
+    "frame_viewer_pct": 71.7,
+    "frame_tailscaled_pct": 8.4,
+    "frame_sshd_pct": 0.9,
+    "frame_gamescope_pct": 8.4,
+    "frame_total_pct": 30.1
+   },
+   "viewer": "h264 software; ANGLE (Mesa, zink Vulkan 1.4(Turnip Adreno (TM) 750 (MESA_TURNIP)), OpenGL 4.6); raf 139 Hz",
+   "show_s": 1.25,
+   "panel": "valve.steam.desktopgame.2001195625",
+   "src": "separate:8050"
+  }
+ ]
+}
@@ -0,0 +1,682 @@
+{
+ "label": "wifi2",
+ "date": "2026-09-28T21:18:40",
+ "commit": "1d05f57",
+ "config": {
+  "quality": "balanced",
+  "mode": "separate",
+  "net": "none",
+  "delay_ms": 0,
+  "buffer_ms": 250,
+  "host": "frame",
+  "ssh_opts": [],
+  "encoder": "",
+  "duration_s": 15.0,
+  "browser_flags": []
+ },
+ "frame_build": "20260925.6191901",
+ "mac": "26.5.2",
+ "headset": "",
+ "scenarios": [
+  {
+   "scenario": "test",
+   "frames_sent": 902,
+   "frames_drawn": 902,
+   "frames_shown": 604,
+   "duration_s": 15.2,
+   "stages_ms": {
+    "capture": {
+     "p50": 0.0,
+     "p95": 0.0,
+     "p99": 0.0,
+     "n": 902
+    },
+    "queue": {
+     "p50": 0.0,
+     "p95": 0.1,
+     "p99": 0.1,
+     "n": 902
+    },
+    "encode": {
+     "p50": 4.3,
+     "p95": 5.0,
+     "p99": 5.3,
+     "n": 902
+    },
+    "socket": {
+     "p50": 0.1,
+     "p95": 0.2,
+     "p99": 0.3,
+     "n": 902
+    },
+    "network": {
+     "p50": 3.8,
+     "p95": 5.3,
+     "p99": 8.0,
+     "n": 902
+    },
+    "decode": {
+     "p50": 1.1,
+     "p95": 2.8,
+     "p99": 3.8,
+     "n": 902
+    },
+    "draw": {
+     "p50": 0.3,
+     "p95": 0.8,
+     "p99": 1.2,
+     "n": 902
+    },
+    "present": {
+     "p50": 5.9,
+     "p95": 17.0,
+     "p99": 18.6,
+     "n": 604
+    },
+    "content": {
+     "p50": 9.8,
+     "p95": 12.1,
+     "p99": 14.4,
+     "n": 902
+    },
+    "content_shown": {
+     "p50": 15.2,
+     "p95": 27.0,
+     "p99": 28.4,
+     "n": 604
+    }
+   },
+   "fps": 59.3,
+   "fps_shown": 39.7,
+   "late_pct": 0.55,
+   "stall_max": 187.5,
+   "stalls_over_100ms": 1,
+   "mbps": 0.47,
+   "keyframes": 1,
+   "size": "1280x720",
+   "captured": 913,
+   "viewer_never_drawn": 0,
+   "input_ms": {
+    "p50": 27.8,
+    "p95": 34.9,
+    "p99": 206.9,
+    "n": 30
+   },
+   "input_shown_ms": {
+    "p50": 35.1,
+    "p95": 52.6,
+    "p99": 207.2,
+    "n": 24
+   },
+   "input_parts_ms": {
+    "uplink": {
+     "p50": 9.8,
+     "p95": 22.5,
+     "p99": 184.8,
+     "n": 30
+    },
+    "mac": {
+     "p50": 5.3,
+     "p95": 14.4,
+     "p99": 19.6,
+     "n": 30
+    },
+    "back": {
+     "p50": 9.8,
+     "p95": 13.4,
+     "p99": 13.9,
+     "n": 30
+    }
+   },
+   "inputs": {
+    "asked": 30,
+    "sent": 30,
+    "seen": 30
+   },
+   "timeline": [
+    {
+     "t": 0,
+     "fps": 60,
+     "mbps": 0.48,
+     "content_p50": 9.3,
+     "content_p95": 11.2,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 1,
+     "fps": 50,
+     "mbps": 0.39,
+     "content_p50": 9.1,
+     "content_p95": 14.4,
+     "bitrate": 2764800,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 2,
+     "fps": 60,
+     "mbps": 0.48,
+     "content_p50": 9.2,
+     "content_p95": 11.8,
+     "bitrate": 2764800,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 3,
+     "fps": 60,
+     "mbps": 0.48,
+     "content_p50": 9.4,
+     "content_p95": 10.5,
+     "bitrate": 2764800,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 4,
+     "fps": 60,
+     "mbps": 0.48,
+     "content_p50": 9.5,
+     "content_p95": 12.5,
+     "bitrate": 3091280,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 5,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 10.0,
+     "content_p95": 12.8,
+     "bitrate": 4757991,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 6,
+     "fps": 60,
+     "mbps": 0.57,
+     "content_p50": 9.5,
+     "content_p95": 11.7,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 7,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 9.8,
+     "content_p95": 11.3,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 8,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 10.2,
+     "content_p95": 12.2,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 9,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 10.0,
+     "content_p95": 13.2,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 10,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 10.2,
+     "content_p95": 12.5,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 11,
+     "fps": 60,
+     "mbps": 0.45,
+     "content_p50": 9.9,
+     "content_p95": 11.6,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 12,
+     "fps": 60,
+     "mbps": 0.46,
+     "content_p50": 10.0,
+     "content_p95": 11.6,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 13,
+     "fps": 60,
+     "mbps": 0.47,
+     "content_p50": 10.3,
+     "content_p95": 12.3,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 14,
+     "fps": 60,
+     "mbps": 0.48,
+     "content_p50": 9.8,
+     "content_p95": 11.6,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 15,
+     "fps": 12,
+     "mbps": 0.1,
+     "content_p50": 9.3,
+     "content_p95": 11.5,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    }
+   ],
+   "adapt": [],
+   "content_p50": 9.8,
+   "content_p95": 12.1,
+   "input_p50": 27.8,
+   "input_p95": 34.9,
+   "grades": {
+    "input_replies": "local",
+    "content_p50": "local",
+    "content_p95": "local",
+    "input_p50": "local",
+    "input_p95": "local",
+    "fps": "local",
+    "late_pct": "local",
+    "stall_max": "acceptable"
+   },
+   "controller": {
+    "adapt": true,
+    "baseRtt": 5.084,
+    "ceiling": 5529600,
+    "fps": 60,
+    "inFlight": 1,
+    "scale": 1,
+    "slack": 49.603,
+    "target": 5529600,
+    "tier": 0
+   },
+   "events": [
+    {
+     "t": 1.77,
+     "e": "down to 2764 kbit/s: queue 23 ms, held 5, oldest 0 ms, base 5 ms, sent 315 got 315 wants 472"
+    }
+   ],
+   "source_fps": 60.1,
+   "cpu": {
+    "mac_agent_pct": 16.1,
+    "frame_viewer_pct": 40.9,
+    "frame_tailscaled_pct": 5.2,
+    "frame_sshd_pct": 0.7,
+    "frame_gamescope_pct": 8.4,
+    "frame_total_pct": 25.2
+   },
+   "viewer": "h264 software; ANGLE (Mesa, zink Vulkan 1.4(Turnip Adreno (TM) 750 (MESA_TURNIP)), OpenGL 4.6); raf 136 Hz",
+   "show_s": 1.27,
+   "panel": "valve.steam.desktopgame.2001639889",
+   "src": "test"
+  },
+  {
+   "scenario": "scroll",
+   "frames_sent": 832,
+   "frames_drawn": 832,
+   "frames_shown": 602,
+   "duration_s": 15.1,
+   "stages_ms": {
+    "capture": {
+     "p50": -3.9,
+     "p95": 0.8,
+     "p99": 6.8,
+     "n": 832
+    },
+    "queue": {
+     "p50": 0.0,
+     "p95": 0.1,
+     "p99": 1.9,
+     "n": 832
+    },
+    "encode": {
+     "p50": 6.8,
+     "p95": 10.5,
+     "p99": 18.8,
+     "n": 832
+    },
+    "socket": {
+     "p50": 0.1,
+     "p95": 0.2,
+     "p99": 0.5,
+     "n": 832
+    },
+    "network": {
+     "p50": 5.8,
+     "p95": 18.7,
+     "p99": 56.7,
+     "n": 832
+    },
+    "decode": {
+     "p50": 5.9,
+     "p95": 11.7,
+     "p99": 26.0,
+     "n": 832
+    },
+    "draw": {
+     "p50": 0.7,
+     "p95": 1.4,
+     "p99": 1.7,
+     "n": 832
+    },
+    "present": {
+     "p50": 4.4,
+     "p95": 18.5,
+     "p99": 25.0,
+     "n": 602
+    },
+    "content": {
+     "p50": 20.2,
+     "p95": 36.9,
+     "p99": 90.7,
+     "n": 832
+    },
+    "content_shown": {
+     "p50": 26.4,
+     "p95": 45.4,
+     "p99": 101.0,
+     "n": 602
+    }
+   },
+   "fps": 54.9,
+   "fps_shown": 39.7,
+   "late_pct": 6.96,
+   "stall_max": 204.9,
+   "stalls_over_100ms": 2,
+   "mbps": 9.04,
+   "keyframes": 1,
+   "size": "1920x1290",
+   "captured": 852,
+   "viewer_never_drawn": 0,
+   "input_ms": {
+    "n": 0
+   },
+   "input_shown_ms": {
+    "n": 0
+   },
+   "input_parts_ms": {
+    "uplink": {
+     "n": 0
+    },
+    "mac": {
+     "n": 0
+    },
+    "back": {
+     "n": 0
+    }
+   },
+   "inputs": {
+    "asked": 0,
+    "sent": 0,
+    "seen": 0
+   },
+   "timeline": [
+    {
+     "t": 0,
+     "fps": 49,
+     "mbps": 8.68,
+     "content_p50": 19.8,
+     "content_p95": 119.1,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 1,
+     "fps": 47,
+     "mbps": 6.06,
+     "content_p50": 19.8,
+     "content_p95": 88.6,
+     "bitrate": 7430400,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 2,
+     "fps": 52,
+     "mbps": 8.07,
+     "content_p50": 19.4,
+     "content_p95": 25.5,
+     "bitrate": 10055362,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 3,
+     "fps": 56,
+     "mbps": 9.79,
+     "content_p50": 21.4,
+     "content_p95": 29.1,
+     "bitrate": 13549185,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 4,
+     "fps": 57,
+     "mbps": 8.93,
+     "content_p50": 21.5,
+     "content_p95": 35.1,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 5,
+     "fps": 57,
+     "mbps": 10.59,
+     "content_p50": 21.1,
+     "content_p95": 35.4,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 6,
+     "fps": 56,
+     "mbps": 11.55,
+     "content_p50": 19.7,
+     "content_p95": 66.9,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 7,
+     "fps": 57,
+     "mbps": 8.8,
+     "content_p50": 18.8,
+     "content_p95": 24.8,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 8,
+     "fps": 57,
+     "mbps": 9.17,
+     "content_p50": 19.5,
+     "content_p95": 24.4,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 9,
+     "fps": 57,
+     "mbps": 9.39,
+     "content_p50": 19.4,
+     "content_p95": 23.1,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 10,
+     "fps": 57,
+     "mbps": 8.53,
+     "content_p50": 19.5,
+     "content_p95": 38.1,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 11,
+     "fps": 56,
+     "mbps": 9.23,
+     "content_p50": 20.7,
+     "content_p95": 30.9,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 12,
+     "fps": 53,
+     "mbps": 8.87,
+     "content_p50": 20.6,
+     "content_p95": 35.4,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 13,
+     "fps": 57,
+     "mbps": 8.94,
+     "content_p50": 20.4,
+     "content_p95": 28.2,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 14,
+     "fps": 57,
+     "mbps": 9.14,
+     "content_p50": 21.0,
+     "content_p95": 35.5,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 15,
+     "fps": 7,
+     "mbps": 1.2,
+     "content_p50": 21.8,
+     "content_p95": 23.1,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    }
+   ],
+   "adapt": [],
+   "content_p50": 20.2,
+   "content_p95": 36.9,
+   "input_p50": null,
+   "input_p95": null,
+   "grades": {
+    "content_p50": "local",
+    "content_p95": "local",
+    "fps": "acceptable",
+    "late_pct": "bad",
+    "stall_max": "acceptable"
+   },
+   "controller": {
+    "adapt": true,
+    "baseRtt": 6.833,
+    "ceiling": 14860800,
+    "fps": 60,
+    "inFlight": 2,
+    "scale": 1,
+    "slack": 44.791,
+    "target": 14860800,
+    "tier": 0
+   },
+   "events": [
+    {
+     "t": 1.22,
+     "e": "down to 7430 kbit/s: queue 74 ms, held 7, oldest 235 ms, base 6 ms, sent 5436 got 4487 wants 11082"
+    }
+   ],
+   "source_fps": 56.4,
+   "cpu": {
+    "mac_agent_pct": 8.8,
+    "frame_viewer_pct": 69.5,
+    "frame_tailscaled_pct": 9.5,
+    "frame_sshd_pct": 0.8,
+    "frame_gamescope_pct": 8.4,
+    "frame_total_pct": 31.8
+   },
+   "viewer": "h264 software; ANGLE (Mesa, zink Vulkan 1.4(Turnip Adreno (TM) 750 (MESA_TURNIP)), OpenGL 4.6); raf 139 Hz",
+   "show_s": 1.25,
+   "panel": "valve.steam.desktopgame.2001116820",
+   "src": "separate:8327"
+  }
+ ]
+}
@@ -0,0 +1,984 @@
+{
+ "label": "adapt-clean",
+ "date": "2026-09-28T17:13:59",
+ "commit": "a3c6e5c+dirty",
+ "config": {
+  "quality": "balanced",
+  "mode": "separate",
+  "net": "none",
+  "delay_ms": 0,
+  "buffer_ms": 250,
+  "host": "frame",
+  "ssh_opts": [],
+  "encoder": "",
+  "duration_s": 20.0,
+  "browser_flags": []
+ },
+ "frame_build": "20260925.6191901",
+ "mac": "26.5.2",
+ "headset": ":39.639829 [Info] - 0 - entering standby",
+ "scenarios": [
+  {
+   "scenario": "test",
+   "frames_sent": 653,
+   "frames_drawn": 653,
+   "frames_shown": 649,
+   "duration_s": 20.7,
+   "stages_ms": {
+    "capture": {
+     "p50": 0.0,
+     "p95": 0.0,
+     "p99": 0.0,
+     "n": 653
+    },
+    "queue": {
+     "p50": 11.2,
+     "p95": 16.1,
+     "p99": 17.1,
+     "n": 653
+    },
+    "encode": {
+     "p50": 3.5,
+     "p95": 4.2,
+     "p99": 4.4,
+     "n": 653
+    },
+    "socket": {
+     "p50": 0.1,
+     "p95": 0.2,
+     "p99": 0.2,
+     "n": 653
+    },
+    "network": {
+     "p50": 2.9,
+     "p95": 7.0,
+     "p99": 12.0,
+     "n": 653
+    },
+    "decode": {
+     "p50": 1.0,
+     "p95": 2.4,
+     "p99": 2.8,
+     "n": 653
+    },
+    "draw": {
+     "p50": 0.3,
+     "p95": 0.6,
+     "p99": 0.9,
+     "n": 653
+    },
+    "present": {
+     "p50": 0.5,
+     "p95": 0.9,
+     "p99": 5.9,
+     "n": 649
+    },
+    "content": {
+     "p50": 19.8,
+     "p95": 26.2,
+     "p99": 30.1,
+     "n": 653
+    },
+    "content_shown": {
+     "p50": 20.3,
+     "p95": 26.9,
+     "p99": 31.1,
+     "n": 649
+    }
+   },
+   "fps": 31.5,
+   "fps_shown": 31.3,
+   "late_pct": 98.01,
+   "stall_max": 144.0,
+   "stalls_over_100ms": 2,
+   "mbps": 0.13,
+   "keyframes": 2,
+   "size": "960x540",
+   "captured": 1336,
+   "viewer_never_drawn": 0,
+   "input_ms": {
+    "p50": 43.5,
+    "p95": 61.8,
+    "p99": 63.4,
+    "n": 40
+   },
+   "input_shown_ms": {
+    "p50": 44.3,
+    "p95": 62.3,
+    "p99": 63.9,
+    "n": 40
+   },
+   "input_parts_ms": {
+    "uplink": {
+     "p50": 11.1,
+     "p95": 29.2,
+     "p99": 36.1,
+     "n": 40
+    },
+    "mac": {
+     "p50": 10.4,
+     "p95": 28.2,
+     "p99": 32.4,
+     "n": 40
+    },
+    "back": {
+     "p50": 17.5,
+     "p95": 27.9,
+     "p99": 28.8,
+     "n": 40
+    }
+   },
+   "inputs": {
+    "sent": 40,
+    "seen": 40
+   },
+   "timeline": [
+    {
+     "t": 0,
+     "fps": 31,
+     "mbps": 0.13,
+     "content_p50": 21.5,
+     "content_p95": 27.1,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 1,
+     "fps": 31,
+     "mbps": 0.12,
+     "content_p50": 22.4,
+     "content_p95": 26.3,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 2,
+     "fps": 32,
+     "mbps": 0.13,
+     "content_p50": 18.9,
+     "content_p95": 24.8,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 3,
+     "fps": 31,
+     "mbps": 0.12,
+     "content_p50": 18.2,
+     "content_p95": 24.6,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 4,
+     "fps": 32,
+     "mbps": 0.13,
+     "content_p50": 21.0,
+     "content_p95": 25.6,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 5,
+     "fps": 32,
+     "mbps": 0.12,
+     "content_p50": 19.5,
+     "content_p95": 24.3,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 6,
+     "fps": 31,
+     "mbps": 0.12,
+     "content_p50": 22.2,
+     "content_p95": 25.6,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 7,
+     "fps": 32,
+     "mbps": 0.13,
+     "content_p50": 19.9,
+     "content_p95": 24.0,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 8,
+     "fps": 31,
+     "mbps": 0.21,
+     "content_p50": 22.7,
+     "content_p95": 26.9,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 9,
+     "fps": 31,
+     "mbps": 0.12,
+     "content_p50": 20.2,
+     "content_p95": 26.1,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 10,
+     "fps": 32,
+     "mbps": 0.12,
+     "content_p50": 18.2,
+     "content_p95": 23.2,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 11,
+     "fps": 31,
+     "mbps": 0.12,
+     "content_p50": 19.8,
+     "content_p95": 24.2,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 12,
+     "fps": 32,
+     "mbps": 0.13,
+     "content_p50": 19.2,
+     "content_p95": 24.8,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 13,
+     "fps": 32,
+     "mbps": 0.12,
+     "content_p50": 21.0,
+     "content_p95": 24.3,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 14,
+     "fps": 30,
+     "mbps": 0.13,
+     "content_p50": 20.4,
+     "content_p95": 96.7,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 15,
+     "fps": 32,
+     "mbps": 0.13,
+     "content_p50": 18.2,
+     "content_p95": 24.4,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 16,
+     "fps": 31,
+     "mbps": 0.12,
+     "content_p50": 18.4,
+     "content_p95": 25.1,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 17,
+     "fps": 32,
+     "mbps": 0.13,
+     "content_p50": 17.2,
+     "content_p95": 23.5,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 18,
+     "fps": 32,
+     "mbps": 0.19,
+     "content_p50": 19.4,
+     "content_p95": 25.2,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 19,
+     "fps": 31,
+     "mbps": 0.14,
+     "content_p50": 21.1,
+     "content_p95": 84.3,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 20,
+     "fps": 24,
+     "mbps": 0.08,
+     "content_p50": 20.8,
+     "content_p95": 25.1,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    }
+   ],
+   "adapt": [],
+   "content_p50": 19.8,
+   "content_p95": 26.2,
+   "input_p50": 43.5,
+   "input_p95": 61.8,
+   "grades": {
+    "content_p50": "local",
+    "content_p95": "local",
+    "input_p50": "local",
+    "input_p95": "local",
+    "fps": "bad",
+    "late_pct": "bad",
+    "stall_max": "acceptable"
+   },
+   "source_fps": 64.5,
+   "cpu": {
+    "mac_agent_pct": 11.3,
+    "frame_viewer_pct": 35.9,
+    "frame_tailscaled_pct": 3.5,
+    "frame_sshd_pct": 0.6,
+    "frame_gamescope_pct": 7.5,
+    "frame_total_pct": 16.1
+   },
+   "viewer": "h264 software; ANGLE (Mesa, zink Vulkan 1.4(Turnip Adreno (TM) 750 (MESA_TURNIP)), OpenGL 4.6); raf 141 Hz",
+   "show_s": 4.68,
+   "panel": "valve.steam.desktopgame.2001639889",
+   "src": "test"
+  },
+  {
+   "scenario": "scroll",
+   "frames_sent": 937,
+   "frames_drawn": 937,
+   "frames_shown": 720,
+   "duration_s": 20.7,
+   "stages_ms": {
+    "capture": {
+     "p50": -3.9,
+     "p95": 0.9,
+     "p99": 8.2,
+     "n": 937
+    },
+    "queue": {
+     "p50": 0.0,
+     "p95": 16.3,
+     "p99": 20.1,
+     "n": 937
+    },
+    "encode": {
+     "p50": 6.4,
+     "p95": 7.2,
+     "p99": 9.3,
+     "n": 937
+    },
+    "socket": {
+     "p50": 0.1,
+     "p95": 0.2,
+     "p99": 0.2,
+     "n": 937
+    },
+    "network": {
+     "p50": 7.3,
+     "p95": 13.0,
+     "p99": 22.8,
+     "n": 937
+    },
+    "decode": {
+     "p50": 6.2,
+     "p95": 12.7,
+     "p99": 18.6,
+     "n": 937
+    },
+    "draw": {
+     "p50": 0.6,
+     "p95": 1.1,
+     "p99": 1.9,
+     "n": 937
+    },
+    "present": {
+     "p50": 2.9,
+     "p95": 18.5,
+     "p99": 24.5,
+     "n": 720
+    },
+    "content": {
+     "p50": 19.9,
+     "p95": 37.2,
+     "p99": 50.2,
+     "n": 937
+    },
+    "content_shown": {
+     "p50": 27.0,
+     "p95": 45.7,
+     "p99": 55.0,
+     "n": 720
+    }
+   },
+   "fps": 45.3,
+   "fps_shown": 34.8,
+   "late_pct": 28.63,
+   "stall_max": 99.3,
+   "stalls_over_100ms": 0,
+   "mbps": 7.59,
+   "keyframes": 2,
+   "size": "1920x1290",
+   "captured": 1215,
+   "viewer_never_drawn": 0,
+   "input_ms": {
+    "n": 0
+   },
+   "input_shown_ms": {
+    "n": 0
+   },
+   "input_parts_ms": {
+    "uplink": {
+     "n": 0
+    },
+    "mac": {
+     "n": 0
+    },
+    "back": {
+     "n": 0
+    }
+   },
+   "inputs": {
+    "sent": 0,
+    "seen": 0
+   },
+   "timeline": [
+    {
+     "t": 0,
+     "fps": 30,
+     "mbps": 4.01,
+     "content_p50": 31.9,
+     "content_p95": 52.3,
+     "bitrate": 4867140,
+     "tier": 2,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 1,
+     "fps": 30,
+     "mbps": 4.12,
+     "content_p50": 29.9,
+     "content_p95": 45.3,
+     "bitrate": 3807054,
+     "tier": 2,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 2,
+     "fps": 30,
+     "mbps": 3.26,
+     "content_p50": 31.8,
+     "content_p95": 52.5,
+     "bitrate": 3235995,
+     "tier": 2,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 3,
+     "fps": 30,
+     "mbps": 3.72,
+     "content_p50": 24.5,
+     "content_p95": 37.7,
+     "bitrate": 4238740,
+     "tier": 2,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 4,
+     "fps": 30,
+     "mbps": 5.09,
+     "content_p50": 25.2,
+     "content_p95": 39.6,
+     "bitrate": 5992063,
+     "tier": 2,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 5,
+     "fps": 31,
+     "mbps": 7.29,
+     "content_p50": 29.1,
+     "content_p95": 53.5,
+     "bitrate": 5526067,
+     "tier": 2,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 6,
+     "fps": 35,
+     "mbps": 4.58,
+     "content_p50": 25.3,
+     "content_p95": 32.4,
+     "bitrate": 6018152,
+     "tier": 1,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 7,
+     "fps": 46,
+     "mbps": 7.12,
+     "content_p50": 25.1,
+     "content_p95": 34.3,
+     "bitrate": 8412933,
+     "tier": 1,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 8,
+     "fps": 46,
+     "mbps": 8.25,
+     "content_p50": 27.1,
+     "content_p95": 40.3,
+     "bitrate": 10760191,
+     "tier": 1,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 9,
+     "fps": 47,
+     "mbps": 9.3,
+     "content_p50": 27.8,
+     "content_p95": 41.2,
+     "bitrate": 13717060,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 10,
+     "fps": 55,
+     "mbps": 8.98,
+     "content_p50": 16.5,
+     "content_p95": 21.7,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 11,
+     "fps": 51,
+     "mbps": 9.32,
+     "content_p50": 18.6,
+     "content_p95": 35.4,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 12,
+     "fps": 56,
+     "mbps": 9.53,
+     "content_p50": 15.4,
+     "content_p95": 26.2,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 13,
+     "fps": 55,
+     "mbps": 8.46,
+     "content_p50": 18.6,
+     "content_p95": 23.2,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 14,
+     "fps": 54,
+     "mbps": 9.82,
+     "content_p50": 15.9,
+     "content_p95": 28.4,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 15,
+     "fps": 54,
+     "mbps": 11.11,
+     "content_p50": 15.5,
+     "content_p95": 26.1,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 16,
+     "fps": 54,
+     "mbps": 9.79,
+     "content_p50": 17.4,
+     "content_p95": 25.3,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 17,
+     "fps": 57,
+     "mbps": 9.3,
+     "content_p50": 16.3,
+     "content_p95": 23.5,
+     "bitrate": 14860800,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 18,
+     "fps": 55,
+     "mbps": 8.3,
+     "content_p50": 19.5,
+     "content_p95": 31.0,
+     "bitrate": 7984791,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 19,
+     "fps": 51,
+     "mbps": 8.56,
+     "content_p50": 16.4,
+     "content_p95": 29.1,
+     "bitrate": 10220855,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    },
+    {
+     "t": 20,
+     "fps": 40,
+     "mbps": 7.13,
+     "content_p50": 18.3,
+     "content_p95": 23.9,
+     "bitrate": 12025604,
+     "tier": 0,
+     "w": 1920,
+     "net": null
+    }
+   ],
+   "adapt": [],
+   "content_p50": 19.9,
+   "content_p95": 37.2,
+   "input_p50": null,
+   "input_p95": null,
+   "grades": {
+    "content_p50": "local",
+    "content_p95": "local",
+    "fps": "acceptable",
+    "late_pct": "bad",
+    "stall_max": "local"
+   },
+   "source_fps": 58.7,
+   "cpu": {
+    "mac_agent_pct": 6.6,
+    "frame_viewer_pct": 72.4,
+    "frame_tailscaled_pct": 8.1,
+    "frame_sshd_pct": 0.9,
+    "frame_gamescope_pct": 10.0,
+    "frame_total_pct": 24.2
+   },
+   "viewer": "h264 software; ANGLE (Mesa, zink Vulkan 1.4(Turnip Adreno (TM) 750 (MESA_TURNIP)), OpenGL 4.6); raf 141 Hz",
+   "show_s": 6.43,
+   "panel": "valve.steam.desktopgame.2001414593",
+   "src": "separate:9384"
+  },
+  {
+   "scenario": "type",
+   "frames_sent": 25,
+   "frames_drawn": 25,
+   "frames_shown": 25,
+   "duration_s": 14.7,
+   "stages_ms": {
+    "capture": {
+     "p50": -0.3,
+     "p95": 0.9,
+     "p99": 1.1,
+     "n": 25
+    },
+    "queue": {
+     "p50": 0.0,
+     "p95": 0.0,
+     "p99": 0.0,
+     "n": 25
+    },
+    "encode": {
+     "p50": 5.0,
+     "p95": 33.4,
+     "p99": 33.6,
+     "n": 25
+    },
+    "socket": {
+     "p50": 0.1,
+     "p95": 0.1,
+     "p99": 0.2,
+     "n": 25
+    },
+    "network": {
+     "p50": 6.1,
+     "p95": 14.9,
+     "p99": 24.4,
+     "n": 25
+    },
+    "decode": {
+     "p50": 3.0,
+     "p95": 10.5,
+     "p99": 11.4,
+     "n": 25
+    },
+    "draw": {
+     "p50": 0.6,
+     "p95": 1.0,
+     "p99": 1.1,
+     "n": 25
+    },
+    "present": {
+     "p50": 0.8,
+     "p95": 1.4,
+     "p99": 1.7,
+     "n": 25
+    },
+    "content": {
+     "p50": 12.6,
+     "p95": 41.0,
+     "p99": 46.3,
+     "n": 25
+    },
+    "content_shown": {
+     "p50": 13.9,
+     "p95": 41.8,
+     "p99": 46.6,
+     "n": 25
+    }
+   },
+   "fps": 1.7,
+   "fps_shown": 1.7,
+   "late_pct": 100.0,
+   "stall_max": 3099.2,
+   "stalls_over_100ms": 21,
+   "mbps": 0.04,
+   "keyframes": 4,
+   "size": "960x646",
+   "captured": 79,
+   "viewer_never_drawn": 0,
+   "input_ms": {
+    "p50": 41.3,
+    "p95": 60.3,
+    "p99": 68.6,
+    "n": 18
+   },
+   "input_shown_ms": {
+    "p50": 43.0,
+    "p95": 61.1,
+    "p99": 69.6,
+    "n": 18
+   },
+   "input_parts_ms": {
+    "uplink": {
+     "p50": 5.3,
+     "p95": 11.6,
+     "p99": 14.2,
+     "n": 18
+    },
+    "mac": {
+     "p50": 23.1,
+     "p95": 35.0,
+     "p99": 42.6,
+     "n": 18
+    },
+    "back": {
+     "p50": 11.8,
+     "p95": 19.2,
+     "p99": 22.7,
+     "n": 18
+    }
+   },
+   "inputs": {
+    "sent": 66,
+    "seen": 18
+   },
+   "timeline": [
+    {
+     "t": 3,
+     "fps": 1,
+     "mbps": 0.01,
+     "content_p50": 11.8,
+     "content_p95": 11.8,
+     "bitrate": 1224075,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 4,
+     "fps": 1,
+     "mbps": 0.02,
+     "content_p50": 14.7,
+     "content_p95": 14.7,
+     "bitrate": 2091896,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 5,
+     "fps": 4,
+     "mbps": 0.1,
+     "content_p50": 18.6,
+     "content_p95": 37.4,
+     "bitrate": 3071304,
+     "tier": 3,
+     "w": 1440,
+     "net": null
+    },
+    {
+     "t": 6,
+     "fps": 5,
+     "mbps": 0.13,
+     "content_p50": 4.2,
+     "content_p95": 19.2,
+     "bitrate": 4031277,
+     "tier": 3,
+     "w": 1440,
+     "net": null
+    },
+    {
+     "t": 7,
+     "fps": 6,
+     "mbps": 0.04,
+     "content_p50": 12.0,
+     "content_p95": 46.3,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 10,
+     "fps": 1,
+     "mbps": 0.01,
+     "content_p50": 12.2,
+     "content_p95": 12.2,
+     "bitrate": 300000,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 12,
+     "fps": 1,
+     "mbps": 0.02,
+     "content_p50": -1.2,
+     "content_p95": -1.2,
+     "bitrate": 842857,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    },
+    {
+     "t": 14,
+     "fps": 1,
+     "mbps": 0.02,
+     "content_p50": 14.4,
+     "content_p95": 14.4,
+     "bitrate": 2091896,
+     "tier": 4,
+     "w": 960,
+     "net": null
+    }
+   ],
+   "adapt": [],
+   "content_p50": 12.6,
+   "content_p95": 41.0,
+   "input_p50": 41.3,
+   "input_p95": 60.3,
+   "grades": {
+    "content_p50": "local",
+    "content_p95": "acceptable",
+    "input_p50": "local",
+    "input_p95": "local"
+   },
+   "source_fps": 5.4,
+   "cpu": {
+    "mac_agent_pct": 1.2,
+    "frame_viewer_pct": 5.0,
+    "frame_tailscaled_pct": 1.5,
+    "frame_sshd_pct": 0.2,
+    "frame_gamescope_pct": 4.7,
+    "frame_total_pct": 8.5
+   },
+   "viewer": "h264 software; ANGLE (Mesa, zink Vulkan 1.4(Turnip Adreno (TM) 750 (MESA_TURNIP)), OpenGL 4.6); raf 140 Hz",
+   "show_s": 5.12,
+   "panel": "valve.steam.desktopgame.2001910179",
+   "src": "separate:9510"
+  }
+ ]
+}
@@ -0,0 +1,311 @@
+{
+ "label": "pc-agent-frame-arm64-hairpin",
+ "date": "2026-09-28T22:36:19",
+ "commit": "c5fc552+dirty",
+ "config": {
+  "quality": "balanced",
+  "mode": "separate",
+  "net": "none",
+  "delay_ms": 0,
+  "buffer_ms": 250,
+  "host": "frame",
+  "usb": false,
+  "ssh_opts": [],
+  "encoder": "",
+  "duration_s": 10.0,
+  "browser_flags": []
+ },
+ "frame_build": "20260925.6191901",
+ "mac": "26.5.2",
+ "host_platform": "Linux aarch64, SteamOS Frame, BUILD_ID 20260925.6191901 (synthetic host only)",
+ "source": "",
+ "pc_host": true,
+ "headset": "0:03.056203 [Info] - 1 - leaving standby",
+ "scenarios": [
+  {
+   "scenario": "test",
+   "frames_sent": 351,
+   "frames_drawn": 351,
+   "frames_shown": 324,
+   "duration_s": 10.1,
+   "stages_ms": {
+    "capture": {
+     "p50": 0.1,
+     "p95": 0.2,
+     "p99": 0.5,
+     "n": 351
+    },
+    "queue": {
+     "p50": 0.1,
+     "p95": 0.3,
+     "p99": 0.4,
+     "n": 351
+    },
+    "encode": {
+     "p50": 2.0,
+     "p95": 2.9,
+     "p99": 3.5,
+     "n": 351
+    },
+    "socket": {
+     "p50": 0.1,
+     "p95": 0.4,
+     "p99": 0.6,
+     "n": 351
+    },
+    "network": {
+     "p50": 11.6,
+     "p95": 75.7,
+     "p99": 130.1,
+     "n": 351
+    },
+    "decode": {
+     "p50": 0.7,
+     "p95": 2.1,
+     "p99": 4.1,
+     "n": 351
+    },
+    "draw": {
+     "p50": 0.4,
+     "p95": 0.8,
+     "p99": 1.1,
+     "n": 351
+    },
+    "present": {
+     "p50": 0.8,
+     "p95": 18.1,
+     "p99": 24.9,
+     "n": 324
+    },
+    "content": {
+     "p50": 15.1,
+     "p95": 80.3,
+     "p99": 134.2,
+     "n": 351
+    },
+    "content_shown": {
+     "p50": 21.0,
+     "p95": 71.8,
+     "p99": 146.3,
+     "n": 324
+    }
+   },
+   "fps": 34.8,
+   "fps_shown": 32.0,
+   "late_pct": 53.41,
+   "stall_max": 226.8,
+   "stalls_over_100ms": 7,
+   "mbps": 0.26,
+   "keyframes": 6,
+   "size": "1280x720",
+   "captured": 607,
+   "viewer_never_drawn": 0,
+   "input_ms": {
+    "p50": 43.3,
+    "p95": 98.4,
+    "p99": 110.1,
+    "n": 18
+   },
+   "input_shown_ms": {
+    "p50": 54.7,
+    "p95": 98.9,
+    "p99": 114.0,
+    "n": 18
+   },
+   "input_parts_ms": {
+    "uplink": {
+     "p50": 17.1,
+     "p95": 48.3,
+     "p99": 53.1,
+     "n": 18
+    },
+    "mac": {
+     "p50": 12.5,
+     "p95": 32.9,
+     "p99": 35.7,
+     "n": 18
+    },
+    "back": {
+     "p50": 16.8,
+     "p95": 50.0,
+     "p99": 55.9,
+     "n": 18
+    }
+   },
+   "inputs": {
+    "asked": 18,
+    "sent": 18,
+    "seen": 18
+   },
+   "timeline": [
+    {
+     "t": 0,
+     "fps": 38,
+     "mbps": 0.3,
+     "content_p50": 14.8,
+     "content_p95": 25.3,
+     "bitrate": 2764800,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 1,
+     "fps": 33,
+     "mbps": 0.24,
+     "content_p50": 15.6,
+     "content_p95": 103.5,
+     "bitrate": 3845448,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 2,
+     "fps": 39,
+     "mbps": 0.28,
+     "content_p50": 15.6,
+     "content_p95": 29.9,
+     "bitrate": 5283790,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 3,
+     "fps": 25,
+     "mbps": 0.21,
+     "content_p50": 17.8,
+     "content_p95": 175.9,
+     "bitrate": 2764800,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 4,
+     "fps": 34,
+     "mbps": 0.22,
+     "content_p50": 15.4,
+     "content_p95": 69.8,
+     "bitrate": 2764800,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 5,
+     "fps": 38,
+     "mbps": 0.29,
+     "content_p50": 17.7,
+     "content_p95": 68.7,
+     "bitrate": 2764800,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 6,
+     "fps": 38,
+     "mbps": 0.26,
+     "content_p50": 15.0,
+     "content_p95": 18.2,
+     "bitrate": 3091280,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 7,
+     "fps": 30,
+     "mbps": 0.22,
+     "content_p50": 19.2,
+     "content_p95": 111.0,
+     "bitrate": 1725204,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 8,
+     "fps": 35,
+     "mbps": 0.26,
+     "content_p50": 13.4,
+     "content_p95": 102.6,
+     "bitrate": 1725204,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 9,
+     "fps": 38,
+     "mbps": 0.28,
+     "content_p50": 12.3,
+     "content_p95": 20.9,
+     "bitrate": 1725204,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 10,
+     "fps": 3,
+     "mbps": 0.02,
+     "content_p50": 20.9,
+     "content_p95": 24.7,
+     "bitrate": 1725204,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    }
+   ],
+   "adapt": [],
+   "content_p50": 15.1,
+   "content_p95": 80.3,
+   "input_p50": 43.3,
+   "input_p95": 98.4,
+   "grades": {
+    "input_replies": "local",
+    "content_p50": "local",
+    "content_p95": "acceptable",
+    "input_p50": "local",
+    "input_p95": "acceptable",
+    "fps": "bad",
+    "late_pct": "bad",
+    "stall_max": "acceptable"
+   },
+   "controller": {
+    "target": 3442081,
+    "ceiling": 5529600,
+    "tier": 0,
+    "fps": 60,
+    "scale": 1.0,
+    "baseRtt": 13.334,
+    "inFlight": 1,
+    "slack": 71.738,
+    "adapt": true
+   },
+   "events": [
+    {
+     "t": 3.77,
+     "e": "target 2764800 bit/s; tier 0"
+    },
+    {
+     "t": 7.42,
+     "e": "target 1725204 bit/s; tier 0"
+    }
+   ],
+   "source_fps": 60.1,
+   "cpu": {
+    "host": "not sampled"
+   },
+   "viewer": "h264 software; ANGLE (Mesa, zink Vulkan 1.4(Turnip Adreno (TM) 750 (MESA_TURNIP)), OpenGL 4.6); raf 71 Hz",
+   "show_s": 1.88,
+   "panel": "valve.steam.desktopgame.2001932457",
+   "src": "test",
+   "route": "network"
+  }
+ ],
+ "proof": "Synthetic x264 PC agent runs on the Frame itself, forwarded via Mac and back. Not Windows/Linux desktop capture or laser input."
+}
@@ -0,0 +1,302 @@
+{
+ "label": "pc-agent-frame-arm64-final",
+ "date": "2026-09-28T22:55:27",
+ "commit": "cd20243",
+ "config": {
+  "quality": "balanced",
+  "mode": "native",
+  "net": "none",
+  "delay_ms": 0,
+  "buffer_ms": 250,
+  "host": "frame",
+  "usb": false,
+  "ssh_opts": [],
+  "encoder": "",
+  "duration_s": 10.0,
+  "browser_flags": []
+ },
+ "frame_build": "20260925.6191901",
+ "mac": "26.5.2",
+ "host_platform": "Linux aarch64, SteamOS Frame, BUILD_ID 20260925.6191901 (synthetic host only)",
+ "source": "",
+ "pc_host": true,
+ "headset": "0:03.056203 [Info] - 1 - leaving standby",
+ "scenarios": [
+  {
+   "scenario": "test",
+   "frames_sent": 400,
+   "frames_drawn": 400,
+   "frames_shown": 339,
+   "duration_s": 10.2,
+   "stages_ms": {
+    "capture": {
+     "p50": 0.1,
+     "p95": 0.1,
+     "p99": 0.2,
+     "n": 400
+    },
+    "queue": {
+     "p50": 0.1,
+     "p95": 0.2,
+     "p99": 0.4,
+     "n": 400
+    },
+    "encode": {
+     "p50": 2.0,
+     "p95": 2.7,
+     "p99": 3.1,
+     "n": 400
+    },
+    "socket": {
+     "p50": 0.1,
+     "p95": 0.3,
+     "p99": 0.4,
+     "n": 400
+    },
+    "network": {
+     "p50": 11.8,
+     "p95": 25.2,
+     "p99": 37.8,
+     "n": 400
+    },
+    "decode": {
+     "p50": 0.7,
+     "p95": 1.5,
+     "p99": 2.0,
+     "n": 400
+    },
+    "draw": {
+     "p50": 0.3,
+     "p95": 0.7,
+     "p99": 1.0,
+     "n": 400
+    },
+    "present": {
+     "p50": 0.8,
+     "p95": 19.3,
+     "p99": 26.8,
+     "n": 339
+    },
+    "content": {
+     "p50": 15.1,
+     "p95": 28.6,
+     "p99": 41.3,
+     "n": 400
+    },
+    "content_shown": {
+     "p50": 17.6,
+     "p95": 39.0,
+     "p99": 47.2,
+     "n": 339
+    }
+   },
+   "fps": 39.3,
+   "fps_shown": 33.4,
+   "late_pct": 50.25,
+   "stall_max": 69.3,
+   "stalls_over_100ms": 0,
+   "mbps": 0.35,
+   "keyframes": 6,
+   "size": "1280x720",
+   "captured": 612,
+   "viewer_never_drawn": 0,
+   "input_ms": {
+    "p50": 76.6,
+    "p95": 98.2,
+    "p99": 127.8,
+    "n": 19
+   },
+   "input_shown_ms": {
+    "p50": 80.5,
+    "p95": 107.8,
+    "p99": 128.1,
+    "n": 16
+   },
+   "input_parts_ms": {
+    "uplink": {
+     "p50": 36.5,
+     "p95": 57.6,
+     "p99": 83.2,
+     "n": 19
+    },
+    "mac": {
+     "p50": 14.6,
+     "p95": 31.3,
+     "p99": 31.7,
+     "n": 19
+    },
+    "back": {
+     "p50": 15.1,
+     "p95": 32.8,
+     "p99": 36.2,
+     "n": 19
+    }
+   },
+   "inputs": {
+    "asked": 19,
+    "sent": 19,
+    "seen": 19
+   },
+   "timeline": [
+    {
+     "t": 0,
+     "fps": 40,
+     "mbps": 0.34,
+     "content_p50": 11.7,
+     "content_p95": 27.4,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 1,
+     "fps": 39,
+     "mbps": 0.37,
+     "content_p50": 11.9,
+     "content_p95": 28.2,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 2,
+     "fps": 41,
+     "mbps": 0.35,
+     "content_p50": 11.4,
+     "content_p95": 16.6,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 3,
+     "fps": 39,
+     "mbps": 0.37,
+     "content_p50": 12.7,
+     "content_p95": 28.0,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 4,
+     "fps": 38,
+     "mbps": 0.33,
+     "content_p50": 15.3,
+     "content_p95": 30.0,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 5,
+     "fps": 39,
+     "mbps": 0.37,
+     "content_p50": 16.7,
+     "content_p95": 28.5,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 6,
+     "fps": 41,
+     "mbps": 0.38,
+     "content_p50": 15.9,
+     "content_p95": 29.4,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 7,
+     "fps": 41,
+     "mbps": 0.31,
+     "content_p50": 15.6,
+     "content_p95": 28.2,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 8,
+     "fps": 38,
+     "mbps": 0.38,
+     "content_p50": 15.5,
+     "content_p95": 25.9,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 9,
+     "fps": 38,
+     "mbps": 0.29,
+     "content_p50": 16.7,
+     "content_p95": 28.8,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    },
+    {
+     "t": 10,
+     "fps": 6,
+     "mbps": 0.03,
+     "content_p50": 17.6,
+     "content_p95": 23.1,
+     "bitrate": 5529600,
+     "tier": 0,
+     "w": 1280,
+     "net": null
+    }
+   ],
+   "adapt": [],
+   "content_p50": 15.1,
+   "content_p95": 28.6,
+   "input_p50": 76.6,
+   "input_p95": 98.2,
+   "grades": {
+    "input_replies": "local",
+    "content_p50": "local",
+    "content_p95": "local",
+    "input_p50": "acceptable",
+    "input_p95": "acceptable",
+    "fps": "bad",
+    "late_pct": "bad",
+    "stall_max": "local"
+   },
+   "controller": {
+    "target": 5529600,
+    "ceiling": 5529600,
+    "tier": 0,
+    "fps": 60,
+    "scale": 1.0,
+    "baseRtt": 12.332,
+    "inFlight": 0,
+    "slack": 74.12,
+    "adapt": true
+   },
+   "events": [],
+   "source_fps": 60.0,
+   "cpu": {
+    "host": "not sampled"
+   },
+   "viewer": "h264 software; ANGLE (Mesa, zink Vulkan 1.4(Turnip Adreno (TM) 750 (MESA_TURNIP)), OpenGL 4.6); raf 72 Hz",
+   "show_s": 1.75,
+   "panel": "valve.steam.desktopgame.2001932457",
+   "src": "test",
+   "route": "network"
+  }
+ ],
+ "proof": "Synthetic x264 PC agent runs on the Frame itself, forwarded via Mac and back. Not Windows/Linux desktop capture or laser input."
+}
@@ -1,9 +1,32 @@
 # compat-db: Frame Control's compatibility database

 A private [Lakebed](https://docs.lakebed.dev/) capsule holding compatibility
-reports for Android apps on the Steam Frame. For now only the maintainer's
-copy of Frame Control has the key to read or write it. Everyone else's reports
-stay on their own Mac (see `shared()` in `ui/frame_compat_db.py`).
+reports for Android apps on the Steam Frame. Only the maintainer's copy of
+Frame Control has the key to read or write it (see `shared()` in
+`ui/frame_compat_db.py`). Everyone else's reports stay on their computer
+unless they turn on **Share compatibility results**. Then the reports also go
+to PostHog as `compat_report` events, and the maintainer syncs them in (below).
+
+## Community reports
+
+```sh
+python3 ui/frame_compat_db.py sync --dry-run   # what would be added
+python3 ui/frame_compat_db.py sync             # add them
+```
+
+`sync` reads `compat_report` events through PostHog's query API and adds
+them with `via` set to `community`, `community-probe` or `community-install`.
+It skips invalid reports and anything over 30 per reporter per day. Each run
+re-reads the last 30 days, because an offline copy sends its reports late,
+with the time they were made. `posthog-sync.json`, next to the outbox,
+remembers which reports it has handled and each reporter's daily count, so
+nothing is added twice and the cap holds across runs.
+
+It needs:
+
+- the PostHog project id: `"project"` in `ui/telemetry.json`
+- a personal API key with `query:read`: `POSTHOG_PERSONAL_API_KEY`, or in the
+  Keychain (service `frame-control-posthog`, account `personal-api-key`)

 - Live: `https://frame-compat.lakebed.app` (deploy `dep_dDmcsosVSiFirpW6`,
  claimed, so it doesn't expire). The browser page only says it's private.
@@ -0,0 +1,16 @@
+#!/usr/bin/env python3
+"""Build the common controller for the Python PC host (a C11 compiler needed)."""
+import os
+from pathlib import Path
+import subprocess
+import sys
+
+root = Path(__file__).resolve().parent
+suffix = '.dll' if sys.platform == 'win32' else '.dylib' if sys.platform == 'darwin' else '.so'
+out = root / ('controller' + suffix)
+if sys.platform == 'win32' and os.environ.get('CC', 'cl') == 'cl':
+    cmd = ['cl', '/nologo', '/O2', '/LD', '/std:c11', str(root / 'controller.c'), '/link', '/OUT:' + str(out)]
+else:
+    cmd = [os.environ.get('CC', 'cc'), '-O2', '-std=c11', '-shared', '-fPIC', str(root / 'controller.c'), '-o', str(out)]
+subprocess.run(cmd, cwd=root, check=True)
+print(out)
@@ -0,0 +1,126 @@
+#!/usr/bin/env python3
+"""Build our native PC adapter and bundle its ordinary shared libraries.
+
+Run on the target architecture after installing GStreamer development packages.
+No GStreamer executable or desktop streaming application is shipped or invoked.
+Linux libc, display/GPU drivers and the portal service remain platform pieces.
+
+x264 is GPL, so the bundle the Windows installer ships leaves it out: Windows
+encodes with Media Foundation. --with-x264 adds it for CI's headless protocol
+tests, which have no GPU or desktop to encode with.
+"""
+import argparse
+import json
+import os
+from pathlib import Path
+import re
+import shutil
+import subprocess
+import sys
+
+ROOT = Path(__file__).resolve().parent
+
+
+def output(*args):
+    return subprocess.check_output(args, text=True).strip()
+
+
+def build(destination, with_x264=False):
+    win = sys.platform == 'win32'
+    destination.mkdir(parents=True, exist_ok=True)
+    modules = ['gstreamer-1.0', 'gstreamer-app-1.0', 'gstreamer-video-1.0']
+    if not win:
+        modules += ['gio-unix-2.0']
+    import shlex
+    flags = shlex.split(output('pkg-config', '--cflags', '--libs', *modules))
+    lib = destination / ('pc-host.dll' if win else 'pc-host.so')
+    cmd = [os.environ.get('CC', 'gcc' if win else 'cc'), '-std=c11', '-O2', '-shared', '-Wall', '-Wextra']
+    if not win:
+        cmd += ['-fPIC', '-Wl,-rpath,$ORIGIN/lib']
+    subprocess.run(cmd + [str(ROOT / f) for f in ('controller.c', 'capture.c', 'portal.c')] + flags + ['-o', str(lib)], check=True)
+    prefix = Path(output('pkg-config', '--variable=prefix', 'gstreamer-1.0'))
+    plugin_dir = Path(output('pkg-config', '--variable=pluginsdir', 'gstreamer-1.0'))
+    plugins_out, libs_out = destination / 'lib' / 'gstreamer-1.0', destination / ('bin' if win else 'lib')
+    plugins_out.mkdir(parents=True, exist_ok=True)
+    libs_out.mkdir(parents=True, exist_ok=True)
+    names = ['coreelements', 'videotestsrc', 'videoconvertscale', 'app', 'jpeg', 'videoparsersbad']
+    if with_x264:
+        names.append('x264')
+    names += ['d3d11', 'mediafoundation'] if win else ['pipewire', 'va']
+    pending = [lib]
+    for name in names:
+        matches = list(plugin_dir.glob('*gst' + name + ('.dll' if win else '.so')))
+        if not matches:
+            raise SystemExit('Missing GStreamer library: ' + name)
+        for source in matches:
+            target = plugins_out / source.name
+            shutil.copy2(source, target)
+            pending.append(source)
+    if not win:
+        # PipeWire dynamically loads support and protocol modules. ldd cannot
+        # discover those; bundle them with a private client-only config instead
+        # of accidentally loading a different distro's client modules/config.
+        libdir = plugin_dir.parent
+        for relative in ('pipewire-0.3/libpipewire-module-protocol-native.so',
+                         'pipewire-0.3/libpipewire-module-client-node.so',
+                         'pipewire-0.3/libpipewire-module-adapter.so',
+                         'spa-0.2/support/libspa-support.so',
+                         'spa-0.2/support/libspa-dbus.so',
+                         'spa-0.2/videoconvert/libspa-videoconvert.so'):
+            source = libdir / relative
+            if not source.is_file():
+                raise SystemExit('Missing PipeWire runtime library: ' + str(source))
+            target = destination / 'lib' / relative
+            target.parent.mkdir(parents=True, exist_ok=True)
+            shutil.copy2(source, target)
+            pending.append(source)
+        config = destination / 'share' / 'pipewire'
+        config.mkdir(parents=True, exist_ok=True)
+        (config / 'client.conf').write_text(
+            'context.spa-libs = { support.* = support/libspa-support video.* = videoconvert/libspa-videoconvert }\n'
+            'context.modules = [\n'
+            '  { name = libpipewire-module-protocol-native }\n'
+            '  { name = libpipewire-module-client-node }\n'
+            '  { name = libpipewire-module-adapter }\n'
+            ']\n')
+    copied = set()
+    while pending:
+        binary = pending.pop()
+        if win:
+            imported = re.findall(r'DLL Name:\s*(\S+)', output('objdump', '-p', str(binary)))
+            dependencies = [prefix / 'bin' / name for name in imported]
+        else:
+            dependencies = [Path(p) for p in re.findall(r'=>\s+(/\S+)', output('ldd', str(binary)))]
+        for dep in dependencies:
+            if not dep.is_file() or dep.name in copied:
+                continue
+            if not win and re.match(r'lib(c|m|dl|rt|pthread|resolv)\.so', dep.name):
+                continue
+            shutil.copy2(dep, libs_out / dep.name)
+            copied.add(dep.name)
+            pending.append(dep)
+    # License texts and package provenance accompany the dynamically linked
+    # libraries. Distribution builders keep the upstream package/source URLs.
+    licenses = destination / 'licenses'
+    licenses.mkdir(exist_ok=True)
+    if win:
+        source = prefix / 'share' / 'licenses'
+        if source.exists():
+            shutil.copytree(source, licenses, dirs_exist_ok=True)
+        packages = output('pacman', '-Q') if shutil.which('pacman') else 'See MSYS2 build log'
+    else:
+        packages = output('dpkg-query', '-W', '-f=${Package} ${Version}\n') if shutil.which('dpkg-query') else ''
+        # Debian copyright files contain licenses and source homepage details.
+        for path in Path('/usr/share/doc').glob('*/copyright'):
+            shutil.copy2(path, licenses / (path.parent.name + '.copyright'))
+    (destination / 'libraries.json').write_text(json.dumps(dict(gstreamer=output('pkg-config', '--modversion', 'gstreamer-1.0'),
+        plugins=names, libraries=sorted(copied), packages=packages), indent=2))
+    print(lib)
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('--out', type=Path, default=ROOT / 'bundle')
+    parser.add_argument('--with-x264', action='store_true', help='add the GPL x264 encoder (CI tests only, never shipped)')
+    args = parser.parse_args()
+    build(args.out.resolve(), args.with_x264)
@@ -0,0 +1,125 @@
+/* Frame Control's capture/encode adapter. GStreamer is a bundled library;
+ * capture uses WGC on Windows and the consented PipeWire fd on Linux. */
+#include "controller.h"
+#include <gst/gst.h>
+#include <gst/app/gstappsink.h>
+#include <gst/video/video-event.h>
+#include <string.h>
+
+typedef int (*Gate)(int stage, int64_t pts, int64_t capture, int64_t arrived);
+typedef struct {
+    GstElement *pipeline, *encoder, *sink, *raw_queue;
+    Gate gate;
+    GstSample *sample;
+    GstMapInfo map;
+    int mapped;
+    char error[512];
+} Capture;
+typedef struct { const unsigned char *data; int size, key, width, height; int64_t pts; } Encoded;
+FC_API int64_t fc_now(void) {return g_get_monotonic_time();}
+FC_API void fc_gst_init(void) {gst_init(NULL,NULL);}
+FC_API int fc_has_element(const char *name) {
+    GstElementFactory *f=gst_element_factory_find(name);
+    if(!f)return 0;gst_object_unref(f);return 1;
+}
+static GstPadProbeReturn probe(GstPad *pad,GstPadProbeInfo *info,gpointer data) {
+    Capture *c=data; GstBuffer *b=GST_PAD_PROBE_INFO_BUFFER(info);
+    if(!b)return GST_PAD_PROBE_OK;
+    int64_t now=fc_now(), cap=now;
+    GstClock *clock=gst_element_get_clock(c->pipeline);
+    if(clock && GST_BUFFER_PTS_IS_VALID(b)) {
+        GstClockTime current=gst_clock_get_time(clock),origin=gst_element_get_base_time(c->pipeline);
+        if(current>=origin && current-origin>=GST_BUFFER_PTS(b))
+            cap-= (int64_t)((current-origin-GST_BUFFER_PTS(b))/1000);
+    }
+    if(clock)gst_object_unref(clock);
+    int stage=GPOINTER_TO_INT(g_object_get_data(G_OBJECT(pad),"stage"));
+    if(stage==1) return c->gate(stage,(int64_t)GST_BUFFER_PTS(b),cap,now)>0 ? GST_PAD_PROBE_OK : GST_PAD_PROBE_DROP;
+    int phase=0;
+    for(;;) {
+        /* While congested, prefer a newer raw picture queued upstream. If
+         * nothing changed, retain this last picture until the gate opens;
+         * an idle window must not stay stale after a dropped final update. */
+        guint queued=0;
+        if(phase && c->raw_queue)g_object_get(c->raw_queue,"current-level-buffers",&queued,NULL);
+        if(queued)return GST_PAD_PROBE_DROP;
+        int decision=c->gate(phase,(int64_t)GST_BUFFER_PTS(b),cap,now);
+        if(decision)return decision>0 ? GST_PAD_PROBE_OK : GST_PAD_PROBE_DROP;
+        phase=2;g_usleep(2000);
+    }
+}
+FC_API Capture *fc_capture_open(const char *pipeline,Gate gate,char *error,int capacity) {
+    GError *e=NULL;Capture *c=g_new0(Capture,1);c->gate=gate;
+    c->pipeline=gst_parse_launch(pipeline,&e);
+    if(e || !c->pipeline) {
+        g_strlcpy(error,e?e->message:"No pipeline",capacity);
+        if(e)g_error_free(e);if(c->pipeline)gst_object_unref(c->pipeline);g_free(c);return NULL;
+    }
+    c->encoder=gst_bin_get_by_name(GST_BIN(c->pipeline),"enc");
+    c->sink=gst_bin_get_by_name(GST_BIN(c->pipeline),"out");
+    GstElement *raw=gst_bin_get_by_name(GST_BIN(c->pipeline),"gate");
+    if(!c->encoder || !c->sink || !raw) {
+        g_strlcpy(error,"Pipeline is missing enc, out or gate",capacity);
+        if(raw)gst_object_unref(raw);if(c->encoder)gst_object_unref(c->encoder);
+        if(c->sink)gst_object_unref(c->sink);gst_object_unref(c->pipeline);g_free(c);return NULL;
+    }
+    c->raw_queue=gst_bin_get_by_name(GST_BIN(c->pipeline),"raw_queue");
+    GstPad *p=gst_element_get_static_pad(raw,"src");
+    g_object_set_data(G_OBJECT(p),"stage",GINT_TO_POINTER(0));
+    gst_pad_add_probe(p,GST_PAD_PROBE_TYPE_BUFFER,probe,c,NULL);gst_object_unref(p);gst_object_unref(raw);
+    p=gst_element_get_static_pad(c->encoder,"sink");
+    g_object_set_data(G_OBJECT(p),"stage",GINT_TO_POINTER(1));
+    gst_pad_add_probe(p,GST_PAD_PROBE_TYPE_BUFFER,probe,c,NULL);gst_object_unref(p);
+    gst_element_set_state(c->pipeline,GST_STATE_PLAYING);
+    return c;
+}
+FC_API int fc_capture_pull(Capture *c,Encoded *out) {
+    if(c->mapped) {gst_buffer_unmap(gst_sample_get_buffer(c->sample),&c->map);c->mapped=0;}
+    if(c->sample){gst_sample_unref(c->sample);c->sample=NULL;}
+    GstBus *bus=gst_element_get_bus(c->pipeline);
+    GstMessage *m=gst_bus_pop_filtered(bus,GST_MESSAGE_ERROR);gst_object_unref(bus);
+    if(m) {
+        if(GST_MESSAGE_TYPE(m)==GST_MESSAGE_ERROR) {
+            GError *e=NULL;char *debug=NULL;gst_message_parse_error(m,&e,&debug);
+            g_strlcpy(c->error,e->message,sizeof(c->error));g_error_free(e);g_free(debug);
+        } else g_strlcpy(c->error,"The capture source closed",sizeof(c->error));
+        gst_message_unref(m);return -1;
+    }
+    c->sample=gst_app_sink_try_pull_sample(GST_APP_SINK(c->sink),100*GST_MSECOND);
+    if(!c->sample) {
+        if(gst_app_sink_is_eos(GST_APP_SINK(c->sink))) {
+            g_strlcpy(c->error,"The capture source closed",sizeof(c->error));return -1;
+        }
+        return 0;
+    }
+    GstBuffer *b=gst_sample_get_buffer(c->sample);
+    if(!gst_buffer_map(b,&c->map,GST_MAP_READ))return 0;
+    c->mapped=1;out->data=c->map.data;out->size=(int)c->map.size;
+    out->key=!GST_BUFFER_FLAG_IS_SET(b,GST_BUFFER_FLAG_DELTA_UNIT);out->pts=(int64_t)GST_BUFFER_PTS(b);
+    const GstStructure *s=gst_caps_get_structure(gst_sample_get_caps(c->sample),0);
+    gst_structure_get_int(s,"width",&out->width);gst_structure_get_int(s,"height",&out->height);return 1;
+}
+FC_API const char *fc_capture_error(Capture *c) {return c->error;}
+FC_API void fc_capture_bitrate(Capture *c,int bps) {
+    g_object_set(c->encoder,"bitrate",(guint)MAX(1,bps/1000),NULL);
+}
+FC_API void fc_capture_test(Capture *c,uint32_t input) {
+    GstElement *source=gst_bin_get_by_name(GST_BIN(c->pipeline),"source");
+    if(source) {
+        /* Each benchmark click visibly changes the ball, before injection is
+         * timestamped. This is a response, not merely a protocol input echo. */
+        guint color=0xff000000u | ((input*2654435761u)&0x00ffffffu);
+        g_object_set(source,"foreground-color",color,NULL);gst_object_unref(source);
+    }
+}
+FC_API void fc_capture_key(Capture *c) {
+    GstPad *p=gst_element_get_static_pad(c->encoder,"src");
+    gst_pad_send_event(p,gst_video_event_new_upstream_force_key_unit(GST_CLOCK_TIME_NONE,TRUE,0));gst_object_unref(p);
+}
+FC_API void fc_capture_close(Capture *c) {
+    gst_element_set_state(c->pipeline,GST_STATE_NULL);
+    if(c->mapped)gst_buffer_unmap(gst_sample_get_buffer(c->sample),&c->map);
+    if(c->sample)gst_sample_unref(c->sample);
+    if(c->raw_queue)gst_object_unref(c->raw_queue);
+    gst_object_unref(c->sink);gst_object_unref(c->encoder);gst_object_unref(c->pipeline);g_free(c);
+}
@@ -0,0 +1,130 @@
+/* A C port of the Mac agent's RateController (mac/frame-mac-view/Sources/
+ * Controller.swift) for the Windows host. Same gate, demand protection, bitrate
+ * recovery and tier hysteresis; bounded sample buffers. The Mac agent keeps its
+ * Swift original, so a change to either must be made to both:
+ * tests/test_pc_controller.py checks this one against the Swift trace. */
+#include "controller.h"
+#include <stdlib.h>
+#include <string.h>
+#define CAP 1024
+#define MIN(a,b) ((a)<(b)?(a):(b))
+#define MAX(a,b) ((a)>(b)?(a):(b))
+typedef struct { int64_t t, v; uint32_t seq; } Sample;
+typedef struct { Sample a[CAP]; int n; } Samples;
+struct FCController {
+    int max_fps, enabled, ceiling, target, tier, frame_bytes, held, signal, saw_ack;
+    int64_t decrease, increase, below, above, slack;
+    Samples flight, rtts, acked, sent, captures;
+};
+static const int fps[] = {60,45,30,30,30}, scale[] = {100,100,100,75,50};
+static const double floors[] = {.45,.28,.16,.08,0};
+static void remove_first(Samples *s, int n) {
+    s->n -= n; memmove(s->a, s->a+n, (size_t)s->n*sizeof(Sample));
+}
+static void add(Samples *s, Sample v, int cap) {
+    if (s->n >= cap) remove_first(s, 1);
+    s->a[s->n++] = v;
+}
+static void expire(Samples *s, int64_t oldest) {
+    int n=0; while(n<s->n && s->a[n].t<oldest) n++;
+    remove_first(s,n);
+}
+static int64_t base(FCController *c) {
+    int64_t v=0;
+    for(int i=0;i<c->rtts.n;i++) if(!i || c->rtts.a[i].v<v) v=c->rtts.a[i].v;
+    return v;
+}
+static int compare(const void *a, const void *b) {
+    int64_t x=*(const int64_t*)a, y=*(const int64_t*)b;
+    return (x>y)-(x<y);
+}
+static int64_t quantile(Samples *s,int64_t since,int numerator,int denominator) {
+    int64_t a[CAP]; int n=0;
+    for(int i=0;i<s->n;i++) if(s->a[i].t>since) a[n++]=s->a[i].v;
+    if(!n) return 0;
+    qsort(a,(size_t)n,sizeof(int64_t),compare);
+    return a[n*numerator/denominator];
+}
+static int rate(Samples *s) {
+    int64_t total=0; for(int i=0;i<s->n;i++) total+=s->a[i].v;
+    return (int)MIN(total*16,2147483647);
+}
+FCController *fc_new(int max_fps,int enabled) {
+    FCController *c=calloc(1,sizeof(*c));
+    if(c) {c->max_fps=MAX(1,max_fps);c->enabled=enabled;c->slack=40000;}
+    return c;
+}
+void fc_free(FCController *c) {free(c);}
+void fc_ceiling(FCController *c,int bps) {
+    if(!c->target || c->target>bps) c->target=bps;
+    c->ceiling=bps;
+}
+int fc_gate(FCController *c,int64_t now,int counts) {
+    if(!c->enabled || !c->saw_ack) return 1;
+    expire(&c->flight,now-2000000);
+    if(!c->flight.n) return 1;
+    int64_t interval=1000000/MIN(c->max_fps,fps[c->tier]);
+    int window=MAX(3,(int)((base(c)+c->slack)/interval)+1);
+    if(c->flight.n<window && now-c->flight.a[0].t<=base(c)+c->slack) return 1;
+    if(counts) c->held++;
+    return 0;
+}
+void fc_capture(FCController *c,int64_t now) {add(&c->captures,(Sample){now,0,0},256);}
+void fc_sent(FCController *c,uint32_t seq,int bytes,int64_t now) {
+    Sample s={now,bytes,seq};add(&c->flight,s,512);add(&c->sent,s,CAP);
+}
+int fc_ack(FCController *c,uint32_t seq,int64_t now) {
+    c->saw_ack=1;
+    for(int i=0;i<c->flight.n;i++) if(c->flight.a[i].seq==seq) {
+        Sample s=c->flight.a[i];remove_first(&c->flight,i+1);
+        add(&c->rtts,(Sample){now,now-s.t,0},CAP);
+        add(&c->acked,(Sample){now,s.v,0},CAP);return 1;
+    }
+    return 0;
+}
+int fc_update(FCController *c,int64_t now) {
+    expire(&c->rtts,now-10000000);expire(&c->acked,now-500000);
+    expire(&c->sent,now-500000);expire(&c->flight,now-2000000);
+    expire(&c->captures,now-1000000);
+    if(!c->enabled || c->ceiling<=0) return 0;
+    int64_t baseline=base(c), spread=quantile(&c->rtts,now-2000000,9,10);
+    int64_t jitter=spread ? spread-baseline : 0;
+    c->slack=1000000/MIN(c->max_fps,fps[c->tier])+MIN(MAX(jitter*3/2,25000),80000);
+    int64_t recent=quantile(&c->rtts,now-300000,1,2);
+    int64_t queue=recent ? recent-baseline : 0;
+    int64_t age=c->flight.n ? now-c->flight.a[0].t : 0;
+    int delivered=rate(&c->acked),sending=rate(&c->sent);
+    if(c->sent.n) c->frame_bytes=sending/16/c->sent.n;
+    int demand=(int)MIN((int64_t)MIN(c->captures.n,MIN(c->max_fps,fps[c->tier]))*c->frame_bytes*8,2147483647);
+    int signal=(sending>=c->target/2 && queue>40000)||c->held>=3||age>baseline+100000;
+    int congested=signal && c->signal;c->signal=signal;c->held=0;
+    if(congested && now-c->decrease>300000) {
+        int next=MAX(300000,MIN((int64_t)c->target*4/5,MAX((int64_t)delivered*9/10,c->target/2)));
+        if(demand>0 && (int64_t)demand*2<=(int64_t)c->target*5/4) next=MAX(next,MIN(c->target,(int64_t)demand*2));
+        c->target=next;c->decrease=now;
+    } else if(!congested && now-c->decrease>1000000 && now-c->increase>250000 && c->target<c->ceiling &&
+              (sending>(int64_t)c->target*6/10 || now-c->decrease>3000000)) {
+        c->target=MIN(c->ceiling,(int64_t)(c->target*1.1)+50000);c->increase=now;
+    }
+    double share=(double)c->target/MAX(c->ceiling,1);
+    if(c->tier<4 && share<floors[c->tier] && sending>=(int64_t)c->target*7/10) {
+        if(!c->below)c->below=now;
+        if(now-c->below>500000) {
+            for(c->tier=0;c->tier<4 && share<floors[c->tier];c->tier++) {}
+            c->below=0;
+        }
+    } else c->below=0;
+    if(c->tier>0 && share>floors[c->tier-1]*1.25) {
+        if(!c->above)c->above=now;
+        if(now-c->above>2000000) {c->tier--;c->above=0;}
+    } else c->above=0;
+    return c->target;
+}
+int64_t fc_value(FCController *c,int field) {
+    switch(field) {
+    case 0:return c->target;case 1:return c->ceiling;case 2:return c->tier;
+    case 3:return MIN(c->max_fps,fps[c->tier]);case 4:return scale[c->tier];
+    case 5:return base(c);case 6:return c->flight.n;case 7:return c->slack;
+    default:return 0;
+    }
+}
@@ -0,0 +1,29 @@
+/* Shared latency controller. Times are monotonic host microseconds.
+ * Callers serialize access. Capture is gated BEFORE encoding: dropping an
+ * encoded inter frame would break the decoder's reference chain. */
+#ifndef FRAME_CONTROLLER_H
+#define FRAME_CONTROLLER_H
+#include <stdint.h>
+#ifdef _WIN32
+#define FC_API __declspec(dllexport)
+#else
+#define FC_API
+#endif
+#ifdef __cplusplus
+extern "C" {
+#endif
+typedef struct FCController FCController;
+FC_API FCController *fc_new(int fps, int enabled);
+FC_API void fc_free(FCController *c);
+FC_API void fc_ceiling(FCController *c, int bps);
+FC_API int fc_gate(FCController *c, int64_t now, int counts);
+FC_API void fc_capture(FCController *c, int64_t now);
+FC_API void fc_sent(FCController *c, uint32_t seq, int bytes, int64_t now);
+FC_API int fc_ack(FCController *c, uint32_t seq, int64_t now);
+FC_API int fc_update(FCController *c, int64_t now);
+/* target, ceiling, tier, fps, scale percent, baseline us, in flight, slack us */
+FC_API int64_t fc_value(FCController *c, int field);
+#ifdef __cplusplus
+}
+#endif
+#endif
@@ -0,0 +1,147 @@
+/* xdg-desktop-portal RemoteDesktop + ScreenCast, one consented session per
+ * panel. PipeWire receives the portal fd, never the unrestricted daemon fd.
+ * Input uses only the devices granted in Start's response. */
+#ifndef _WIN32
+#include "controller.h"
+#include <gio/gio.h>
+#include <gio/gunixfdlist.h>
+#include <unistd.h>
+#include <string.h>
+#define BUS "org.freedesktop.portal.Desktop"
+#define PATH "/org/freedesktop/portal/desktop"
+#define RD "org.freedesktop.portal.RemoteDesktop"
+#define SC "org.freedesktop.portal.ScreenCast"
+typedef struct {
+    GDBusConnection *bus;
+    GMainContext *context;
+    char *session;
+    int fd;
+    unsigned node, devices;
+    int width,height;
+} Portal;
+typedef struct {GVariant *result;int done;unsigned code;} Response;
+static void response(GDBusConnection *c,const char *sender,const char *path,const char *iface,
+                     const char *signal,GVariant *params,gpointer data) {
+    (void)c;(void)sender;(void)path;(void)iface;(void)signal;
+    Response *r=data;g_variant_get(params,"(u@a{sv})",&r->code,&r->result);r->done=1;
+}
+static void option(GVariantBuilder *b,const char *key,GVariant *v) {g_variant_builder_add(b,"{sv}",key,v);}
+static GVariant *request(Portal *p,const char *iface,const char *method,GVariant *args,
+                         const char *token,char *error,int capacity) {
+    char *sender=g_strdup(g_dbus_connection_get_unique_name(p->bus)+1);
+    for(char *c=sender;*c;c++)if(*c=='.')*c='_';
+    char *path=g_strdup_printf(PATH "/request/%s/%s",sender,token);g_free(sender);
+    Response r={0};GError *e=NULL;
+    guint sub=g_dbus_connection_signal_subscribe(p->bus,BUS,"org.freedesktop.portal.Request","Response",
+                                                 path,NULL,G_DBUS_SIGNAL_FLAGS_NONE,response,&r,NULL);
+    GVariant *reply=g_dbus_connection_call_sync(p->bus,BUS,PATH,iface,method,args,G_VARIANT_TYPE("(o)"),
+                                               G_DBUS_CALL_FLAGS_NONE,10000,NULL,&e);
+    if(reply)g_variant_unref(reply);
+    int64_t deadline=g_get_monotonic_time()+120000000;
+    while(!e && !r.done && g_get_monotonic_time()<deadline) {
+        while(g_main_context_iteration(p->context,FALSE)) {}
+        g_usleep(10000);
+    }
+    if(!r.done) {
+        GVariant *closed=g_dbus_connection_call_sync(p->bus,BUS,path,"org.freedesktop.portal.Request","Close",
+                                NULL,NULL,G_DBUS_CALL_FLAGS_NONE,2000,NULL,NULL);
+        if(closed)g_variant_unref(closed);
+    }
+    g_dbus_connection_signal_unsubscribe(p->bus,sub);g_free(path);
+    if(e || !r.done || r.code) {
+        g_strlcpy(error,e?e->message:!r.done?"Screen sharing request timed out":"Screen sharing was cancelled or refused",capacity);
+        if(e)g_error_free(e);if(r.result)g_variant_unref(r.result);return NULL;
+    }
+    return r.result;
+}
+FC_API void fc_portal_close(Portal *p) {
+    if(!p)return;
+    if(p->session && p->bus) {
+        GVariant *r=g_dbus_connection_call_sync(p->bus,BUS,p->session,"org.freedesktop.portal.Session","Close",
+                                                NULL,NULL,G_DBUS_CALL_FLAGS_NONE,2000,NULL,NULL);
+        if(r)g_variant_unref(r);
+    }
+    if(p->fd>=0){close(p->fd);p->fd=-1;}
+    g_free(p->session);if(p->bus)g_object_unref(p->bus);
+    if(p->context)g_main_context_unref(p->context);g_free(p);
+}
+/* Each pipeline gets a fresh restricted PipeWire connection. A dup of a
+ * previously consumed protocol socket is not a new connection.
+ * Ownership: the Portal owns p->fd for its whole life and is its only closer.
+ * pipewiresrc never takes the fd it is given: its core connects with
+ * pw_context_connect_fd(ctx, fcntl(fd, F_DUPFD_CLOEXEC, 3), ...) and that
+ * duplicate is what PipeWire closes on teardown (src/gst/gstpipewirecore.c,
+ * unchanged from 0.3.19 through 1.x). So closing p->fd here is not a double
+ * close; not closing it would leak one socket per pipeline reopen. The
+ * caller closes the previous pipeline before asking for a new fd. */
+FC_API int fc_portal_refresh(Portal *p,char *error,int capacity) {
+    GVariantBuilder b;g_variant_builder_init(&b,G_VARIANT_TYPE_VARDICT);
+    GUnixFDList *fds=NULL;GError *e=NULL;
+    GVariant *r=g_dbus_connection_call_with_unix_fd_list_sync(p->bus,BUS,PATH,SC,"OpenPipeWireRemote",
+         g_variant_new("(oa{sv})",p->session,&b),G_VARIANT_TYPE("(h)"),G_DBUS_CALL_FLAGS_NONE,10000,NULL,&fds,NULL,&e);
+    if(!r){g_strlcpy(error,e->message,capacity);g_error_free(e);return -1;}
+    int handle;g_variant_get(r,"(h)",&handle);g_variant_unref(r);
+    int fd=g_unix_fd_list_get(fds,handle,&e);g_object_unref(fds);
+    if(fd<0){g_strlcpy(error,e->message,capacity);g_error_free(e);return -1;}
+    if(p->fd>=0)close(p->fd);
+    p->fd=fd;return fd;
+}
+FC_API Portal *fc_portal_select(char *error,int capacity) {
+    Portal *p=g_new0(Portal,1);p->fd=-1;p->context=g_main_context_new();
+    g_main_context_push_thread_default(p->context);
+    GError *e=NULL;GVariant *r=NULL;
+    p->bus=g_bus_get_sync(G_BUS_TYPE_SESSION,NULL,&e);
+    if(!p->bus) {g_strlcpy(error,e->message,capacity);g_error_free(e);goto fail;}
+    GVariantBuilder b;char token[64],session[64];
+    g_snprintf(session,sizeof(session),"fc_%u",g_random_int());
+    g_snprintf(token,sizeof(token),"fc_%u",g_random_int());
+    g_variant_builder_init(&b,G_VARIANT_TYPE_VARDICT);
+    option(&b,"handle_token",g_variant_new_string(token));option(&b,"session_handle_token",g_variant_new_string(session));
+    r=request(p,RD,"CreateSession",g_variant_new("(a{sv})",&b),token,error,capacity);if(!r)goto fail;
+    g_variant_lookup(r,"session_handle","s",&p->session);g_variant_unref(r);r=NULL;
+    if(!p->session){g_strlcpy(error,"Portal did not return a session",capacity);goto fail;}
+    g_snprintf(token,sizeof(token),"fc_%u",g_random_int());g_variant_builder_init(&b,G_VARIANT_TYPE_VARDICT);
+    option(&b,"handle_token",g_variant_new_string(token));option(&b,"types",g_variant_new_uint32(3));
+    r=request(p,RD,"SelectDevices",g_variant_new("(oa{sv})",p->session,&b),token,error,capacity);if(!r)goto fail;g_variant_unref(r);
+    g_snprintf(token,sizeof(token),"fc_%u",g_random_int());g_variant_builder_init(&b,G_VARIANT_TYPE_VARDICT);
+    option(&b,"handle_token",g_variant_new_string(token));option(&b,"types",g_variant_new_uint32(3));
+    option(&b,"multiple",g_variant_new_boolean(FALSE));option(&b,"cursor_mode",g_variant_new_uint32(2));
+    r=request(p,SC,"SelectSources",g_variant_new("(oa{sv})",p->session,&b),token,error,capacity);if(!r)goto fail;g_variant_unref(r);
+    g_snprintf(token,sizeof(token),"fc_%u",g_random_int());g_variant_builder_init(&b,G_VARIANT_TYPE_VARDICT);
+    option(&b,"handle_token",g_variant_new_string(token));
+    r=request(p,RD,"Start",g_variant_new("(osa{sv})",p->session,"",&b),token,error,capacity);if(!r)goto fail;
+    g_variant_lookup(r,"devices","u",&p->devices);
+    GVariant *streams=g_variant_lookup_value(r,"streams",G_VARIANT_TYPE("a(ua{sv})"));
+    if(streams && g_variant_n_children(streams)>0) {
+        GVariant *entry=g_variant_get_child_value(streams,0),*props=NULL;
+        g_variant_get(entry,"(u@a{sv})",&p->node,&props);
+        g_variant_lookup(props,"size","(ii)",&p->width,&p->height);
+        g_variant_unref(props);g_variant_unref(entry);
+    }
+    if(streams)g_variant_unref(streams);g_variant_unref(r);r=NULL;
+    if(!p->node || p->width<=0 || p->height<=0) {g_strlcpy(error,"Portal returned no stream size",capacity);goto fail;}
+    if(fc_portal_refresh(p,error,capacity)<0)goto fail;
+    g_main_context_pop_thread_default(p->context);return p;
+fail:
+    g_main_context_pop_thread_default(p->context);fc_portal_close(p);return NULL;
+}
+FC_API int fc_portal_value(Portal *p,int field) {
+    switch(field){case 0:return p->fd;case 1:return (int)p->node;case 2:return p->width;
+                 case 3:return p->height;case 4:return (int)p->devices;default:return 0;}
+}
+FC_API int fc_portal_input(Portal *p,int type,double x,double y,int code,int down) {
+    const char *method=NULL;GVariant *args=NULL;GVariantBuilder b;g_variant_builder_init(&b,G_VARIANT_TYPE_VARDICT);
+    if(type==0 && (p->devices&2)) {
+        method="NotifyPointerMotionAbsolute";args=g_variant_new("(oa{sv}udd)",p->session,&b,p->node,x,y);
+    } else if(type==1 && (p->devices&2)) {
+        method="NotifyPointerButton";args=g_variant_new("(oa{sv}iu)",p->session,&b,code,(guint)down);
+    } else if(type==2 && (p->devices&2)) {
+        method="NotifyPointerAxis";args=g_variant_new("(oa{sv}dd)",p->session,&b,x,y);
+    } else if(type==3 && (p->devices&1)) {
+        method="NotifyKeyboardKeysym";args=g_variant_new("(oa{sv}iu)",p->session,&b,code,(guint)down);
+    } else return 0;
+    GError *e=NULL;GVariant *r=g_dbus_connection_call_sync(p->bus,BUS,PATH,RD,method,args,NULL,
+                                                       G_DBUS_CALL_FLAGS_NONE,2000,NULL,&e);
+    if(e)g_error_free(e);if(r)g_variant_unref(r);return r!=NULL;
+}
+#endif
@@ -0,0 +1,222 @@
+# Frame Control for AI agents
+
+**Documented interface:** Frame Control's own stdlib Python MCP adapter wraps
+its loopback HTTP API. No API key, hosted service, model SDK or third-party
+helper app is needed. The assistant is our HTML/Python implementation hosted
+in the platform Chromium browser. Its optional LLM endpoint is user configuration.
+Installing other apps is an optional management action, never a prerequisite.
+
+## Connect an MCP client
+
+The default MCP command starts a private HTTP backend on a free loopback port,
+with a fresh local access key. It stops that backend when the MCP client closes
+stdin or sends SIGTERM. It uses its own SSH control socket, so closing it does
+not close the desktop app's connection. No manually started server is needed.
+
+Add this stdio server to your MCP client (use absolute paths):
+
+```json
+{
+  "mcpServers": {
+    "frame-control": {
+      "command": "python3",
+      "args": ["/absolute/path/frame-control/ui/frame_mcp.py"]
+    }
+  }
+}
+```
+
+For Codex, the equivalent registration is:
+
+```sh
+codex mcp add frame-control -- python3 /absolute/path/frame-control/ui/frame_mcp.py
+```
+
+New agent sessions load the entry. An already running session may need its MCP
+connections reloaded; registration does not retroactively add tools to its
+initial tool inventory. Keep the checkout at that path while it is registered.
+Use `codex mcp remove frame-control` to remove only this registration.
+
+To reuse a running server instead, pass `--url http://127.0.0.1:47810`.
+The desktop app uses a random port; use that port with `--url`, or run the
+checkout server above. If the HTTP server uses `FRAME_UI_KEY`, pass the same
+value in the MCP process environment. This is local access control, not an LLM
+API key. The adapter only accepts loopback HTTP servers, refuses redirects and
+ignores environment proxies. Stdout contains newline-delimited JSON-RPC only.
+It supports MCP initialization, ping, tool listing and tool calls; no sampling,
+resources, prompts or streaming transport.
+
+| Tool | Arguments | Effect |
+|---|---|---|
+| `computer_state` | none | Read-only gamescope window IDs/focus and bounded AT-SPI tree; reports incomplete observations |
+| `status` | none | Battery, services, installed games and Flatpaks |
+| `screenshot` | `view`: `headset` (default) or `desktop` | Returns PNG image content to the MCP client |
+| `job` | `id` | Background install status; poll until `done`, inspect `error` |
+| `launch` | `appid` | Launch an installed Steam app |
+| `install` / `uninstall` | `id` | Install from Flathub / remove a user Flatpak |
+| `send_text` | `text` | Frame desktop clipboard; desktop must be open |
+| `send_file` | `path` | File on the HTTP server computer, up to 16 MiB, copied to Frame `~/Downloads` |
+| `panel` | `id` | Launch an installed Flatpak as a panel using the existing launcher |
+| `power` | `action`: `suspend`, `reboot`, `poweroff` | Open a terminal for the user to enter the sudo password |
+| `keep_awake` | `action`: `on`, `off`, `status` | Optional keep-awake script interface |
+
+Only install free software with its developer's consent. There is no purchase,
+entitlement bypass or arbitrary shell tool. `install` returns a background job
+ID; it does not claim the installation has finished. APK and sideloaded title
+installs remain in the main UI for now.
+
+### Approval is a separate human action
+
+Every mutation first returns an `approvalUrl`, exact action and `confirmation`
+token. Ask the user to open that URL and choose **Approve this action** or
+**Reject**. Then repeat the same tool and arguments with the token in
+`confirmation`. The server refuses execution before approval, changed arguments,
+expired tokens and reuse. A file approval binds the content hash as well as the
+path. Approvals last five minutes and disappear when the HTTP server restarts.
+A failed execution also consumes the approval; review a fresh request to retry.
+The panel does not execute an action merely because it was approved.
+
+MCP has no approval tool. This is protection against accidental model tool
+calls, not a sandbox against a client with independent shell/HTTP access to your
+computer. Grant the MCP client only the access you intend. Status, captures and computer-state observations
+are returned directly to that client, which may forward them to its configured
+model. The assistant's separate opt-in does not govern an external MCP client.
+
+Power still requires the existing password prompt in a local terminal. MCP
+never receives passwords. Power via `FRAME_LOCAL=1` is unsupported: use the main
+UI. The panel launcher and keep-awake adapter require zsh on the computer.
+
+[PR #16](https://github.com/saphid/frame-control/pull/16) owns
+`scripts/keep-awake.sh on|off|status`. This branch does not copy or change it.
+Until that script is present, the tool reports it unavailable. Keep-awake is
+never automatic: `on` changes the shared idle timers; explicitly approve `off`
+to restore them after work. It is not a per-agent lease; coordinate with other
+users. No changes are made to the analytics/update interfaces in
+[PR #17](https://github.com/saphid/frame-control/pull/17). Prompts, keys, model
+replies, screenshots and approval payloads are not sent to analytics.
+
+## Assistant panel
+
+Open **Settings** (the gear in the header) **→ Open assistant settings**, or `http://127.0.0.1:47810/assistant`.
+To put the same page in the headset, with the HTTP server still running:
+
+```sh
+python3 scripts/assistant-on-frame.py --port 47810
+```
+
+This starts an SSH reverse forward bound to Frame loopback (port 47812 by
+default), then a dedicated Chromium profile tagged as a SteamVR panel. Keep the
+command running. Ctrl-C closes this browser profile and the tunnel; it leaves
+other Chromium windows and the existing HTTP server alone. A failed cleanup
+prints the temporary profile path so it can be removed when the Frame returns.
+Use `--frame-port` if the default is busy. Chromium must already be available as
+`org.chromium.Chromium`; the launcher never installs anything automatically.
+Place the panel with SteamVR's normal docking controls.
+
+Enter your full **chat-completions endpoint**, model name and optional key.
+An OpenAI-compatible local server works without a key; no OpenAI account is
+required. HTTP is allowed only on loopback; other endpoints require HTTPS.
+Loopback refers to the computer running the HTTP server, even in the headset.
+Endpoints with embedded credentials, query strings or redirects are refused.
+
+Check the message consent box and press **Send message**. Screenshot context is
+a separate unchecked box and sends one fresh capture with that request. Both
+boxes reset after sending, and changing endpoint/model revokes consent. Nothing
+is sent when opening the page or entering configuration. There is no model
+list fetch, saved history, automatic screenshot capture or assistant telemetry.
+Each send is independent: previous messages and replies are not included.
+
+Configuration, credentials and chat remain in page memory; close/reload the page
+or choose **Clear everything** to clear them. A request already sent cannot be
+recalled. Only the chosen endpoint gets the request; proxy environment variables
+and redirects are disabled. Its privacy and retention policy still applies.
+Replies are plain text and cannot call tools or operate the Frame. A model must
+support image inputs to accept screenshot context.
+
+## Evidence and limits
+
+**Verified 2026-09-28, SteamOS 0.4.1, BUILD_ID 20260925.6191901:** loopback HTTP
+status through an SSH reverse tunnel; platform Chromium created a separate
+SteamVR panel (confirmed in `GAMESCOPE_FOCUSABLE_APPS`); headset capture returned
+a PNG. These checks preceded the UI implementation. No power or global settings
+were changed.
+
+**Inferred:** visual comfort and controller keyboard usability while wearing
+the headset; panel creation in gamescope alone does not establish these.
+Windows/Linux launcher support, live third-party model endpoints, installs,
+uninstalls, power and keep-awake changes are not covered by that feasibility
+check. See the PR for the final unit and end-to-end results.
+
+**Verified end to end on the same Frame/build (2026-09-28):** a stdio MCP client
+initialized, read status, retrieved a headset PNG, and transferred a test file
+only after approval through the Chromium page. Remote file bytes matched;
+reusing the confirmation was rejected. The actual headset Chromium page sent
+text and then separately opted-in image context to a local test endpoint and
+displayed its replies. Without consent there were zero endpoint requests.
+The test endpoint returned canned replies: model inference and a live external
+provider remain **unverified**. The launcher’s Ctrl-C cleanup was checked;
+profiles, SSH tunnels and the test file were removed. No installs, removals,
+launches of user games, power operations or keep-awake changes were performed.
+
+**Verified locally:** unit coverage includes the stdio subprocess, approval
+binding/expiry/replay/concurrency, file-change rejection, and a real local HTTP
+endpoint for opt-in, text/image payloads and redirect refusal. Fake-Frame
+regressions are in `tests/e2e/test_agents.py`; local Docker execution was blocked
+because the Docker daemon was unavailable. The ARM64 fake-Frame CI job passed
+on this branch (run 36421345682).
+
+**Verified on the same Frame/build:** both Ctrl-C and SIGTERM close the dedicated
+browser profile and SSH tunnel and remove the profile and panel log.
+
+**Verified end to end on the same Frame/build (2026-09-29), with mutations:**
+a stdio MCP client started `ui/frame_mcp.py` in its default mode (private
+backend, no API key, no prestarted server) and a human approved or rejected
+each change in the approval page in a real Chrome window:
+
+| Tool | Result on the Frame |
+|---|---|
+| `send_file` | Approved; the file arrived in `~/Downloads` with identical contents |
+| `install` | Approved; `io.github.fizzyizzy05.binary` job finished in about 35 s |
+| `panel` | Approved; gamescope listed a new panel window, and `computer_state` reported the same window ID and PID. The app rendered in that window (below) |
+| `launch` | Approved; Keep Talking and Nobody Explodes (341800) started under Proton and `computer_state` reported it as the focused app |
+| `uninstall` | Approved; app and locale removed |
+| `power` | Rejected in the page. Unapproved retries, the same token used for `uninstall`, and a retry after rejection were all refused. Nothing was powered off |
+| `send_text` | Approved, then refused because the Plasma desktop was not open (documented requirement) |
+| `keep_awake` | `status` reports the script unavailable until PR #16 lands |
+
+A separate Claude Code CLI session, with only this server configured, read
+status, `computer_state` and a headset capture, and requested an install. It
+received an approval URL and did not execute anything.
+
+The assistant opened as a Frame panel through `scripts/assistant-on-frame.py`.
+Against a loopback stub model, a send without consent made zero requests. With
+consent it made exactly one, carrying the text and a fresh Frame screenshot.
+Consent unticked itself after sending. SIGTERM removed the panel, profile, log
+and tunnel.
+
+**Not verified while unworn:** every headset capture was a uniform dark frame,
+so SteamVR's rendered view of panels and the game could not be checked; window
+captures (`xwd`) were used instead. MCP can launch a game or panel but has no
+tool to stop one: the tester stopped them over SSH. Removing an app leaves any
+runtime it pulled in; Flatpak may also remove related extensions when that
+runtime is removed by hand.
+
+![Approval page showing the exact install action](img/mcp-approval-install.png)
+
+![The installed Flatpak rendering in its own gamescope panel window](img/mcp-panel-binary.png)
+
+![Assistant in Frame Chromium, after an opted-in request to the local test endpoint](img/assistant-panel.png)
+
+## Computer-use coverage
+
+MCP is the tool transport, not a limit on what an agent can do. A screenshot,
+accessibility snapshot, click or keystroke can all be MCP tools when we have a
+reliable underlying implementation. See [the investigation](computer-use.md)
+for the verified boundaries. `computer_state` adds observation, not an input
+channel: it cannot click an approval button or send keyboard/mouse events.
+
+**Verified 2026-09-29, SteamOS 0.4.1, BUILD_ID 20260925.6191901:** the command saved
+by `codex mcp add` launched without a prestarted server, negotiated MCP, listed
+12 tools, read live Frame status and returned X11 window state plus AT-SPI
+observations. It exited 0 at EOF. Steam's accessibility tree had inaccessible
+children, reported as `incomplete: true`; this is not a complete actionable UI.
@@ -0,0 +1,57 @@
+# Import an Android game backup
+
+A SideQuest backup can contain a small APK and much larger `main.…obb` and
+`patch.…obb` files. All of them belong to the same game. Installing only the APK
+can leave the game unable to start because its assets are missing.
+
+In **Add / Install → Import Android game** (or **Apps → Browse → Import Android game**), choose the APK and its OBB files together.
+You can also drop them anywhere in the window. In the desktop app, dropping the
+APK alone finds matching OBB files beside it; dropping a folder containing the
+APK and OBBs also works. Browsers can read only the files you explicitly choose.
+
+The import screen shows the game app and its game-data files together. **Import
+game & data** installs the APK, starts its own Steam library instance, waits for
+Android and the installed package, copies the OBBs, and closes the game. Open it
+again from the Steam library to play. The screen explains this before import;
+choosing files alone never starts a game.
+
+Files are matched using the package inside the APK and the standard
+`main|patch.<version>.<package>.obb` name, not the APK's filename. Both main and
+patch files are preserved. OBB version numbers need not equal the APK's current
+version code: updates can reuse older expansion files. The copy uses the existing
+SHA-256 transfer check inside Android and writes to that game's own
+`/sdcard/Android/obb/<package>` folder.
+
+If no data files are selected, the screen reminds you to include any OBBs in the
+backup before offering **Install game app**. An APK cannot reliably tell Frame
+Control whether separate expansion files are required. Game-data files are never
+silently sent to Downloads by this flow.
+
+If setup fails, the import screen keeps the error and retry instructions visible.
+The APK may already be installed, but the game is not reported ready. Import the
+same backup again, or choose just its OBB files to add data to an existing install.
+The importer checks the installed package and version before writing data and
+closes the instance after a failed copy. Unrelated or duplicate files, running
+games that would be replaced, and downgrades are rejected. Uploaded staging files
+are discarded after use or cancellation; abandoned imports expire after an hour
+when the importer is next used, and dead-server staging folders are swept at
+startup.
+
+This flow handles one ordinary APK and standard OBB files for one game at a time.
+Split APKs, XAPK/APKS archives, nested Android/data folders and private save
+archives need their own import handling. PC ZIP/folder/EXE imports keep their
+existing route. Quest USB and Meta library imports keep their existing workflow.
+
+Validation covers actual APK parsing, the three-file backup layout, package
+matching, sibling discovery, browser upload staging and cleanup, instance/version
+guards and failure recovery with mocked headset operations. Browser verification
+uses the real page and upload/inspection endpoints with simulated installation.
+**Verified 2026-10-20 on a real Steam Frame**, BUILD_ID `20260925.6191901`:
+F-Droid Basic 2.0.2 with two synthetic main/patch OBBs completed the automatic
+install/start/copy/close sequence. Reopening its persistent instance confirmed
+both SHA-256 hashes in `/sdcard/Android/obb/org.fdroid.basic`. The test app,
+shortcut and instance were removed afterward, the shared device lock was
+released, and the battery remained at 99%. See [device evidence](evidence/android-import/device.json).
+The headset was unworn; this checks transfer and persistence, not rendering,
+Resident Evil 4 gameplay, or a game that exits before data can be copied. This
+flow does not change APK compatibility or remove licence checks.
@@ -0,0 +1,144 @@
+# Announcing changes
+
+How we tell people about Frame Control features and fixes as they merge. The
+same few sentences feed the X post, the release notes and the website, so they
+are written once, in the pull request, while the change is fresh.
+
+## What gets announced
+
+| Kind | Announce? | Example |
+|---|---|---|
+| **New** — something you can now do | Yes, its own post | Stream Mac windows into the Frame as panels |
+| **Better** — something existing got noticeably easier, faster or wider | Yes, its own post or a roundup | APKs install without the Android SDK |
+| **Fixed** — something broken that users hit | Yes if people reported it or it blocked a flow; otherwise the next roundup | Mac mirror showed a zoomed-in corner |
+| **Release** — a tagged build | Always, one post linking the release | Frame Control 0.3.1 |
+| Tests, refactors, CI, docs-only, website polish | No | Fake Frame tests, screenshot crop |
+
+If a change isn't worth a sentence to someone who owns a Frame, it isn't
+announced.
+
+## The voice
+
+Write it the way the README and release notes already read.
+
+- **Lead with what the person can now do**, in their words: "Install older
+  versions of an app when the newest won't run on the Frame", not "Add APK
+  version fallback resolver".
+- **Plain and specific.** Name the thing, give the number: "about 30 fps",
+  "4,500 apps", "up to 8 older versions". No "blazing", "game-changing",
+  "excited to announce", "huge", or exclamation marks.
+- **Say where it works.** Platforms and what it was tested on, briefly:
+  "Tested on a real Frame from macOS 27." Don't claim what wasn't tested.
+- **Say the catch.** If it needs a setup step, an unsigned build, or only works
+  on one OS, say so in the same post.
+- **Sentence case**, full sentences, British spelling to match the docs.
+  Contractions are fine.
+- **No emoji in the text.** One image, GIF or short clip carries the tone
+  instead. The only symbol is the kind label below.
+- **Unofficial, always.** Never imply Valve made or endorses it. Say "Steam
+  Frame" for the headset and "Frame Control" for the app.
+- **Credit people.** If a user reported the bug or suggested the feature and is
+  happy to be named, thank them by handle.
+
+## The formats
+
+Every announceable PR ends with an `## Announcement` section holding these.
+The reviewer checks it like code.
+
+### 1. The post (X, and any other social account)
+
+```
+<Kind>: <what you can do now, one sentence>
+
+<one or two sentences: how it works, the catch, or what it was tested on>
+
+<link>
+```
+
+- `<Kind>` is `New`, `Better` or `Fixed`.
+- 280 characters maximum including the link (X counts any link as 23).
+- One link: the release if it has shipped, otherwise the PR.
+- One visual when the change is visible: a screenshot from the app, a GIF, or a
+  short clip from the headset. Alt text describes what it shows.
+- No hashtags, except `#SteamFrame` on releases and on posts about something
+  new, because people search for it.
+
+### 2. The release-note line
+
+One bullet under **New in x.y.z**, same as the current release notes: the
+first half of the post's first sentence, no kind label, no link.
+
+### 3. Release post
+
+```
+Frame Control <version>: <the headline change>
+
+<one sentence on the headline change>. Also: <two or three short items>.
+
+Windows, macOS and Linux: <release link>
+#SteamFrame
+```
+
+The release title on GitHub uses the same `Frame Control <version>: <headline>`
+line, as 0.3.0 and 0.3.1 already do.
+
+### Roundups
+
+Small fixes that don't earn their own post wait for a roundup, posted with the
+next release or when three or more have piled up:
+
+```
+Fixed in Frame Control this week:
+- <fix>
+- <fix>
+- <fix>
+
+<link>
+```
+
+## Examples from what has already merged
+
+**#11, older APK versions**
+
+```
+New: when an Android app is too new for the Frame, Frame Control now offers
+older versions that will install.
+
+It checks F-Droid, its archive and IzzyOnDroid, and verifies each download
+before it goes on the headset.
+
+https://github.com/saphid/steam-frame/pull/11
+```
+
+**#8, Mac mirror fixes**
+
+```
+Fixed: mirroring your Mac into the Steam Frame now fits the whole desktop in
+the panel, asks for the right password, and shows the cursor.
+
+Tested end to end on a real Frame from macOS 27.
+
+https://github.com/saphid/steam-frame/pull/8
+```
+
+**v0.3.1**
+
+```
+Frame Control 0.3.1: install APKs without the Android SDK
+
+Frame Control now reads APK files itself, so there's nothing extra to install.
+Also: Linux and Windows game sideloading, and one-click install links.
+
+Windows, macOS and Linux: https://github.com/saphid/steam-frame/releases/tag/v0.3.1
+#SteamFrame
+```
+
+## Posting
+
+Nothing is posted without a person approving it. The flow is:
+
+1. The PR carries its `## Announcement` section.
+2. On merge, the post is drafted from that section (manually for now).
+3. Alex approves or edits it, then it's posted from the project account.
+4. Replies and questions that turn out to be bugs become GitHub issues labelled
+   `feedback`, same as the website form.
@@ -0,0 +1,197 @@
+# APK repositories
+
+Frame Control supports **F-Droid-format repositories**, including F-Droid,
+F-Droid archive, IzzyOnDroid and user-provided HTTPS repositories. Repository
+indexes are authenticated before their apps appear. Search lists builds with
+Android API ≤30 and arm64-v8a or no native libraries, using the same streaming
+reducer as the existing catalogue. This does not guarantee an app works in Lepton.
+
+## Formats considered
+
+| Format | Users and purpose | Support in this source |
+|---|---|---|
+| F-Droid v2 | F-Droid, IzzyOnDroid, self-hosted fdroidserver repositories; consumed by F-Droid clients including Droid-ify and Neo Store | Preferred: signed `entry.jar` authenticates `entry.json`; its SHA-256 authenticates `index-v2.json`, which supplies APK SHA-256 hashes |
+| F-Droid v1 | Older F-Droid servers and clients | Fallback: verify `index-v1.jar`, then read its signed `index-v1.json` |
+| Obtainium configurations / exports | Obtainium users share app URLs plus source-specific filters and update settings; exports can contain a list of app configuration objects | Not imported here: configurations describe how to find releases, not one signed repository index |
+| SideQuest listings / custom feeds | SideQuest's own app discovery and installation service | No interoperable signed custom-repository specification was established from the public project documentation examined; SideQuest needs its own adapter |
+| GitHub release lists | Developers publish APK assets on release pages; community lists link to projects | Not a repository standard: asset naming, build selection and publisher verification vary; handled separately from this F-Droid source |
+| Minimal JSON list | A private list could contain package, title, APK URL and SHA-256 | Deliberately not introduced: unsigned hashes downloaded alongside files do not authenticate their publisher; another bespoke signing/update protocol would duplicate F-Droid |
+
+Research references (checked 2026-09-28):
+
+- [F-Droid APIs](https://f-droid.org/docs/All_our_APIs/) and
+  [repository setup](https://f-droid.org/docs/Setup_an_F-Droid_App_Repo/).
+- [F-Droid signing keys](https://f-droid.org/docs/Release_Channels_and_Signing_Keys/)
+  and [IzzyOnDroid's repository page and fingerprint](https://apt.izzysoft.de/fdroid/).
+- [Droid-ify](https://github.com/Droid-ify/client) and
+  [Neo Store](https://github.com/NeoApplications/Neo-Store).
+- [Obtainium](https://github.com/ImranR98/Obtainium), its
+  [configuration/deep-link format](https://wiki.obtainium.imranr.dev/deep_links/),
+  and [community app configurations](https://apps.obtainium.imranr.dev/).
+- [SideQuest's public client](https://github.com/SideQuestVR/SideQuest).
+  The absence of a specification in these materials is not proof that no
+  historical or private custom-feed format exists.
+
+## Add a repository in Frame Control
+
+From the Frame Control checkout, use its source-management CLI:
+
+```sh
+python3 ui/apk_sources/fdroid.py add 'https://example.org/fdroid/repo?fingerprint=YOUR_64_HEX_CERTIFICATE_FINGERPRINT' --name 'My apps'
+python3 ui/apk_sources/fdroid.py list
+python3 ui/apk_sources/fdroid.py search SOURCE_ID 'music'
+python3 ui/apk_sources/fdroid.py download SOURCE_ID org.example.app
+python3 ui/apk_sources/fdroid.py remove SOURCE_ID
+```
+
+Replace `SOURCE_ID` with the `id` printed by `add` or `list`. `--fingerprint`
+can also supply the pin. `fdroidrepos://example.org/fdroid/repo?fingerprint=…`
+links are accepted and converted to HTTPS. Conflicting fingerprints are refused.
+A URL must identify the repository directory, not its website or an index file.
+
+Adding fetches and validates the complete index **before saving** the source.
+Without a fingerprint, Frame Control verifies the JAR signature and remembers
+its signer: trust on first use (TOFU). This establishes continuity with the
+first server response, not independent publisher identity. Obtain the published
+fingerprint through a trusted channel when possible; the store's Add a source
+form shows the pinned one ("Trusted on first use: …") so you can compare it.
+Re-adding an existing URL preserves its pin; changing it requires deliberately
+removing and re-adding it.
+
+### Repositories behind a login
+
+Some self-hosted repositories sit behind HTTP basic auth. As in F-Droid's own
+client, the login can be part of the address
+(`https://user:pass@example.org/fdroid/repo`), or given separately: in the
+store's **Add a source** form under "This repository needs a login", in the
+server API (`username` and `password` beside `url`), or on the command line,
+where the password is read from standard input so it stays out of shell history
+and `ps`:
+
+```sh
+printf '%s\n' "$PASSWORD" | python3 ui/apk_sources/fdroid.py add 'https://example.org/fdroid/repo' --username me --password-stdin
+```
+
+How the login is kept and used:
+
+- **Kept out of the address.** The `user:pass@` part is removed before anything
+  is saved, so the saved repository URL, its id, the source list, job results
+  and the page never contain it. The store only says "Uses a login".
+- **Stored separately.** Logins live in
+  `frame_host.data_dir('source-credentials.json')`
+  (`~/Library/Application Support/Frame Control/source-credentials.json` on
+  macOS), written with mode 0600 and keyed by source id. They are stored in
+  plain text there, protected by file permissions only (on Windows,
+  the folder's normal ACLs). Removing the repository deletes its login, and
+  the file goes when the last login is removed. Each change holds a lock
+  file (`source-credentials.lock`), so the app and the command line can't undo
+  each other's saves or removals.
+- **HTTPS only.** An `http://` address with a login is refused before any request.
+- **Inside the repository only.** The `Authorization` header is decided
+  separately for every request, redirects included: index, `entry.jar`,
+  APKs and the store's artwork proxy. It is sent only when the request's
+  scheme, host and port match the repository's and its path is inside the
+  repository's directory by whole path segments. A redirect to another host
+  or port, or to another path on the same server, gets no login, and that
+  request's own `Authorization` header is removed. Paths with `.` or `..`
+  segments (plain or percent-encoded), encoded `/` or `\`, double encoding
+  or `;` get no login, since a server could normalise them to a path
+  outside the repository. Index file names like that are refused outright.
+- **Never echoed.** Logins are not written to logs, error messages, API
+  responses or the cache. The CLI prints only the saved source.
+- **Checked before saving.** As with any repository, the whole index must
+  download and verify before anything is saved. HTTP 401 (or 403 when a login
+  was sent) stops the add with "check the username and password" and saves
+  nothing: no repository, no login, no rollback state. A 403 with a login is
+  not treated as rate limiting.
+- **Same verification.** The login changes only who may download. Signature,
+  fingerprint pin or TOFU, index hash and rollback checks apply exactly as for
+  any other repository.
+
+Re-adding the same address with a new login replaces the saved one. Re-adding
+it without a login keeps the saved one.
+
+The API for the search/server integration is in `ui/apk_sources/fdroid.py`:
+`add_repo(url, fingerprint=None, name=None, username=None, password=None)`, `remove_repo(source_id)`,
+`set_enabled(source_id, enabled)`, and `user_repos()`. The module also exposes
+`sources`, `search`, `details`, and `download` from the shared source contract.
+This change supplies the CLI and API; the integrated source-management UI is
+separate work. Built-in sources can be disabled but cannot be removed.
+
+Settings and pins live in `frame_host.data_dir('apk-repos.json')`
+(`~/Library/Application Support/Frame Control/apk-repos.json` on macOS).
+Authenticated reduced indexes and APKs live under
+`frame_host.cache_dir('apk-sources')`; indexes refresh after 24 hours. An
+expired index is still served (marked stale in the store) while it refreshes in
+the background; a failed refresh is retried after 10 minutes.
+Only the running Frame Control app prunes cached APKs (at start and after store
+downloads); the command-line tools never do.
+The existing catalogue's unverified index cache is never treated as authenticated.
+
+Rollback protection: each repository's newest accepted signed index timestamp
+is kept in `apk-repo-state.json` next to the settings, and an older index is
+refused. Once a repository has served a v2 `entry.jar`, a missing `entry.jar`
+is an error rather than a reason to fall back to `index-v1.jar`. `entry.jar`
+must be signed with SHA-2 (SHA-1 is still accepted for legacy `index-v1.jar`).
+Removing a repository clears its state.
+
+## Publish your own repository
+
+Only publish free APKs you own or have the developer's permission to distribute.
+Do not publish paid app mirrors or bypass store licences. Check distribution
+terms before adding someone else's repository; this module does not infer legal
+permission from a signature or automatically audit a repository's terms.
+
+Install a current [fdroidserver](https://f-droid.org/docs/Installing_the_Server_and_Repo_Tools/)
+and its documented Android/Java dependencies on the publishing machine, then:
+
+```sh
+mkdir my-fdroid
+cd my-fdroid
+fdroid init
+# Set repo_url in config.yml to https://example.org/fdroid/repo
+# Also set repo_name and repo_description; keep the generated signing key safe.
+cp /path/to/your-free-app.apk repo/
+fdroid update --create-metadata
+# Review the generated metadata (name, summary, licence, source and website).
+fdroid update
+```
+
+Serve the generated **repo directory** at that HTTPS URL, including APKs,
+icons, `entry.jar`, `index-v2.json` and `index-v1.jar`. Do not publish the
+private signing keystore or configuration passwords. Configure fdroidserver's
+`serverwebroot` and run `fdroid deploy` for managed publication, or copy the
+public directory with your existing deployment tool. Publish the SHA-256
+repository certificate fingerprint displayed by fdroidserver in a link such as
+`https://example.org/fdroid/repo?fingerprint=…`.
+
+Keep the repository signing key backed up: changing it breaks existing pins.
+For updates, add the new APK, edit metadata as needed, run `fdroid update` and
+publish again. Test the published URL with Frame Control's `add`, `search` and
+`download` commands. The above publisher setup is documented from fdroidserver;
+it was not executed as part of this implementation.
+
+## Verification and limits
+
+The stdlib verifier supports one RSA PKCS#1 v1.5 JAR/CMS signer with a key of
+2048–8192 bits; SHA-256/384/512 and legacy SHA-1 digest encodings are
+recognized. It checks the signer certificate pin, the signature over `.SF`,
+the whole-manifest digest, and the manifest's digest of the JSON member.
+ECDSA, DSA, RSA-PSS, multiple signers and section-only `.SF` manifests are
+rejected. Certificates are pinned identities, not validated as Web PKI chains.
+HTTPS certificates are separately checked by Python's normal TLS validation.
+
+v1 fallback occurs only when `entry.jar` returns HTTP 404 or 410. Signature,
+fingerprint, index hash, TLS and server errors never trigger an unsigned
+fallback. APKs are cached by SHA-256 and checked again before reuse. Here,
+`verified: true` means the bytes match the signed repository's APK hash; it
+is not an independent APK publisher-signature or runtime compatibility verdict.
+There is no repository timestamp rollback/expiry policy or automated signing-key
+rotation yet. An old correctly signed index can still validate.
+
+Offline fixtures exercise v2, v1, TOFU, pin changes, disabled sources, cache
+reuse, URL rejection and corruption of every signature/hash layer. On the Mac,
+the real IzzyOnDroid repository was added with its published pin, searched for
+Tiny Music Player, and its 16,520-byte APK downloaded with SHA-256
+`d7bcb24d101b04beb3394b695b24be4e2c3d6ed702f1d0e06bc4dd707f64d86a`.
+No headset connection or installation was performed.
@@ -0,0 +1,116 @@
+# Developer-consented APK sources
+
+Surveyed 2026-09-28. Free access is not proof of redistribution permission or
+Frame compatibility. These adapters fetch only public publisher releases or
+link to publisher pages. They do not acquire store entitlements, defeat access
+checks, install anything, or rehost APKs. See [VR compatibility](vr-apks.md).
+
+| Source | Developer consent and automated-access position | API/feed; VR coverage | Decision |
+|---|---|---|---|
+| [itch.io](https://itch.io/docs/legal/terms) | Publishers warrant distribution rights (§4). Users may access content through the service; this is not blanket scraping permission. Main robots excludes `/game/download/`; author subdomains exclude `/*/download/`. No challenge bypass. | Public free Android RSS for `openxr` and `oculus-quest`; substantial indie VR. Server API is mostly authenticated publisher/account functionality, not a general anonymous store-download API. | Implement RSS search, artwork and page links; `downloadable: False`. The supplied free-download script follows keyed download pages excluded by robots, so it is not shipped. |
+| [GitHub releases](https://docs.github.com/en/rest/releases/releases) | Maintainers publish assets; curated repositories below establish provenance. Public hosting or an open-source topic alone does not establish rights to every uploaded binary. Use supported REST API under [API terms](https://docs.github.com/en/site-policy/github-terms/github-terms-of-service#h-api-terms), not HTML crawling. | Releases API includes APK assets and sometimes SHA-256. Topic search finds OpenXR/Quest projects. 60 unauthenticated requests/hour; authenticated user limits are generally 5,000/hour, with separate search/secondary limits. | Implement curated downloads and explicit topic discovery. Unreviewed topic results are page-only. |
+| [Uptodown](https://www.uptodown.com/aboutus) | Developer distribution program exists, but that does not prove publisher authorization for every catalog item. [Privacy policy](https://www.uptodown.com/aboutus/privacy) explicitly describes protection against automated access. General automation permission was not established. | Broad Android catalog, limited VR focus; no supported public consumer-download API established in this survey. | Page links only; no downloader. Do not infer consent from an unchanged APK signature. |
+| [APKPure](https://apkpure.com/terms) | Third-party APK catalog; individual publisher consent and automation rights were not established. Terms request returned HTTP 403; no bypass attempted. | Broad Android coverage, incidental VR; internal endpoints are not permission to automate. | Exclude automatic indexing/downloading; user may open site. |
+| [APKMirror](https://www.apkmirror.com/faq/) | Publisher-signed files and a free-app policy are not a blanket developer-consent or automation grant. FAQ request returned HTTP 403, so current terms could not be confirmed. | General Android/version archive; APK bundles often need another installer; little VR focus. No supported consumer-download API established. | Page links only, no scraping or bundle conversion. |
+| [Aptoide](https://en.aptoide.com/company/legal) | Terms define an app supplier as developer, owner or authorized distributor; user stores still require per-item provenance. API availability alone does not settle third-party access rights. | API ecosystem and general Android catalog; weak VR focus. | Defer until a publisher-owned store and its API terms can be approved. No blanket community-store downloader. |
+| [Amazon Appstore](https://developer.amazon.com/docs/app-submission/understanding-submission.html) | Official developer submissions; store account, device and license rules apply. Publisher submission APIs do not authorize public binary extraction. | Fire-device distribution; Android-device Appstore support ended in 2025; little Quest relevance. | Official product links only; no account or entitlement extraction. |
+| [PICO / ByteDance store](https://developer.picoxr.com/document/distribute) | Official publisher channel with store/device entitlements. No public unauthenticated binary-download grant established; documentation request encountered a redirect error. | Strong standalone VR; PICO builds may depend on PICO services/extensions. | Store links only. A developer's independently published GitHub/itch build can qualify separately. |
+| [Meta Horizon Store / former App Lab](https://www.meta.com/experiences/) | Official developer submissions. A free store entitlement is still an entitlement; no license bypass or authenticated store extraction. App Lab was folded into the main store in 2024. | Strongest Quest coverage; no supported anonymous APK-download API established. | Store links only; independently distributed free builds use their publisher source. |
+| [Khronos samples](https://github.com/KhronosGroup/OpenXR-SDK-Source) | Official upstream, Apache-2.0 sample; developer-published release APKs. GitHub API terms apply. | `hello_xr` Vulkan/OpenGL ES APKs; excellent OpenXR diagnostics. | Included in GitHub curated list, Vulkan variant selected. |
+| [Meta OpenXR samples](https://github.com/meta-quest/Meta-OpenXR-SDK) | Official upstream; check each sample's license. Source availability does not imply a published APK, and some samples require Meta extensions/services. | Source/build examples, inconsistent ready-made APK releases. | Link to upstream; add specific free APKs only after release/provenance review. |
+| [Godot XR demos](https://github.com/GodotVR/godot-xr-tools) | Official project source and publisher demo pages; licenses and dependencies vary by demo. | OpenXR examples on GitHub/itch. Older Godot builds can fail on Lepton's missing clipboard service. | Covered by source discovery; no compatibility promise from an OpenXR tag. |
+
+The table distinguishes observed restrictions from unknown permission. An
+unverified policy is a reason to defer automation, not a claim that a site is
+unlawful. Only the two implemented source kinds are registered by their own
+`sources()` functions; the other rows are recommendations, not new UI entries.
+
+## Adapters
+
+`ui/apk_sources/github.py` uses `github_curated.json`: Khronos `hello_xr`,
+[Open Brush](https://github.com/icosa-foundation/open-brush), and
+[SuperTux 3D](https://github.com/SgtBilko76/SuperTux-3D). These have official
+OpenXR project/release evidence, not a blanket claim of headset compatibility.
+Open Brush's compatibility evidence is recorded in [vr-apks.md](vr-apks.md).
+Open Brush and SuperTux publish the selected builds as prereleases; curated
+opt-ins preserve that label in version records. Exact APK filename patterns
+avoid downloading desktop archives or alternate non-Quest builds.
+[OpenSaberPlus](https://github.com/arpruss/OpenSaberPlus) was examined but not
+curated: GitHub reports its license as `NOASSERTION`, and current OpenXR APK
+provenance was not established in this pass.
+
+Default GitHub search is offline against this small list. Queries
+`topic:openxr`, `topic:oculus-quest`, and `topic:quest` explicitly call repository
+search. Results outside the curated list stay page-only, even if a repository
+claims an open-source license. This prevents an arbitrary tagged mirror from
+becoming a trusted downloader. Extend the curated JSON after provenance review.
+
+For a higher API quota, add a GitHub access token in the store's source
+settings (the GitHub row's token field) or through the server API
+(`POST /api/sources` with `action: "set-token"` or `"clear-token"` and
+`source: "github"`). `FRAME_GITHUB_TOKEN` in the process environment still
+works and wins over a saved token. A saved token is kept with repository
+logins in `source-credentials.json` (mode 0600, see
+[APK repositories](apk-repos.md#repositories-behind-a-login)). The source list
+reports only whether a token is set, where it came from and whether GitHub
+rejected it, never its value. Tokens are sent as `Authorization: Bearer` only to
+`api.github.com`, never written to the cache, never sent to asset hosts (public
+release downloads don't need them), and removed on redirects. If GitHub answers
+401 to the token, the request is retried once without it, as Obtainium does.
+The rejection is logged without the token and shown in source settings. The
+token is not tried again until it is saved again or the app restarts. Rate-limit
+backoff is kept per login (anonymous, or each token), so saving a token after
+anonymous requests were limited lets the next request go out with it. The
+adapter does not read `gh` credentials automatically. Metadata is cached for one hour under
+`frame_host.cache_dir('apk-sources', 'publisher')`. A cold details request
+fetches at most ten releases. Rate-limit errors are surfaced without retry
+loops. Asset IDs and release tags are not Android version codes: metadata
+leaves the latter unknown and rejects a requested `version_code` rather than
+silently fetching a different build.
+
+`itch.py` exposes separate OpenXR and Quest feed sources, so one feed's failure
+does not suppress the other at the aggregator level. Queries filter the current
+feed window locally: this is not an exhaustive historical itch search. Only
+explicit zero-price Android entries are returned. Covers are exposed in
+`images`; absent screenshots, APK version, ABI and minimum SDK stay unknown.
+Curated GitHub entries include publisher artwork and plain-language summaries.
+Repository image URLs are pinned to inspected commits. Open Brush screenshots
+come from its README-linked Steam listing; SuperTux uses the upstream gameplay
+preview embedded in the port's README (not a headset capture). The hello_xr
+sample has a launcher icon and GitHub social banner; no published screenshot
+was found in the inspected repository/README, so its screenshot list is empty.
+Uncurated topic results use the owner's avatar and GitHub's repository social
+preview. These are repository placeholders, not app screenshots. Itch's recorded
+RSS includes only covers, so screenshot lists remain empty without page scraping. VR is
+based on curated evidence or a VR-specific feed/topic, not a compatibility claim.
+
+Downloads stream to unique temporary files, require an APK manifest entry,
+restrict HTTPS origins and redirects, and enforce a 2 GiB ceiling. `verified`
+means the downloaded SHA-256 matches GitHub's published digest. Without such a
+digest, the computed SHA-256 is returned with `verified: False`; neither value
+claims publisher-signature validation. Installation must inspect the APK as
+usual. OBBs, split APKs, paid assets and external release-body download links
+are unsupported.
+
+## Evidence and limits
+
+On this Mac, Python 3.9 downloaded the real Khronos Vulkan 1.1.63 APK through
+the GitHub adapter, matched its published SHA-256
+`f24bbe8ba6f6339fca658628868ba8189cbc33390d6ac508f69d76fb67b5fa34`, and
+`python3 ui/frame_android.py info <apk>` exited 0: package
+`org.khronos.openxr.hello_xr.vulkan`, version code 1063, minimum API 24,
+arm64-v8a present, OpenXR detected. No Frame connection or installation occurred.
+
+The itch OpenXR RSS was fetched successfully and recorded as a fixture.
+Subsequent live adapter search encountered HTTP 429; it is not claimed as a
+successful live end-to-end search. Fixture search finds Off Nominal and parses
+nine Android entries from the ten-item feed (one has only an HTML platform).
+A real itch download and APK inspection were deliberately not performed:
+robots restrictions take precedence over that requested proof. No current
+policy text is claimed verified where the table records failed access.
+
+Tests use recorded, reduced API/RSS fixtures with network access blocked in
+the new test class. They cover selection, prereleases, unknown topic results,
+paid/non-Android exclusion, URL restrictions, redirect credential removal,
+caching, rate limits, checksum mismatch, non-APK rejection and partial-file
+cleanup. See `.claude/NOTES-more-sources.md` for commands and local evidence.
@@ -5,6 +5,12 @@ in **Lepton**, Valve's Waydroid-based container. Lepton is built for games,
 not general Android use
 ([GamingOnLinux](https://www.gamingonlinux.com/2026/09/lepton-from-valve-to-run-android-games-on-linux-is-now-open-source/)).

+**VR streaming clients:** WiVRn 26.9 and ALVR 20.14.1 install, but both fail
+OpenXR instance creation on the checked Frame because its Android runtime lacks
+`XR_KHR_convert_timespec_time` (**verified** 2026-09-28, SteamOS 0.4.1,
+BUILD_ID 20260925.6191901). They are not Frame Control dependencies. See
+[the feasibility results and options](linux-vr-streaming.md).
+
 ## Install from the Mac: one app, one Lepton instance (verified 2026-09-25)

 Use Frame Control's **Android apps** section (search, Install, Test, Report), drop
@@ -33,6 +39,12 @@ section below), instead of going into Lepton Development:
   run at once alongside Lepton Development, each in its own container
   (`lepton-steamlaunch-<id>`, ADB on 5556, 5557, …).

+Split APKs (a base plus `split_*.apk` files) install as one app:
+`python3 ui/frame_android.py install base.apk split1.apk …`. Lepton runs one
+APK, so splits that add native libraries or assets are merged into the base
+and re-signed; density and language splits are refused. See
+[quest-import.md](quest-import.md#split-apks). Not yet tried on a real Frame.
+
 Verified with AntennaPod and Tabletop Tools: installed in about 7 s, launched
 from the shortcut, stopped, and relaunched with their data intact. ADB and
 Lepton Development aren't involved.
@@ -46,6 +58,33 @@ Lepton Development must be installed once. Over SSH,
 `ssh frame 'steam steam://install/3056000'` queues it, but the install still
 needs to be confirmed or started in the headset.

+## When an app needs a newer Android
+
+Lepton is Android 11 (API 30), with arm64-v8a only. If Frame Control refuses
+an APK, it shows compatible versions from F-Droid's main and archive repos and
+IzzyOnDroid. It shows at most eight version names, newest first, preferring an
+arm64-only build, and says how many compatible builds it found in total.
+Each index is reduced to its compatible builds once a day and cached (about
+16 MB). The first lookup takes about 30 s and 100 MB of memory; later ones are
+instant.
+Choose **Install** to download a listed version, verify its SHA-256 against
+the index, and install it as its own app.
+
+You can also inspect a file or look up a package from the command line:
+
+```sh
+python3 ui/frame_android.py info some-app.apk
+python3 ui/frame_android.py versions some-app.apk
+python3 ui/frame_android.py versions org.example.app
+```
+
+The search links open APKMirror, APKPure, Uptodown, F-Droid and GitHub. Pick a
+version whose minimum is Android 11 or lower and that has an arm64-v8a build
+(or no native code). Frame Control does not fetch APKs from those search sites.
+Older versions may lack fixes, and being installable does not guarantee an
+app will run: see the missing services below. Android may refuse a downgrade
+or an update signed by a different publisher; removing the app deletes its data.
+
 ## Installed apps disappear when Lepton Development closes (verified 2026-09-25)

 Lepton Development runs in a throwaway "dev" context. When it exits for any
@@ -144,6 +183,63 @@ See
 [compat-db/README.md](../compat-db/README.md) and
 [apk-catalog/README.md](../apk-catalog/README.md).

+## Store and Library in Frame Control
+
+The Android tab has two views. **Store** finds apps across the sources (F-Droid,
+IzzyOnDroid, curated GitHub releases, itch.io). **Library** shows the apps on the
+Frame as cards, with Play and a menu for Test, Report and Remove; the display
+settings and recent reports are below it.
+
+The store groups apps by what a headset is good for, not by F-Droid's list:
+VR, Games, Emulators, Watch & listen, Read & learn, Create, Work, Web & chat
+and Tools (`ui/apk_sources/_categories.py`). A VR app is always in VR. Otherwise
+F-Droid's own categories decide, the finer one first (an app in *Multimedia* and
+*Social Network* is Web & chat).
+
+Apps built around a phone are hidden from browsing unless **Show phone-only
+apps** is on; a search always finds them, labelled *Phone app*. Hidden: apps in
+a category that only makes sense in a phone's own system (launchers, keyboards,
+wallpapers, Xposed modules), apps in a phone-side category (calls and SMS, VPNs,
+fitness, alarms, system tools) unless a useful category says otherwise
+(FluffyChat is in Phone & SMS, but it's a messenger), apps named for a phone
+part (a widget, a launcher, a battery tool), and apps that need hardware the
+Frame lacks (an IR blaster, a watch, NFC, root). They would install and run in
+Lepton, but do nothing useful on the Frame. In the October 2026 indexes that is
+about 1,600 of F-Droid's 4,500 apps and 550 of IzzyOnDroid's 1,300.
+
+Without a search, each source offers its best 300 for the chosen category:
+apps with a banner first, then with screenshots, then the newest.
+
+## The store in the headset: Frame Control's store as a panel (inferred, not yet run on the Frame)
+
+The **APK Store** panel is the Store above, drawn for the laser: the same
+sources, categories and artwork, with Install, Open, Stop and Remove, opened as
+its own floating panel:
+
+```sh
+scripts/store-on-frame.sh open     # copy the server over if it changed, start it, open the panel
+scripts/store-on-frame.sh close
+```
+
+`open` copies the same bundle the iPhone app uses
+(`ios/scripts/make_frame_bundle.py`) to
+`~/.local/share/frame-control/store-server/<version>` on the Frame. It then
+starts `ui/server.py` there with `FRAME_LOCAL=1` on `127.0.0.1:47812`, with a
+fresh `X-Frame-UI` key. Chromium XR (or the `org.chromium.Chromium` Flatpak)
+opens `/store` (`ui/store.html`) in a 1280×800 app window, tagged as panel
+`2001947812` ([panels.md](panels.md)). Downloads, SHA-256 checks and installs
+all run on the Frame, so the Mac can go away after `open`.
+
+The page has a category rail (Home, the store's categories, Installed), eight
+art cards a page, and a featured VR app on Home. It turns pages instead of
+scrolling, because a tagged panel gets mouse clicks and a drag would select
+text; the thumbstick (a mouse wheel) turns pages, and an on-screen keyboard is
+there for search. **Checked on the Mac only** (2026-10-14, 1280×800, against
+the real sources): Home, categories, Installed, the detail sheet. **Still to
+check in the headset:** that the panel opens and takes laser clicks, that an
+install completes when run on the Frame, and that **Open** brings up the app's
+panel.
+
 ## In-headset app store: F-Droid 1.17 (verified 2026-09-25)

 F-Droid 1.23 uses an old Compose and crashes on launch. **F-Droid 1.17.2**, the
@@ -298,3 +394,11 @@ because gamescope scales Lepton's surface to fit the same panel. Also unverified
 whether the settings survive the app or its Lepton instance relaunching.
 Lepton Development rebuilds its Android data on exit, so there they probably
 don't.
+
+## Expansion files and save backups
+
+SideQuest-inspired CLI helpers install local OBB files into an already-running
+app instance and back up/restore a stopped instance's private app data. See
+[SideQuest features and limits](sidequest.md) for commands, archive scope and
+verification status. These paths have offline coverage; real Frame storage and
+permissions remain unverified. They do not change APK install or launch behavior.
@@ -0,0 +1,174 @@
+# Backup and recovery
+
+**Tools → Backup & recovery** copies what a factory reset would lose from the
+Steam Frame to your computer, puts it back afterwards, and gets the headset out
+of Desktop Mode. It runs from Frame Control on a computer, not from the iPhone
+app or the Frame itself.
+
+**Partly tested on a headset.** On a Frame running SteamOS 0.5.6 (build
+20261015.6141207), a full backup saved 26 parts (40.7 MB): 14 games' Proton
+saves, Linux game saves, Steam and desktop settings and 9 Android apps' data,
+with no sign-in files or keys in the archives. Restoring one game's saves on top
+of identical files wrote nothing and left nothing behind. The session check
+reported Game Mode, so **Return to Game Mode** changed nothing. Not tested on a
+headset: a restore after a real factory reset, a restore that replaces files,
+Android data restore, and the switch out of a real Desktop Mode session. Unit
+tests and the fake Frame cover those (see [Verification](#verification)).
+On 2026-10-24 a read-only session check on the same build again reported Game
+Mode (`gamescope=true`, `plasma=false`, `nested_desktop=false`) and the
+`steamos-session-select` command was present. Opening the nested desktop for a
+clipboard test did not enter full Desktop Mode. A full session switch was not
+attempted because it restarts the headset session and a reliable return path
+was not established; the **Return to Game Mode** branch for a real full
+Desktop Mode session remains unverified.
+
+## Back up
+
+Choose **Game saves**, **Steam and desktop settings** and **Android app data**
+(all three by default), then **Back up now**. **See what's included** lists each
+part and its size first. Backing up only reads the headset.
+
+Each backup is a folder named after the time it started:
+
+```
+~/Documents/Frame Control Backups/2026-10-23-142530/
+  manifest.json
+  saves/saves-1234.tar.gz          one per game
+  saves/saves-linux.tar.gz
+  settings/settings-steam.tar.gz
+  settings/settings-system.tar.gz
+  android/org.example.game.tar.gz  one per Android app
+```
+
+`manifest.json` records the headset's name, SteamOS version and BUILD_ID, every
+file with its size and SHA-256, and each archive's SHA-256. It also lists what
+was skipped and why. A backup is written under a hidden `.…partial` name and
+renamed when complete, so a cut-off backup never shows up as one.
+
+### What's saved
+
+| Part | From the headset (`~` is `/home/steamos`) | Restored? |
+|---|---|---|
+| Windows game saves | `~/.local/share/Steam/steamapps/compatdata/<appid>/pfx/drive_c/users/steamuser/` → `AppData/Roaming`, `AppData/Local`, `AppData/LocalLow`, `Documents`, `Saved Games`. Wine's own folders in them (`Temp`, `Microsoft`, `D3DSCache`, `CrashDumps`) are left out. Sideloaded Windows games (non-Steam shortcuts) have prefixes here too | Yes |
+| Linux game saves | `~/.config/unity3d`, `~/.local/share/godot/app_userdata`: where Unity and Godot games save by default | Yes |
+| Steam controller layouts | `userdata/<account>/241100/remote/controller_config`, `steamapps/common/Steam Controller Configs/<account>/config` | Yes |
+| Steam library settings | `userdata/<account>/config/localconfig.vdf`, `shortcuts.vdf`, `grid/` (custom artwork) | Kept for reference |
+| Desktop settings | `~/.config/kdeglobals`, `kwinrc`, `kxkbrc` (keyboard layout), `konsolerc`, `dolphinrc` | Yes |
+| SteamVR settings | `~/.config/openvr/config/steamvr.vrsettings` | Kept for reference |
+| Android app data | Each Frame Control Android app's private data, through the existing `backup-data` helper ([sidequest.md](sidequest.md#private-app-data-backups)) | Yes |
+
+"Kept for reference" files are in the backup but not put back automatically:
+Steam and SteamVR rewrite them while they run, and an old `shortcuts.vdf` would
+point at apps that aren't installed after a reset. Open the backup folder to
+copy one by hand if you need it.
+
+### What's never copied
+
+- **Steam sign-in:** `steam.token`, `config/config.vdf`, `loginusers.vdf`,
+  `registry.vdf` and `ssfn*` files.
+- **Keys and passwords:** `~/.ssh`, `~/.gnupg`, keyrings and KWallet, any
+  `*.token`, `*.key` or `*.pem` file, Tailscale state and proxy settings
+  (`secrets.env`). These are refused by name wherever they turn up, including
+  inside a game's folder, and a restore refuses them too.
+- Games, Proton and the Steam runtimes: reinstall them from Steam.
+- Steam Cloud's own folders (`userdata/<account>/<appid>`), shader caches and
+  screenshots (**Tools → Capture & Live** saves those to Pictures).
+- Android apps themselves, their OBB files and shared storage (`/sdcard`).
+- Symbolic links (never followed), and single files over 2 GiB.
+
+Backups aren't encrypted. Game folders can still hold a launcher's login, and
+Android app data can hold an app's, so keep backups private.
+
+Android apps that are open are skipped (the helper needs the app closed); close
+them and back up again.
+
+## Restore after a factory reset
+
+1. Set up the headset again: Developer Mode, **Set Up Connection**, sign in to
+   Steam with the same account.
+2. Reinstall your games, and **start each once** so Proton creates its folders.
+3. Reinstall your Android apps from Frame Control.
+4. **Tools → Backup & recovery → Restore…** on a backup. Frame Control checks
+   each part against the headset and says what it would do: how many files it
+   replaces and how many are new.
+5. Tick the parts to restore and **Restore selected**. It asks before replacing
+   anything.
+
+Parts that can't go back yet are skipped, not failed, with the reason:
+
+| Reason | What to do |
+|---|---|
+| Install the game first | Install it from Steam, then restore |
+| Start the game once | Proton hasn't made its folders yet; start the game, quit, restore |
+| Sign in to Steam first | Controller layouts go under your account's folder |
+| Install the app first / close the app first | Android app data needs the app installed and closed |
+| The backup file doesn't match its checksum | The archive changed since it was made |
+
+A restore writes only the files in the parts you chose, and only inside that
+part's folders: it never deletes a file, and a save archive can't write outside
+its game. Before replacing a file it copies the current one to
+`~/.local/share/frame-control/restore-previous/<time>/` on the headset. Files
+already identical are left alone. Android app data goes through the existing
+`restore-data` helper, which keeps the previous data folder beside it.
+
+## Stuck in Desktop Mode
+
+On the Frame, Game Mode is gamescope running Steam, and the Linux desktop
+normally runs *inside* it as one panel (`steamos-nested-desktop`, see
+[how-the-frame-works.md](how-the-frame-works.md#the-layer-cake)). If that panel
+is what you see, you're already in Game Mode: close it from the Steam dashboard.
+
+Full Desktop Mode is a different session. SteamOS picks the session through
+SDDM's autologin file, `/etc/sddm.conf.d/zz-steamos-autologin.conf` (or the
+default `/etc/sddm.conf.d/steamos.conf`), and switches with
+`steamos-session-select` (`holo-session-select` on newer images). This is what
+Valve's own devkit tools on the Frame read and run
+([`steamos-get-status`](../frame/devkit-utils/steamos-get-status),
+[`steamos-delete`](../frame/devkit-utils/steamos-delete)).
+
+**Check** reads that file and whether gamescope and Plasma are running; it
+changes nothing. On the Frame gamescope's process is `gamescope-wl`, started by
+`start-gamescope-session`, and either counts as Game Mode. Only when Plasma is
+running without gamescope (Desktop Mode) does **Return to Game Mode** run
+exactly `steamos-session-select gamescope` (or `holo-session-select
+gamescope`) over SSH, after asking. It closes the desktop and anything open in
+it. Frame Control never edits the SDDM files itself, and runs nothing when the
+headset is in Game Mode, when it can't tell which mode it's in, or when the
+image has neither command.
+
+**Uncertain:** whether the Frame offers a full Desktop Mode session at all
+(only the nested desktop has been seen on BUILD_ID 20260922.6101926 and
+20260925.6191901), and whether `steamos-session-select` there works without
+the Developer Mode password. If the switch is refused, Frame Control says so and
+leaves the headset as it was.
+
+### Manual fallback
+
+If the button can't help, or the headset isn't reachable:
+
+1. **Restart it:** hold Power about 10 seconds until the light goes off, then
+   power on. On the Steam Deck, **Switch to Desktop** starts a one-shot
+   Desktop Mode session, so a restart brings Game Mode back. That the Frame
+   behaves the same is **inferred**, not tested.
+2. **Over SSH** (Developer Mode on), run Valve's switch yourself:
+   `ssh frame steamos-session-select gamescope`.
+3. **Boot menu:** hold Power 10 seconds, then power on while holding the AUX
+   button on top of Power. Choose `Current`, or `Previous` to start the other
+   SteamOS slot; both keep your data. Details and the further options
+   (`Repair Steam Installation`, `Erase User Data`, a full re-image) are in
+   [how-the-frame-works.md](how-the-frame-works.md#facts-worth-knowing) and
+   [recovery-and-images.md](recovery-and-images.md). Back up first if you can:
+   `Erase User Data` is a factory reset.
+
+## Verification
+
+`tests/test_backup.py` runs the headset-side script (`ui/frame_backup_remote.py`)
+for real against a folder laid out like the Frame's home, with only SSH faked:
+which paths are picked, that sign-in files and keys never are (also inside game
+folders), links never followed, the manifest and its checksums, restore checks
+after a reset (game not installed, no Proton folders yet, not signed in), that a
+restore replaces only its own files and keeps the old ones, unsafe archives
+(`..`, absolute paths, links, other games' folders) writing nothing, backup
+names that try to leave the backups folder, and the exact Game Mode command.
+`tests/e2e/test_recovery_kit.py` backs up a Proton save from the fake Frame through
+the server, deletes it and restores it. Neither involves a real headset.
@@ -0,0 +1,59 @@
+# Battery and health monitoring
+
+Open **Control Center → Battery & thermal health**, or expand **System
+information**, to read the Frame's charge, charging state, signed battery
+power, battery temperature and driver-reported health. Optional readings show
+charge cycles, full charge capacity, design capacity and full / design capacity.
+The existing battery chip also shows charging state and temperature.
+
+These are read-only readings from the Linux host's `/sys/class/power_supply`,
+not Android's container battery view. The probe prefers the documented Frame
+fuel gauge, `max1720x_bat_7-36`, over other battery nodes, excludes peripheral
+(`scope=Device`) and absent batteries, and otherwise uses a deterministic
+battery fallback. The UI shows the selected source path.
+
+| Reading | Linux attribute | Interpretation |
+|---|---|---|
+| Charge | `capacity` | Integer 0–100 percent; zero is a real reading |
+| Charging state | `status` | Charging, Full, Discharging or Not charging; missing/unknown states stay unavailable |
+| Battery power | `voltage_now × current_now` | Microvolts and microamps converted to signed watts; positive means charging on the Frame |
+| Temperature | `temp` | Tenths of a degree Celsius, including valid negative readings |
+| Battery health | `health` | Driver status such as Good or Overheat; not a wear percentage |
+| Charge cycles | `cycle_count` | Driver-reported count, including zero |
+| Full/design capacity | `charge_full`, `charge_full_design` | Microamp-hours converted to mAh |
+| Full / design capacity | Ratio of those two capacities | Fuel-gauge estimate; not a diagnosis or a guaranteed runtime |
+| Time estimate | `time_to_full_now`, `time_to_empty_now` | Seconds; displayed only for the corresponding charging/discharging state |
+
+Unavailable means the field is absent, unreadable or invalid. One missing
+sensor does not hide the others. Zero design capacity cannot produce a ratio. Time estimates outside 1 second
+to 7 days and battery temperatures outside −273.15 to 200 °C are rejected
+as invalid readings; these bounds are not thermal warning thresholds.
+The ratio may exceed 100% because full capacity is a gauge estimate, so the app
+does not clamp it or invent a wear warning. The app does not estimate health
+from charge percentage or assign temperature thresholds. Existing thermal
+alerts use the kernel's own hot/critical trip points.
+
+A connected charger does not necessarily mean charging: the Frame can still
+drain under load or sit idle near full charge. A 100% capacity reading alone
+does not prove the driver reports Full. Missing status is explicitly unknown;
+near-zero measured battery power with a charger is shown as not charging.
+Failed, stale or disconnected status samples clear Control Center readings.
+
+**No charge controls are provided.** Frame's SteamOS charge-limit backend is
+unsupported; see [charge-limit research](charge-limit.md). Monitoring never
+writes to power supplies, pauses charging, changes current or sets a limit.
+
+## Verification
+
+`tests/test_battery.py` uses temporary sysfs-like files for conversion,
+malformed/missing values, permission failures, source selection and partial
+sensor support. `tests/control_center.cjs` exercises charging states, optional
+metrics and failed/stale/disconnected sample handling. These tests do not
+connect to a headset. On 2026-10-24, the PR's uninstalled parser was also
+piped over SSH to a real Frame running SteamOS 0.5.6 (build
+`20261015.6141207`). Its read-only sysfs result selected
+`max1720x_bat_7-36` and reported 99% charge, Not charging, 24.9 °C,
+Good driver health, 14 cycles, 2,920 mAh full and 2,700 mAh design capacity
+(108.1% fuel-gauge estimate). The Control Center UI was not run against this
+headset; Windows and macOS were not tested. Optional fields are not claimed to
+be available on every firmware build.
@@ -0,0 +1,9 @@
+# Changelog
+
+## Unreleased
+
+- Battery monitoring now selects the Frame fuel gauge predictably, handles
+  missing or invalid readings, and shows optional cycle count, full/design
+  capacity and estimated capacity health in Control Center. Unknown charging
+  status stays explicit. Monitoring is read-only; charge limits remain
+  unsupported on Frame. The read-only parser was checked on a real Frame; the UI was not.
@@ -0,0 +1,168 @@
+# Battery charge limit (80%): research
+
+**Question:** can Frame Control set an 80% battery charge limit on the Frame
+over SSH or adb, safely?
+
+**Verdict: don't build it.** Valve's supported mechanism (the one Steam's own
+charge-limit setting uses on the Deck) is not configured for the Frame. The
+only control the Frame has is a vendor charger attribute that Valve's own
+charging daemon also manages. Research as of 2026-10-23, confirmed with the
+read-only checks below on a Frame running SteamOS 0.5.6 (build 20261015.6141207):
+
+- `steamosctl get-max-charge-level` fails with `UnknownInterface: ...BatteryChargeLimit1`.
+- `/usr/share/steamos-manager/devices/` has `[battery_charge_limit]` only for
+  the Steam Deck, ROG Ally, ROG Xbox Ally and MSI Claw, not the Frame.
+- Power supplies: `max1720x_bat_7-36` (Battery, driver `max1720x`),
+  `pm8550b-charger` (driver `qcom,pm8550b-charger`) and the USB-C `tcpm-source-psy-…`.
+  No `charge_control_end_threshold` anywhere; `pm8550b-charger` has
+  `charging_enable` and `charging_pause`, both `root`-only (`-rw-r--r--`).
+- `deckard-charger.service` (Deckard Battery Charger service) is enabled and running.
+
+For safe read-only readings in Control Center, see [battery and health
+monitoring](battery-health.md). It does not provide charging controls.
+
+## How Steam does it on other devices
+
+- Steam's **Settings → Power → Battery Charge Limit** shipped to the Deck in
+  SteamOS 3.7.8 (beta client 2025-05-08)
+  ([GamingOnLinux](https://www.gamingonlinux.com/2025/05/steamos-3-7-7-beta-brings-a-battery-charge-limit-option-and-improved-lenovo-legion-go-s-support/)).
+- Steam calls SteamOS Manager on the session bus:
+  `com.steampowered.SteamOSManager1.BatteryChargeLimit1`, with properties
+  `MaxChargeLevel` (`i`, read/write; `-1` resets to the default) and
+  `SuggestedMinimumLimit` (`i`, read-only). The CLI is
+  `steamosctl get-max-charge-level` / `set-max-charge-level` /
+  `suggested-minimum-charge-limit`. Source: `data/interfaces/com.steampowered.SteamOSManager1.xml`
+  and `steamos-manager/src/bin/steamosctl.rs` in
+  [holo/steamos-manager](https://gitlab.steamos.cloud/holo/steamos-manager),
+  read at commit `83981ab` (2026-10-21).
+- The root daemon writes the value to sysfs only if the device's TOML in
+  `data/devices/` has a `[battery_charge_limit]` section. Two methods exist
+  (`steamos-manager/src/power.rs`, `BatteryChargeLimitMethod`):
+  - `hwmon_attribute`: the Deck writes `max_battery_charge_level` on
+    `steamdeck_hwmon` (`data/devices/steam-deck.toml`).
+  - `acpi_sb`: ROG Ally, MSI Claw and others write the standard
+    `charge_control_end_threshold` on the ACPI battery
+    (`/sys/bus/{platform,acpi}/drivers/…battery…/PNP0C0A:00/…/power_supply/*/`).
+    On ARM there's no ACPI, so this path can't match there either.
+- With no `[battery_charge_limit]` section, the interface fails with
+  `UnknownInterface` (Steam can still show its toggle on unsupported hardware:
+  [steam-for-linux#12044](https://github.com/ValveSoftware/steam-for-linux/issues/12044)).
+
+## What the Frame has
+
+- **Upstream `data/devices/steam-frame.toml`** (matched by
+  `dt.compatible = "qcom,sm8650"`, variant `Deckard`) configures GPU clocks
+  and performance profiles only. It has **no `[battery_charge_limit]`**, so
+  SteamOS Manager on the Frame has no charge-limit backend. Its git history
+  (2026-09-10 to 2026-10-12) never added one.
+- **Battery driver:** the Frame's battery is
+  `/sys/class/power_supply/max1720x_bat_7-36`, and the charger is a
+  `tcpm-source-psy-…` USB-PD supply ([how-the-frame-works.md](how-the-frame-works.md)).
+  Both were verified 2026-09-25, and `ui/frame_status.py` `battery()` and
+  `power_source()` read them. The Maxim MAX1720x is a **fuel gauge** and
+  can't control charging. Mainline `drivers/power/supply/max1720x_battery.c`
+  exposes only read-only properties (capacity, voltage, current, temp,
+  charge_full…), with no `charge_control_end_threshold`. Valve's build is
+  patched: it reports `time_to_full_now`, which mainline doesn't have. Even
+  so, a charge stop would have to come from the separate charger IC, not this
+  node.
+- **Not the qcom_battmgr path.** Some other SM8650/SM8750 handhelds get
+  `charge_control_end_threshold` from Qualcomm's `qcom_battmgr`
+  (pmic_glink) driver, clamped at 55% or above. A third-party distro enabled
+  Steam's limit on those by patching steamos-manager
+  ([armada-os/armada#758](https://github.com/armada-os/armada/pull/758)). The
+  Frame's battery isn't exposed by `qcom_battmgr`, so that doesn't carry over.
+  (Unverified: whether a `qcom_battmgr` node also exists on the Frame. The
+  read-only checks below answer this.)
+- **What third parties found:**
+  [y0usaf/frame-charge](https://github.com/y0usaf/frame-charge) reports that
+  the Frame has no `charge_control_end_threshold` and falls back to a vendor
+  `charging_pause` attribute. It also reports that `constant_charge_current`
+  changes on its own, which it attributes to Valve's `deckard_charge_daemon`
+  (`deckard-charger.service`). By its own README, that tool has only run its
+  read-only `status` on a Frame and has **not** been tested against real
+  charging. Two questions are still open: whether pausing stops the current,
+  and whether Valve's daemon undoes it.
+
+## Why not build it now
+
+- **No supported endpoint.** The only Valve-supported interface,
+  `BatteryChargeLimit1`, isn't backed on the Frame. Writing sysfs ourselves
+  would bypass Steam and SteamOS Manager, and it needs root (`sudo -S` with
+  the user's password).
+- **A rival owner.** Valve's charge daemon already adjusts charging
+  parameters. Toggling `charging_pause` underneath it could be undone without
+  warning, fight it in a loop, or leave charging paused.
+- **Stuck at a low charge.** If the app or the Frame dies while charging is
+  paused, the Frame could sit at a low charge on the charger. Whether a
+  reboot clears `charging_pause` is unknown.
+- **The UI would mislead.** Steam's setting (if it shows at all) wouldn't
+  reflect our limit.
+- **adb doesn't help.** Wi-Fi adb (`frame:5555`) reaches the Lepton Android
+  container only. USB-C adb reaches the Linux side as `steamos` (non-root, the
+  same as SSH, [ssh.md](ssh.md)). Neither adds a charging-control route.
+  Android's `dumpsys battery` in Lepton is a container view, not the hardware.
+
+**Revisit when** a SteamOS update adds `[battery_charge_limit]` to
+`steam-frame.toml` (or `steamosctl get-max-charge-level` stops returning
+`UnknownInterface` on the Frame). Then the feature is just a SteamOS Manager
+call:
+
+- Read: `steamosctl get-max-charge-level`
+- Set: `steamosctl set-max-charge-level 80`
+- Undo: `steamosctl set-max-charge-level -- -1`
+
+That runs as `steamos` with no sudo. It's the same value Steam's own
+Settings → Power shows, so both stay in sync.
+
+## Headset checks (read-only first)
+
+Run these over SSH as `steamos`. Steps 1–4 only read.
+
+```sh
+# 1. Does SteamOS Manager offer the interface on this build?
+steamosctl get-max-charge-level; steamosctl suggested-minimum-charge-limit
+busctl --user introspect com.steampowered.SteamOSManager1 \
+  /com/steampowered/SteamOSManager1 | grep -i -A3 ChargeLimit
+grep -n battery_charge_limit /usr/share/steamos-manager/devices/*frame* 2>/dev/null
+
+# 2. Which power supplies exist, and which drivers back them?
+for d in /sys/class/power_supply/*; do
+  echo "== $d ($(cat $d/type)) driver=$(basename "$(readlink -f $d/device/driver)" 2>/dev/null)"
+  ls $d
+done
+
+# 3. Any charge-control attributes, with their values and permissions?
+ls -l /sys/class/power_supply/*/{charge_control_*,charging_pause,charging_enable,constant_charge_current*,charge_behaviour} 2>/dev/null
+grep -H . /sys/class/power_supply/*/{charge_control_*,charging_pause,charging_enable,constant_charge_current*,charge_behaviour} 2>/dev/null
+
+# 4. What is Valve's charger daemon, and what does it touch?
+systemctl status deckard-charger.service --no-pager
+systemctl cat deckard-charger.service
+journalctl -b -u deckard-charger.service --no-pager | tail -50
+```
+
+If step 1 succeeds on some build, test `steamosctl set-max-charge-level 80`
+on a charger at a charge above 80%. Watch `current_now` on the
+`max1720x_bat_*` node fall to about 0, then `steamosctl set-max-charge-level -- -1` and
+confirm charging resumes. Do this on a supervised Frame only. If only step 3
+finds a vendor attribute, **stop**: don't write it, and leave the question
+with Valve.
+
+## Sources
+
+- steamos-manager source, commit `83981ab` (2026-10-21):
+  `data/devices/steam-frame.toml`, `data/devices/steam-deck.toml`,
+  `data/interfaces/com.steampowered.SteamOSManager1.xml`,
+  `steamos-manager/src/power.rs`, `steamos-manager/src/manager/user.rs`.
+  [gitlab.steamos.cloud/holo/steamos-manager](https://gitlab.steamos.cloud/holo/steamos-manager)
+- Mainline [`max1720x_battery.c`](https://github.com/torvalds/linux/blob/master/drivers/power/supply/max1720x_battery.c)
+- [ValveSoftware/steam-for-linux#12044](https://github.com/ValveSoftware/steam-for-linux/issues/12044):
+  the Steam toggle shows on unsupported hardware
+- [armada-os/armada#758](https://github.com/armada-os/armada/pull/758):
+  `qcom_battmgr` threshold on other SM8650/SM8750 devices
+- [y0usaf/frame-charge](https://github.com/y0usaf/frame-charge): third-party,
+  untested on real charging
+- [GamingOnLinux: SteamOS 3.7.7 beta charge limit](https://www.gamingonlinux.com/2025/05/steamos-3-7-7-beta-brings-a-battery-charge-limit-option-and-improved-lenovo-legion-go-s-support/)
+- Repo: [how-the-frame-works.md](how-the-frame-works.md) (battery and
+  charger nodes), [ssh.md](ssh.md) (adb scope), `ui/frame_status.py`
@@ -0,0 +1,67 @@
+# Computer use through Frame Control MCP
+
+The MCP transport can carry semantic actions or visual computer-use actions.
+The limits are the Frame's underlying interfaces, permissions and whether an
+action can be targeted and verified. A stereoscopic headset screenshot alone
+is not a reliable coordinate system for clicking a particular app window.
+
+## What exists, and the right route
+
+| Surface | Evidence and route | Remaining work or boundary |
+|---|---|---|
+| Frame management | **Verified:** existing SSH/HTTP operations for status, capture and file transfer work through MCP. Typed install/launch/power tools wrap the existing API. | Extend typed operations before adding generic mouse automation. Preserve explicit approval for consequential changes. |
+| App/window observation | **Verified 2026-09-29:** `computer_state` reads gamescope X11 window/app/process triples, focused app and the installed AT-SPI library. | Bounded to 96 accessible nodes and six levels. Trees may be truncated, stale, hidden or incomplete. Snapshot paths and XIDs are observations, never durable action permissions. |
+| Chromium page content | **Verified previously:** the assistant rendered and could be exercised through CDP in an isolated Frame Chromium profile. | A shipped click/type surface needs exact owned browser/target binding, fresh element references, lifecycle cleanup, consent and post-action readback. Do not expose unrestricted JavaScript or attach to arbitrary existing profiles automatically. |
+| Steam UI | **Verified 2026-09-29:** the AT-SPI service listed the Steam client's Chromium process and frame nodes, but child traversal was incomplete. Existing `frame_steam.py` uses Steam's loopback CDP endpoint for specific operations. | Prefer those narrow Steam interfaces. Presence of AT-SPI does not prove controls are actionable, and generic pointer injection is not proved for VR menus. |
+| Other Linux apps | **Verified 2026-09-29:** Frame ships libX11, libXtst and libatspi; `/dev/uinput` is writable by the current user. | Library presence and access permissions do not prove that a game accepts input. Global virtual input can affect whichever app has focus. Do not ship a blind keyboard/mouse tool on this evidence alone. |
+| Panel focus and layouts | **Documented in [#41](https://github.com/saphid/frame-control/pull/41):** `POST /api/panels` accepts `list`, `focus` and `open`. Focus was verified there. | Reuse that owned interface after integration. Its tested gamescope-owned overlay transform setters return `PermissionDenied`; no reliable saved spatial-layout interface was established. Do not duplicate its implementation here. |
+| Shared keyboard/trackpad | **Documented in [#19](https://github.com/saphid/frame-control/pull/19):** `/api/input` supplies state/start and event submission, implemented with a bundled KDE Connect daemon. | This branch does not import, launch or depend on that daemon. The user's own-implementation rule remains authoritative. A first-party input implementation or permitted bundled-library route needs its own delivery evidence before MCP integration. |
+| Physical/device boundaries | **Documented:** an asleep Frame may be off the network; power authorization can require the user's password; physical pairing and headset fit/comfort require the user. | MCP cannot bypass offline hardware, consent, compositor permissions or physical verification. Keep explicit human handoffs. |
+
+## Reusing the existing computer-use work
+
+**Documented:** the installed `cua-driver` skill has the right control pattern:
+observe an exact window, use a semantic target if available, fall back to pixels
+from that same snapshot, then read back the result. Its browser route requires
+an exact process/window/target binding and session-scoped element references.
+Those are useful design rules for Frame tools.
+
+**Verified locally 2026-09-29:** `cua-driver describe get_window_state` describes
+host-local process/window IDs and macOS AX inspection. It does not establish an
+SSH Frame target. The installed skill's advertised Linux companion file is
+missing. A native ARM64 Frame backend, its dependencies and remote transport
+have not been verified. We therefore do not claim that the existing Mac driver
+can control the Frame by passing it a Frame PID or screenshot, and we do not
+make the feature depend on installing that application.
+
+Frame Control's `computer_state` is our own Python implementation over installed
+platform libraries. It sends the probe over SSH stdin, writes no helper to disk,
+and exits after one observation. Missing displays/libraries return explicit
+errors; a 15-second process deadline prevents a stalled accessibility call from
+leaving a probe behind. Window names and accessibility text are untrusted app
+content, never instructions to an agent.
+
+**Recommended next implementation:** an isolated Chromium session with typed
+snapshot/click/type/scroll tools and exact fresh target binding, then individually
+verified native app actions. Use the headset capture to judge appearance, not to
+invent a screen-to-window coordinate transform. Direct tool calls must retain
+approval rules; a generic computer-use tool must not become a route around the
+MCP approval panel, install confirmation or power confirmation.
+
+## Isolated browser input proof
+
+**Verified 2026-09-29, SteamOS 0.4.1, BUILD_ID 20260925.6191901:** a temporary
+Frame Chromium profile loaded a local test page through an SSH reverse tunnel.
+CDP `Input.insertText` entered the test string in its own input. A CDP
+`Input.dispatchMouseEvent` press/release on its own button copied that string
+to the page's result; DOM readback matched exactly. The browser profile,
+loopback forwards and panel log were removed afterward. No user app was typed
+into, no global settings were changed and no third-party helper app was used.
+
+AT-SPI did **not** expose the test page's controls in that same probe, even with
+Chromium's renderer-accessibility flag. It returned the partial Steam-client
+tree instead. The reason remains **unverified**; this is an evidence gap, not
+proof that Frame accessibility cannot work. For a first implementation,
+Chromium's proven page-specific CDP route is stronger than assuming complete
+AT-SPI coverage. This proof does not ship unrestricted click/type tools or
+establish input delivery to SteamVR's menus.
@@ -0,0 +1,241 @@
+# Headsets, addresses and the connection
+
+Frame Control can manage more than one Steam Frame, and each headset can be
+reached at more than one address: a LAN IP at home, another at the office, its
+mDNS name (`frame.local`), its Tailscale IP or MagicDNS name. The **Devices**
+tab (key 5) lists them. The connection pill in the header says whether the one in
+use is connected and how (Tailscale, or the network's name); click it to add,
+edit or reorder that headset's addresses, or to see each step of connecting.
+
+The code is in three modules, all stdlib-only Python on your computer:
+
+| Module | What it does |
+|---|---|
+| `ui/frame_devices.py` | The registry: headsets, their addresses, networks; importing and updating `~/.ssh/config`; pinned host keys |
+| `ui/frame_network.py` | Which network this computer is on, and Tailscale's state |
+| `ui/frame_link.py` | The connector: finds the headset, keeps the SSH connection, publishes each stage; the Devices API |
+
+## Headsets
+
+Each headset keeps its own SSH alias, as Set Up Connection has always written
+it: the first is `frame`, the next `frame-2`, and so on. Terminal's
+`ssh frame-2` and the helper scripts (`FRAME_ALIAS=frame-2 scripts/push.sh …`)
+work for each one.
+
+- **Nothing to migrate by hand.** On first start, the app imports every
+  `# >>> steam-frame (ALIAS) >>>` block in `~/.ssh/config` as a headset, with
+  the block's HostName as its first address. It also copies the host key your
+  `known_hosts` already trusts for that address into the headset's own
+  known_hosts file, `~/.ssh/frame-control-hosts/<id>`, so nobody is asked to trust it again.
+- **Add a headset** runs Set Up Connection (`scripts/connect.sh` on macOS,
+  `ui/frame_connect.py --alias NAME` elsewhere) in a terminal with a new alias.
+  When it writes its block, the app picks the headset up by itself. If Set Up
+  Connection runs again and finds a headset somewhere new, that address is added
+  at the top of its list.
+- **Use this headset** (or the switcher in the header, or the app's
+  **Frame → Headset** menu) moves the whole app to another headset; every panel
+  reloads from it. From that moment no command goes to the previous headset, even
+  if the new one never answers. It waits while an install is running, since an
+  install reads the SSH settings step by step.
+- **Remove** forgets a headset. Its `~/.ssh/config` block stays unless you tick
+  the box; either way it isn't imported again unless Set Up Connection changes it.
+- A plain `FRAME_ALIAS` that Set Up Connection never configured still works: the
+  app shows it as not set up and lets ssh's own config decide where it goes.
+
+## Addresses
+
+Each address has a kind (LAN, mDNS, Tailscale or Other, guessed from the address
+and changeable), an optional label, the networks it has worked on, and when it
+last worked with its round-trip time.
+
+When connecting, the app **tries all addresses at once** (TCP to the SSH port)
+and ranks them:
+
+1. addresses that worked on the network this computer is on now;
+2. mDNS names;
+3. Tailscale addresses, if Tailscale is running here;
+4. addresses not tried on this network yet;
+5. addresses that only ever worked on other networks;
+6. Tailscale addresses while Tailscale is off.
+
+Your order (on the Devices tab, or in the pill's dialog) breaks ties. The best-ranked address that answers
+wins; one that answers first waits up to 0.35 s for a better-ranked one that is
+still trying. If SSH to the winner fails in a way another address could fix
+(a different device answered there, or the link dropped), the next one that
+answered is tried. Every success records the network on that address, so next
+time on that network it's tried first.
+
+**Test now** probes every address and tries SSH on each one that answers, without
+disturbing the connection in use: "SSH works", "answered as a different
+headset", "refused this computer's key", or why it didn't answer. **Find on
+Tailscale** lists your tailnet's devices (likely headsets first, from `tailscale
+status --json`, including the Mac app's own CLI) with buttons to add their
+MagicDNS name or IP. **Find on this network** asks mDNS for SteamOS devkit
+services and checks `ALIAS.local` and `frame.local`.
+
+**The pill's dialog** lists the same addresses, with what each one answered, and
+can add, edit, reorder and remove them without leaving the page you're on. When
+the headset reports a LAN IP on the same network as this computer and that IP
+isn't saved, it offers to add it. The offer puts the address first in the list,
+so on that network it wins over the Tailscale name; away from home the Tailscale
+name still leads. A new or edited address is tested straight away. While
+connected, **Reconnect** applies your changes now rather than at the next
+connection: it tries the addresses again, ranked as above, and the best-ranked
+one that answers promptly wins. It doesn't pick a particular address.
+
+## Networks
+
+A network is told apart by its default gateway: the router's IP address plus its
+hardware (MAC) address, read with `route`/`arp` (macOS), `ip route`/`ip neigh`
+(Linux) or `route print`/`arp -a` (Windows). That works on wired networks, and
+on macOS 14 and later, which hides the Wi-Fi name from apps without Location
+permission. Where the system does share the Wi-Fi name, it's shown, and you can
+name any network yourself ("Home Wi-Fi") on the Devices tab.
+
+The app rereads the gateway every 5 seconds and Tailscale's state every
+30 seconds. Changing networks reconnects.
+
+## The connection, stage by stage
+
+The connector runs in the server (`frame_link.Link`) and moves through:
+
+1. **Checking this computer's network**: gateway, Wi-Fi, this computer's IP, Tailscale.
+2. **Finding the headset**: each address resolving, trying, answered in N ms,
+   no answer, refused, or can't be found.
+3. **Opening SSH** to the address that answered.
+4. **Checking the headset's identity**: the host key must match the one pinned
+   for this headset.
+5. **Logging in** as the headset's user.
+6. **Connected** via network N, address A, round trip T; or **failed** at a stage
+   with the reason in plain words and a countdown to the next try (5, 10, 20,
+   then every 30 seconds). Retry now skips the wait.
+
+Stages 3 to 5 come from following `ssh -v` as it runs. On macOS and Linux the
+connection is an SSH ControlMaster that every command shares; when it dies (the
+headset slept or left the network) the connector notices and starts again. On
+Windows, where OpenSSH can't share a connection, the same handshake runs once
+and each command then connects on its own; a command that can't reach the
+headset makes the connector start again. There ssh's stderr goes to a file that
+the connector reads as it grows, not a pipe: Windows' OpenSSH can block writing
+to a redirected pipe.
+
+Once connected, every `ssh`, `scp` and `rsync` the app runs gets
+`-o HostName=<address> -o HostKeyAlias=frame-control-<id>
+-o UserKnownHostsFile=~/.ssh/frame-control-hosts/<id> -o HashKnownHosts=no -o User=… -o Port=…`. The
+alias's block in `~/.ssh/config` is also updated to the last address that
+worked (and to the user and port you set), so Terminal's `ssh frame` and the
+scripts follow. Edits to `~/.ssh/config` take a lock file
+(`~/.ssh/config.frame-control.lock`) that Set Up Connection takes too, and never
+write over a change someone else made since the app last read the file.
+
+**Host keys are pinned per headset, not per address.** Your own `known_hosts`
+is keyed by address, so a different device answering at a remembered IP (a DHCP
+lease that moved) would look like a new host there. The app keeps one known_hosts
+file per headset instead, so saving or forgetting one headset's key never touches
+another's: a different device answering at one of its
+addresses is refused, and the pill says so. A headset's first connection trusts
+the key it shows, as Set Up Connection does: while a headset has never had an
+identity here, every ssh command gets `StrictHostKeyChecking=accept-new` (the
+first contact saves it, whichever command makes it), and from then on
+`StrictHostKeyChecking=yes`. `accept-new` never replaces a saved key, so it
+can't hide a change. The saved key's fingerprint is also recorded in
+`devices.json` (`identity`), along with a flag that never clears
+(`identity_established`), as soon as ssh saves the key: even if the login is then
+refused. A headset counts as having had an identity if the flag is set, a
+fingerprint is recorded, its pin file exists at all, or a connection to it ever
+worked (older files get the flag from that when loaded, so editing or removing
+the address that worked doesn't undo it); then
+a pin file that's deleted, empty, unreadable or not parseable **fails closed**
+(*Checking the headset's identity* fails, "missing or can't be read") rather than
+trusting whatever answers, and only **Check its key…** saves a key again. It
+says when the key it reads matches the recorded fingerprint.
+
+**When the key changes** (SteamOS reinstalled or reset, a new headset at the
+same address, or a different device answering there), nothing is accepted by
+itself. The connection fails at *Checking the headset's identity*; the banner
+and the headset on the Devices tab say so, with **Check the new key…** (also
+**Check its key…** on the Devices tab at any time). That reads the key the
+headset shows now, and nothing more: ssh writes it to a throwaway known_hosts
+file during the key exchange, before any login, and is stopped as soon as it
+has. Every other kind of contact is ruled out on the command line, ahead of
+anything `~/.ssh/config` says: no authentication of any kind, no agent
+(`IdentityAgent=none`, `ForwardAgent=no`), no X11 or port forwarding, no local
+or remote command (`-N`), no terminal, and `-F none`. Through a ProxyJump host
+it builds that hop itself, as a ProxyCommand whose ssh forwards nothing and runs
+no local command (ProxyJump's own ssh would take `ForwardAgent` and
+`LocalCommand` from `~/.ssh/config`); it still logs in to the jump host as
+configured. Every word of that ProxyCommand is quoted for the shell that runs it
+(and `%` doubled for ssh's own tokens), the forward is `[%h]:%p` so an IPv6
+headset works, and an IPv6 jump host is passed bare with its port. Through a ProxyCommand from `~/.ssh/config`, or several jump hosts,
+it refuses, since those can't be made safe from here: add an address it can
+reach directly, or compare the key by hand. It shows the key's SHA256 fingerprint next to
+the saved one, with the command that prints it on the headset
+(`ssh-keygen -lf /etc/ssh/ssh_host_ed25519_key.pub`, in Konsole on its desktop).
+**This is my headset: trust its new key** then saves exactly that key, replacing
+the old one, and reconnects. The server keeps the key it showed for 10 minutes
+and saves it only if the page sends back the same fingerprint, and only while
+the headset's port, user and that address are unchanged; it never saves whatever
+answers next. The API's old `forget-identity`, which deleted the pin so that the
+next contact saved any key unseen, is refused. Set Up Connection can't fix a changed key: it checks your
+own `known_hosts`, not the headset's file.
+
+**When the headset refuses this computer's key** (`Permission denied
+(publickey,…)`), the connection stops at *Logging in*, and the banner and the
+Devices tab offer **Set Up Connection…**, which pairs this computer again.
+
+Neither is reported as a fault: the diagnostics report each kind at most once
+a session ([privacy.md](privacy.md)).
+
+## One server at a time
+
+Only one Frame Control server runs per user (a lock file, `server.lock`, in the app's
+data folder). Two would each connect, reconnect and edit the headsets on their own, and
+one could move the other's install to a different headset. A second one, say
+`scripts/frame-ui.sh` while the app is open, exits with "Frame Control is already
+running". `FRAME_CONTROL_DATA_DIR` gives a separate one, with its own headsets.
+
+## API
+
+All under the usual `/api/` guards (loopback `Host`, `X-Frame-UI` header).
+
+| Request | Returns |
+|---|---|
+| `GET /api/connection` | The connection state: `phase` (connecting, connected, failed), `device`, `network`, `stages`, `probes`, `via`, `error`, `retry_at`, `tests`, `version` |
+| `GET /api/connection/events` | The same as server-sent events, one each time it changes (the page reads it with `fetch`, since `EventSource` can't send the header) |
+| `GET /api/devices` | Headsets, the current network, known networks, the next free alias |
+| `GET /api/devices/tailscale?id=` | Tailscale peers, likely headsets first |
+| `GET /api/devices/mdns?id=` | Headsets found on this network |
+| `POST /api/devices` | `{"action": ...}`: `use`, `update` (name, user, port), `remove`, `address-add`, `address-update`, `address-remove`, `address-move`, `test`, `identity-scan` (read the key the headset shows: `scan` with `type`, `fingerprint`, `host`, `saved`, `recorded`, `matches`, `recorded_match`), `identity-trust` (`fingerprint`: save the key `identity-scan` showed), `forget-identity` (refused: use the two before), `name-network`, `setup` (alias, optional host), `retry` |
+
+Every host, alias and user is checked against strict patterns before it's
+stored, because they end up in ssh arguments and `~/.ssh/config`; nothing goes
+through a shell.
+
+## The registry file
+
+`devices.json` in the app's data folder (`~/Library/Application Support/Frame
+Control` on macOS, `%APPDATA%\Frame Control` on Windows,
+`~/.local/share/frame-control` on Linux). It's plain JSON so the iPhone app can
+share the format later (it still connects to one host; see
+[iphone.md](iphone.md)):
+
+```json
+{"version": 1, "active": "f67f8b7e",
+ "devices": [{"id": "f67f8b7e", "name": "Steam Frame", "alias": "frame", "user": "steamos", "port": 22,
+              "identity_files": ["~/.ssh/id_ed25519_frame"], "identity": ["SHA256:…"], "identity_established": true,
+              "addresses": [{"host": "frame.local", "kind": "mdns", "label": "",
+                             "networks": ["n-e0998baa61"], "last_ok": 1790593550.4, "last_rtt_ms": 0.9}]}],
+ "networks": {"n-e0998baa61": {"name": "Home Wi-Fi", "ssid": null, "gateway": "192.168.1.1",
+                               "gateway_mac": "b4:fb:e4:b5:67:55", "wifi": true, "last_seen": 1790593550.0}}}
+```
+
+A network id is `n-` and the first 10 hex digits of SHA-1 of `gateway|mac`.
+
+## Tests
+
+`tests/test_devices.py`, `tests/test_network.py` and `tests/test_link.py` run
+with the other unit tests. They use a stand-in `ssh` (`tests/fakessh/ssh`) that
+prints what `ssh -v` prints and plays a ControlMaster, real sockets on this
+computer for the addresses, and temporary folders for `~/.ssh`
+(`FRAME_CONTROL_SSH_DIR`) and the app data (`FRAME_CONTROL_DATA_DIR`), so they
+never touch yours.
@@ -0,0 +1,34 @@
+{
+  "date": "2026-10-20T03:53:25Z",
+  "build_id": "20260925.6191901",
+  "app": "F-Droid Basic 2.0.2 (org.fdroid.basic, version code 2000052)",
+  "input": "One real APK plus synthetic main and patch OBBs named with older version code 2000050",
+  "sibling_discovery": true,
+  "automatic_import": "passed",
+  "closed_after_import": true,
+  "independent_readback": [
+    "5023b41e7ac80f8b26a868db1f66884976bf864fab81a070ad37539f3481b352  /sdcard/Android/obb/org.fdroid.basic/main.2000050.org.fdroid.basic.obb",
+    "aecbed7433e81aaeef40942069d03ad23130f09d93b7e78bbe224865a9573157  /sdcard/Android/obb/org.fdroid.basic/patch.2000050.org.fdroid.basic.obb"
+  ],
+  "cleanup": {
+    "test_install_removed": true,
+    "running_instances_before": {},
+    "running_instances_after": {},
+    "battery_before": [
+      "100",
+      "Full",
+      "Charging"
+    ],
+    "battery_after": [
+      "99",
+      "Discharging",
+      "Charging"
+    ],
+    "shared_lock_released": true
+  },
+  "limits": [
+    "No headset rendering or gameplay assertion; headset unworn.",
+    "Synthetic expansion files verify transport and persistence, not Resident Evil 4 compatibility.",
+    "A game that exits before its Lepton instance can accept data was not tested."
+  ]
+}
@@ -0,0 +1,47 @@
+Frame client feasibility, 2026-09-28
+SteamOS VERSION_ID=0.4.1 BUILD_ID=20260925.6191901; uname -m=aarch64
+SteamVR: vrserver log reports 2.18.1; process 2256 remained alive across checks.
+Base: dcf9689f6459e576d35fc507eb702ca6b2bf4dad (main).
+Unmodified upstream release APKs; installed with ui/frame_android.py install APK --vr.
+No Linux host attached. No pairing, streamed video, input, audio or worn-headset checks.
+Selected logcat lines only; timestamps in Android logs are UTC.
+
+WiVRn-release.apk
+https://github.com/WiVRn/WiVRn/releases/tag/v26.9
+sha256=1df6649ec77224fcc821af0ab4897222bdf3d7eb6ce6ad636461336724111331
+E/OpenXR-Loader( 1140): Error [GENERAL | xrCreateInstance | OpenXR-Loader] : LoaderInstance::CreateInstance, no support found for requested extension: XR_KHR_convert_timespec_time
+I/WiVRn   ( 1140): [2026-09-28 12:07:58.987] [WiVRn] [info] Failed to create OpenXR instance version 1.1.58: XR_ERROR_EXTENSION_NOT_PRESENT
+E/OpenXR-Loader( 1140): Error [GENERAL | xrCreateInstance | OpenXR-Loader] : LoaderInstance::CreateInstance, no support found for requested extension: XR_KHR_convert_timespec_time
+I/WiVRn   ( 1140): [2026-09-28 12:07:59.034] [WiVRn] [info] Failed to create OpenXR instance version 1.0.58: XR_ERROR_EXTENSION_NOT_PRESENT
+E/WiVRn   ( 1140): [2026-09-28 12:07:59.035] [WiVRn] [error] Error during initialization: Failed to create OpenXR instance: XR_ERROR_EXTENSION_NOT_PRESENT
+
+alvr_client_android.apk
+https://github.com/alvr-org/ALVR/releases/tag/v20.14.1
+sha256=be68feeb02665e3d69f1cdbcabf38ea4d15c42868a7ec6b5e698dbefee4e4e36
+E/OpenXR-Loader( 1139): Error [GENERAL | xrCreateInstance | OpenXR-Loader] : LoaderInstance::CreateInstance, no support found for requested extension: XR_KHR_convert_timespec_time
+I/RustStdoutStderr( 1139): Error [GENERAL | xrCreateInstance | OpenXR-Loader] : LoaderInstance::CreateInstance, no support found for requested extension: XR_KHR_convert_timespec_time
+E/[ALVR NATIVE-RUST]( 1139): ALVR panicked: What happened:
+E/[ALVR NATIVE-RUST]( 1139): panicked at alvr/client_openxr/src/lib.rs:220:10:
+E/[ALVR NATIVE-RUST]( 1139): called `Result::unwrap()` on an `Err` value: ERROR_EXTENSION_NOT_PRESENT
+
+Cleanup verified: both app directories, compatdata directories and Steam shortcuts absent.
+Both test containers stopped and removed. No Steam/SteamVR restart or global setting changes.
+
+Valve release notes fetched from ISteamNews/GetNewsForApp/v2 (appid=250820).
+
+SteamVR Beta Updated - 2.18.1
+https://steamstore-a.akamaihd.net/news/externalpost/steam_community_announcements/1844751498219787
+Added tethered Quest support over USB (must be used with Steam Link Beta)
+
+Introducing SteamVR 2.17
+https://steamstore-a.akamaihd.net/news/externalpost/steam_community_announcements/1843481262693486
+Adds initial support for USB streaming. Note: Requires new Steam Client Beta.
+Fix crash using Steam Link on Linux when games submit invalid textures.
+Improve streaming recovery when using Steam Link on Linux.
+
+SteamVR Beta Updated - 2.17.8
+https://steamstore-a.akamaihd.net/news/externalpost/steam_community_announcements/1842212951314598
+Fix crash using Steam Link on Linux when games submit invalid textures.
+Improve streaming recovery when using Steam Link on Linux.
+Adds initial support for USB streaming.
+USB streaming can be used without WiFi by opting into the Steam Client Beta.
@@ -0,0 +1,29 @@
+Locked real-Frame repeat, 2026-09-29
+SteamOS 0.4.1, BUILD_ID 20260925.6191901; aarch64.
+mkdir /tmp/frame-test.lock succeeded before installs/launches; rmdir issued after cleanup.
+Preflight battery 44%, charging; before WiVRn 46%, before ALVR 47%, cleanup 47%.
+Original Steam PID 49823 and vrserver PID 49571 present after cleanup.
+Same unmodified upstream release APKs and SHA-256s as 2026-09-28.txt.
+Installer functions loaded from a613735 before checkout was fast-forwarded to current main.
+No compatibility layer injected. Per-app immersive Lepton instances, Steam shortcut launches.
+Headset unworn. No Linux gaming host. No pairing or streaming session reached.
+
+WIVRN: selected journal lines (local +1000 prefix, Android timestamps UTC).
+Sep 29 11:03:15 frame lepton-steamlaunch-2817846116[1967]: 09-29 01:03:15.202  1153  1181 E OpenXR-Loader: Error [GENERAL | xrCreateInstance | OpenXR-Loader] : LoaderInstance::CreateInstance, no support found for requested extension: XR_KHR_convert_timespec_time
+Sep 29 11:03:15 frame lepton-steamlaunch-2817846116[1967]: 09-29 01:03:15.210  1153  1181 I WiVRn   : [2026-09-29 01:03:15.210] [WiVRn] [info] Failed to create OpenXR instance version 1.1.58: XR_ERROR_EXTENSION_NOT_PRESENT
+Sep 29 11:03:15 frame lepton-steamlaunch-2817846116[1967]: 09-29 01:03:15.248  1153  1181 E OpenXR-Loader: Error [GENERAL | xrCreateInstance | OpenXR-Loader] : LoaderInstance::CreateInstance, no support found for requested extension: XR_KHR_convert_timespec_time
+Sep 29 11:03:15 frame lepton-steamlaunch-2817846116[1967]: 09-29 01:03:15.256  1153  1181 I WiVRn   : [2026-09-29 01:03:15.256] [WiVRn] [info] Failed to create OpenXR instance version 1.0.58: XR_ERROR_EXTENSION_NOT_PRESENT
+Sep 29 11:03:15 frame lepton-steamlaunch-2817846116[1967]: 09-29 01:03:15.257  1153  1181 E WiVRn   : [2026-09-29 01:03:15.257] [WiVRn] [error] Error during initialization: Failed to create OpenXR instance: XR_ERROR_EXTENSION_NOT_PRESENT
+Screenshot API exit 0; 1920x1080 uniformly dark image; no client scene visible.
+
+ALVR: selected journal lines (local +1000 prefix, Android timestamps UTC).
+Sep 29 11:03:36 frame lepton-steamlaunch-2831623938[1967]: 09-29 01:03:35.553  1139  1167 E OpenXR-Loader: Error [GENERAL | xrCreateInstance | OpenXR-Loader] : LoaderInstance::CreateInstance, no support found for requested extension: XR_KHR_convert_timespec_time
+Sep 29 11:03:36 frame lepton-steamlaunch-2831623938[1967]: 09-29 01:03:35.553  1139  1165 I RustStdoutStderr: Error [GENERAL | xrCreateInstance | OpenXR-Loader] : LoaderInstance::CreateInstance, no support found for requested extension: XR_KHR_convert_timespec_time
+Sep 29 11:03:36 frame lepton-steamlaunch-2831623938[1967]: 09-29 01:03:35.611  1139  1167 E [ALVR NATIVE-RUST]: panicked at alvr/client_openxr/src/lib.rs:220:10:
+Sep 29 11:03:36 frame lepton-steamlaunch-2831623938[1967]: 09-29 01:03:35.611  1139  1167 E [ALVR NATIVE-RUST]: called `Result::unwrap()` on an `Err` value: ERROR_EXTENSION_NOT_PRESENT
+Screenshot API exit 0; 1920x1080 uniformly dark image; no client scene visible.
+
+Cleanup: both test app directories, compatdata, shadercache, containers and shortcuts absent.
+Capture output directory removed. No global settings changed; Steam/SteamVR not stopped.
+The shared lock was subsequently acquired by another thread (new directory timestamp 11:03:49 +1000).
+Native clients and Valve host streaming not exercised: no native build or Linux gaming host available.
@@ -0,0 +1,128 @@
+# Mod feasibility test, 2026-09-28
+
+**Verified observations**, with limits below. Device: `aarch64`, SteamOS
+`VERSION_ID=0.4.1`, `BUILD_ID=20260925.6191901`; Proton version file:
+`1788505046 proton-11.0-2c-arm64`; `vrcmd --stats`: SteamVR `2.18.1`.
+Times below are the Frame's AEST clock.
+
+## Inventory and allowed content
+
+`ssh frame 'python3 - owned' < ui/frame_steam.py` returned 868 games before
+the test. Beat Saber (620980) and Skyrim VR (611670) were absent. Half-Life 2:
+VR Mod – Episode One (2177750) was installed. Hogwarts Legacy (990080),
+Horizon Zero Dawn (1151640) and Horizon Forbidden West (2420110) were listed
+but not installed. The full personal library is not committed.
+
+**Documented, owner report after these tests:** Alex has Beat Saber on a
+Quest 2, which is charging. That copy was not accessed or inspected during
+this test. Its version, transfer path and Frame compatibility remain unknown;
+Steam ownership is still not established.
+
+Gravitas's public Steam metadata reported `is_free: true`, developer Galaxy
+Shark Studios, Windows only. Frame Control's existing Steam helper requested
+its install. Steam first returned free-license state 3, then config state 7,
+then queued the download. The helper did not accept state 3; a later read
+found state 7. The source of that transition was not observed. A later
+`install 1067310` returned `state: installed`. No purchase occurred.
+
+UEVR was downloaded from the author's [1.05 release](https://github.com/praydog/UEVR/releases/tag/1.05),
+not a mirror. `UEVR.zip` was 7,399,455 bytes. Its SHA-256 matched the author's
+`UEVR.zip.sha256` (a UTF-16 text file):
+
+```text
+af4f2f91306802d7ee4e8497d483a547ac8e9a3067dbafb81324100524215d3c
+```
+
+Microsoft's Windows x64 .NET runtime and Windows Desktop runtime 6.0.36 ZIPs
+were downloaded using URLs in the official release metadata. Both SHA-512
+hashes matched that metadata. They were extracted into a test-only user
+directory, not installed globally. No other mod manager was used.
+
+## Half-Life 2 VR: visible setup, not gameplay
+
+Launched the already-installed Episode One using
+`steam steam://rungameid/2177750`. The existing manifest recorded build
+25413453 and a shared base depot from app 658920, build 25413418.
+
+Selected fresh Steam / SteamVR log lines:
+
+```text
+22:07:50 proton waitforexitandrun .../Half-Life 2 VR/ep1vr.exe
+22:08:03 SetApplicationPid: Setting app steam.app.2177750 PID to 27990
+22:08:03 Successfully loaded binding file '.../hlvr/cfg/steamvr/bindings_frame.json' for app 'steam.app.2177750'.
+```
+
+The game's `episodicvr/console.log` reached `Creating VR hand HUD...`,
+`Creating VR weapon HUD...` and `Calibrating VR base position`. It also
+contained missing material/weapon warnings. SteamVR logged a missing
+`frame_hmd` binding as well as the successful controller binding load;
+controller input was not exercised.
+
+`ui/frame_vrshot.py` produced a stereo capture showing the mod's
+“First time setup” and “Dominant hand” dialog. This establishes visible
+startup, not a played level or comfortable performance. The capture includes
+room passthrough and is deliberately not published. The test's `hl2.exe`
+process was terminated; the pre-existing game installation was preserved.
+
+## Gravitas and UEVR: injection unverified
+
+The game was launched through Steam. The injector was then started in the
+same `steamapps/compatdata/1067310` prefix using the shipped
+`SteamLinuxRuntime_4-arm64/_v2-entry-point` and Proton 11 ARM64. The first
+attempt reported:
+
+```text
+Application: UEVRInjector.exe
+Message: You must install .NET to run this application.
+Architecture: x64
+App host version: 6.0.35
+.NET location: Not found
+```
+
+With `DOTNET_ROOT` and `DOTNET_ROOT_X64` pointing at the test-only Windows
+runtime directory, Proton loaded `Microsoft.NETCore.App/6.0.36` and
+`Microsoft.WindowsDesktop.App/6.0.36`. A 25-second launcher timeout was too
+short to establish whether the UI worked. A longer attempt produced a
+625×372 `UEVR` X11 window on display `:1`. This is window creation evidence,
+not a successful injection. That launcher returned exit 0; this was not
+treated as proof that a mod worked.
+
+A combined launch set `PROTON_REMOTE_DEBUG_CMD` to `UEVRInjector.exe` and
+ran Gravitas's `Drop.exe` in the same runtime/prefix. The process
+`SkyArk/Binaries/Win64/Drop-Win64-Shipping.exe` and a 1600×900 window titled
+`SkyArk (64-bit, PCD3D_SM5)` appeared. The launcher exited **1** before an
+injection or stereo game scene could be verified. Output included:
+
+```text
+Proton: Error while copying to ".../windows/system32/amdxcffx64.dll": No such file or directory
+Error [GENERAL | xrCreateInstance | OpenXR-Loader] : xrCreateInstance failed
+X Error of failed request: BadWindow (invalid Window parameter)
+  Major opcode of failed request: 10 (X_UnmapWindow)
+X Error of failed request: XI_BadDevice (invalid Device parameter)
+  Minor opcode of failed request: 28 (X_GetDeviceButtonMapping)
+```
+
+These errors do **not** establish an ARM64/FEX incompatibility. Other work
+was launching apps on the shared Frame. SteamVR's PIDs changed during the
+experiment; `ps` recorded the replacement `vrserver` and `vrcompositor`
+starting at **22:13:03**, corroborated by `steamvr.service` journal startup
+lines. This test did not request a Steam/SteamVR restart. The combined
+attempt ran at 22:14:17–22:14:27, after that restart. A stable, coordinated
+session is needed to distinguish launcher/environment problems from game or
+mod incompatibility.
+
+## Cleanup and remaining checks
+
+- No game executable or OpenVR DLL was replaced. No global runtime or power
+  setting was changed by this test. No R.E.A.L., OpenComposite or Beat Saber
+  payload was installed.
+- Removed the test-only UEVR/.NET directory and downloaded ZIPs from the
+  Frame. Final process checks found no `UEVRInjector.exe`,
+  `Drop-Win64-Shipping.exe`, `ep1vr.exe` or `hl2.exe`. SteamVR was running.
+- Gravitas and its Steam-created prefix remain installed for a repeat test;
+  the pre-existing HL2 VR install remains. Steam may retain normal shader
+  caches, logs and prefix temporary files.
+- Still unverified: UEVR injection and removal, R.E.A.L. releases and
+  permissions, OpenComposite, Beat Saber playback on either build, and our
+  own manager's end-to-end install/uninstall. No installer UI is justified
+  by these results. See the [support table and next checks](../mods.md).
@@ -0,0 +1,122 @@
+# UEVR from Frame Control, 2026-09-29
+
+**Verified observations**, with limits below. Same device and software as
+[2026-09-28](mods-2026-09-28.md): aarch64, SteamOS `0.4.1`, BUILD_ID
+`20260925.6191901`, Proton `11.0-2c` ARM64, SteamVR `2.18.1`. Times are the
+Frame's AEST clock. Headset tests ran under the shared
+`/tmp/frame-test.lock`, taken 11:13–11:28, 11:33–11:49, 12:02–12:40,
+12:44–12:52 and 13:01–13:03. Battery 45 % →
+93 %, on charge throughout.
+
+**Not observed:** the VR image, head tracking and controller input. SteamVR
+reported `activity_level` 3 (user interaction timeout; the headset was not
+being worn) with `average_target_fps` 1. Every `frame_vrshot.py` stereo
+capture was a uniform dark frame. Frame submits prove the app is presenting
+to SteamVR, not that the image is correct.
+
+## Manual injection (11:13–11:28)
+
+Game launched with `steam steam://rungameid/1067310`. `UEVRInjector.exe` was
+started with the game process's Wine variables, and its WPF window drew
+through DXVK's D3D9. It listed `Drop-Win64-Shipping (pid: 320) (SkyArk
+(64-bit, PCD3D_SM5))`, a Wine-side process id.
+
+- `--attach=Drop-Win64-Shipping` didn't inject. The button still read
+  **Inject**, and the saved `OpenVRRadio=True` in `AppData/Local/praydog/…/user.config`
+  was not restored on the next start.
+- XTest clicks landed on the game: `XQueryPointer` reported child
+  `0x2000001` (the game) at the injector's coordinates, and
+  `XLowerWindow`/`XRaiseWindow`/`XSetInputFocus` didn't change that.
+- `XSendEvent` button events to the injector window worked. After OpenVR and
+  Inject, `/proc/<game>/maps` contained
+  `drive_c/frame26/uevr/UEVRBackend.dll` and `openvr_api.dll`, next to
+  Proton's `vrclient_x64.dll`. UEVR's log:
+
+```text
+[11:17:32.948] Hooking D3D11
+[11:17:33.020] Hooked DirectX 11
+[11:17:34.263] Framework initialized
+[11:17:34.475] [VR] Requested runtime: openvr_api.dll
+[11:17:39.791] UGameViewportClient::Draw called for the first time.
+[11:17:39.792] is stereo enabled called!
+```
+
+It also logged repeated `Failed to initialize Framework on DirectX 12` (probing
+before it chose D3D11) and failed Unreal scans, for example `Failed to locate
+r.EnableStereoEmulation cvar` and `Failed to find FSceneView constructor`.
+SteamVR loaded UEVR's action manifest and `bindings_oculus_touch.json` for
+`steam.app.1067310`. `vrcmd --stats` then showed `"key" :
+"steam.app.1067310"` with `frame_submits` 4439, then 5506 a few seconds later.
+
+A second session's injector exited within about 20 seconds. The third,
+started fresh, injected again (`frame_submits` 1412). After the game was
+stopped, a plain launch had no UEVR modules and no SteamVR scene app. All test
+files were removed.
+
+## Through Frame Control (11:33–11:49)
+
+Local `ui/server.py` against the `frame` alias. First the API, then the
+page's **VR mod** dialog, driven with a headless Chromium:
+
+| Step | Result |
+|---|---|
+| `status` for 2177750 (HL2 VR) | `eligible: false`; `install` refused ("doesn't look like an Unreal Engine game") |
+| `install` for 1067310 | 3 s. Three downloads, hashes matched, receipt written |
+| `start`, game running | First run refused: `the UEVR window is 960 px wide, not 625`. It was measured before WPF laid it out; `start` now waits for the width to settle. Rerun: injected, `frame_submits` 27 (36 s) |
+| `uninstall` while running | Refused: "the game is running; quit it first" |
+| `start`, game not running | Launched the game, injected, `frame_submits` 45 (55 s) |
+| UI: Remove, Install | Worked; the prefix listing matched its pre-install state except an empty `UnrealVRMod`, which uninstall now also removes |
+| UI: Play in VR | Failed once: the injector died with `Process terminated. … Resource name: Arg_AccessViolationException`, after `Fontconfig error: No writable cache directories` (it had no `HOME`). Fixed by keeping the normal environment under the game's (the relaunch-on-failure added next is gone with the GUI injector; see below) |
+| UI: Play in VR ×4 | 4 of 4 injected. The three cold starts took 54, 58 and 58 s; frame submits 56, 71, 80 |
+| UI: Remove | `drive_c`, `AppData/Roaming` and `AppData/Local` listings matched the pre-install state; download cache and receipt gone |
+
+## The GUI injector wasn't reliable (12:02–12:40)
+
+After review fixes, cold starts through the API with UEVR's own injector:
+
+- Batches of 4, 6 and 8 runs: 3/4, 4/6 and 6/8 injected. A retry that
+  relaunched the injector didn't rescue a bad session. In both failures of
+  the last batch, the first injector ignored the clicks and the next two
+  never showed a window within 60 s.
+- Captures of the ignoring injector (`xwd`) showed its first-drawn frame,
+  with OpenVR still unselected. The injector's log had no error.
+
+Lock released at 12:40 before redesigning.
+
+## Frame Control's own injector (12:44–12:52)
+
+`start` now runs `frame_inject.py` under python.org's embeddable Python
+3.14.7 (SHA-256 `d297e5ff…1f15`, matching python.org's release-file API), in
+the game's Wine session. The old install was removed and the new one
+installed without the headset lock, since neither launches anything. The
+prefix listing after removal matched the pre-install state; the new install
+is 47 MB.
+
+| Step | Result |
+|---|---|
+| API `start`, cold | Injected in 35 s, `frame_submits` 39. UEVR log: `Hooked DirectX 11`, `Framework initialized`, `Requested runtime: openvr_api.dll`; `config.txt` has `Frontend_RequestedRuntime=openvr_api.dll` |
+| API `start` ×8, cold | **8/8** injected, 33–34 s each; `frame_submits` 33–42 at the check |
+| UI: dialog while running | "UEVR is running in the game now." Remove disabled (screenshot in `docs/img/uevr-dialog.png`) |
+| UI: Remove → Install → Play in VR → Remove | Each worked; Play in VR injected (`frame_submits` 106 a few seconds later) |
+| After the last Remove | `drive_c`, `AppData/Roaming` and `AppData/Local` listings matched the pre-install state; download cache and receipts gone; no game or `python.exe` processes. The empty `/tmp/frame-control-mods.lock` was deleted by hand (tmpfs) |
+
+Battery 98–100 % on charge. Lock released at 12:52.
+
+## After the second review (13:01–13:03)
+
+`frame_inject.py` now declares every Win32 signature, checks each result,
+frees its remote buffers and closes its handles; uninstall accepts only our
+folder and UEVR's own settings folders. Rechecked: install; three cold starts
+injected in 35, 34 and 33 s (`frame_submits` 42, 38, 47), UEVR logging
+`Framework initialized` and `Requested runtime: openvr_api.dll`; uninstall
+left `drive_c`, `AppData/Roaming` and `AppData/Local` as before, with no cache,
+receipts or processes. Frame Control's injector total: 12 of 12 cold starts
+through the API, plus the UI run.
+
+## Left on the Frame
+
+- Gravitas (free, 7.4 GB, installed 2026-09-28 for this work) is still
+  installed; nothing else from this work is left.
+- No `UEVRInjector.exe` or game processes were left running. Steam and
+  SteamVR were running at the end.
+- The test lock was released at 13:03.
@@ -0,0 +1,53 @@
+{
+  "date": "2026-09-28",
+  "os": {
+    "version": "0.4.1",
+    "build": "20260925.6191901",
+    "variant": "vr"
+  },
+  "performance": {
+    "compositorFps": 72.0,
+    "frameMs": 13.89,
+    "appFps": null,
+    "gpuMs": 3.07,
+    "compositorCpuMs": 0.61,
+    "cpuPercent": 40.6,
+    "gpuMHz": 903.0
+  },
+  "batteryPercent": 16,
+  "maxTempC": 73.5,
+  "ownership": [
+    {
+      "id": 1009850,
+      "owned": false,
+      "installed": false,
+      "frame": 0
+    },
+    {
+      "id": 1173510,
+      "owned": false,
+      "installed": false,
+      "frame": 0
+    },
+    {
+      "id": 1068820,
+      "owned": false,
+      "installed": false,
+      "frame": 0
+    },
+    {
+      "id": 908520,
+      "owned": false,
+      "installed": false,
+      "frame": 0
+    },
+    {
+      "id": 1494460,
+      "owned": false,
+      "installed": false,
+      "frame": 0
+    }
+  ],
+  "hudLifecycle": "open, duplicate-open, close passed before control-test pause; overlay probe removal verified",
+  "comfort": "Initial 1cm seated probe restored exactly. Later commits/readback disagreed while Frame in use; Alex paused control tests. No controls shipped."
+}
@@ -0,0 +1,58 @@
+# Independent review attempts — 2026-09-28
+
+Target: the implementation and evidence in
+[3fb541d](https://github.com/saphid/frame-control/commit/3fb541d) and
+[6a63a5a](https://github.com/saphid/frame-control/commit/6a63a5a), supplied as a
+frozen diff before those commits were made. No executable code changed after
+the final review snapshot.
+
+Requested model: **SWE-2 Max**, explicitly selected with `--model swe-2-max`.
+No completed verdict or model self-identification was returned. This is not a
+passed review, and there is no "no actionable findings" claim.
+
+## First attempt
+
+Launcher:
+
+```sh
+devin -p --model swe-2-max --permission-mode auto --respect-workspace-trust false --prompt-file /tmp/frame-vr-review-prompt.txt
+```
+
+The prompt required read-only review, no delegation, no edits and no Frame
+access. The reviewer inspected surrounding code, then stopped while checking
+the public OpenVR header. The tool runner reported:
+
+> warning: rejected a tool call that requires confirmation. Running in non-interactive mode.
+
+The real launcher exit status was **0**, but no verdict was returned. A zero
+process status here is not evidence that the review completed.
+
+## Tool-free retry
+
+Launcher:
+
+```sh
+devin -p --model swe-2-max --permission-mode auto --respect-workspace-trust false --prompt-file /tmp/frame-vr-review-final-prompt.txt
+```
+
+The self-contained prompt supplied the complete frozen changes, surrounding
+code, standards and locally fetched authoritative OpenVR header excerpts. It
+explicitly prohibited tools, edits, delegation and device access. This avoided
+the first attempt's permission boundary without escalating permissions.
+
+No output or verdict arrived within the fifteen-minute review window. The
+process was sent SIGTERM at 918 seconds; the shell recorded real exit status
+**143**. Findings are unavailable. Review must be completed before considering
+this partial draft ready; playspace feasibility is also still paused.
+
+## Other validation
+
+- 166 Python unit tests passed locally.
+- 8 website tests and UI JavaScript syntax passed locally.
+- Desktop and phone-width attached-preview checks passed using live telemetry;
+  paid optional install buttons were disabled, and unavailable metrics cleared.
+- [Fake-Frame CI](https://github.com/saphid/frame-control/actions/runs/36423548487/job/108931878908)
+  passed, as did Windows/Linux server tests and the main checks job. Docker was
+  unavailable locally. macOS/iOS jobs were still queued at this handoff.
+- Real-device evidence and the paused-control limitation are in
+  [VR utilities](../../vr-utilities.md).
@@ -0,0 +1,122 @@
+# Family and comfort
+
+Frame Control's Home tab has a **Family and comfort** card, on desktop and
+on iPhone. No third-party notification or parental-control app is needed.
+This is Frame Control code using Python, Steam and SteamVR already on the Frame.
+
+![Family and comfort controls in the desktop app](img/comfort-desktop.png)
+
+## Sessions
+
+Set a limit of 1–240 minutes, optional break and check-in intervals, then
+**Start session**. Break and check-in intervals of 0 turn those reminders off.
+**Cancel session** cancels the timer and monitoring without changing the game.
+Cancel before starting a session with different settings.
+
+The Frame shows a one-minute warning, then opens Steam Home in its dashboard.
+**Games stay running**: save and pause before the limit. Some games pause when
+the dashboard opens; others do not. There is no kill, power-off, Steam restart,
+account restriction or parental lock. The wearer can return to the game.
+
+**Documented implementation:** the timer is a single, opt-in Python worker in
+the Frame user's account. Desktop and iPhone share its state. It keeps going
+when the companion disconnects, closes or is suspended. It exits after
+completion or cancellation (normally within five seconds). Cancellation waits
+for any in-flight SteamVR action to finish within its timeout; it is not a boot
+service. A Frame reboot invalidates the session. Suspend counts toward the
+limit, using Linux's boot-time clock. If a warning was delayed by suspend or a
+SteamVR failure, Home waits until at least a full minute after a successful
+warning. A failed Home transition remains active and retries, with an error
+shown in the companion. A stale worker is reported as unverified enforcement.
+
+## Alerts and breaks
+
+During a session, battery, overheating and check-in alerts go to connected
+companions. Break reminders and session warnings also appear on the headset.
+
+- **Low battery:** 15% or below while discharging. One alert until charging or
+  recovery to 20%, so values around 15% do not produce repeated notifications.
+- **Overheating:** a thermal zone reaches its own kernel-reported hot/critical
+  trip, or the battery reports `Overheat`. Missing sensors mean unknown, not
+  safe. These are status alerts, not medical advice or an extra thermal governor.
+- **Check in:** an alert after the chosen number of active minutes.
+- **Breaks:** a SteamVR reminder and companion notification at the chosen interval.
+
+**Inferred:** SteamVR activity levels 1 and 2 are a useful proxy for use, not
+proof someone is wearing the headset. Inactive readings reset continuous use;
+missing readings add no time. Long gaps count at most 30 seconds. Breaks and
+check-ins are distinct from the elapsed-time session limit.
+
+Click **Enable / test notifications** on each companion. iOS asks for permission;
+macOS, Windows and Linux follow their notification settings. The page also shows
+recent events and errors. Keep Frame Control open and connected for companion
+alerts. **Phone alerts are local, not push notifications:** iOS suspension,
+force-quit or a lost SSH connection prevents live delivery. Old alerts are not
+replayed as a notification burst on reconnect. Headset warnings and the session
+limit continue without the phone. A physical iPhone's background delivery has
+not been verified and is not guaranteed.
+
+## Casting
+
+**Cast to this screen** (Home → **Right now**) starts the existing headset Live view and requests full
+screen where supported. Show that screen to people in the room, or use the
+computer/phone's own screen mirroring. It creates no new stream transport,
+public URL or LAN server. iPhone uses the inline viewer if full screen is not
+available. The image includes private content visible to the wearer.
+
+## What is installed
+
+The shared authenticated `/api/comfort` endpoint copies three bundled Python
+files to `~/.cache/frame-control/comfort/<content-hash>/`. Session state and
+locks live in `~/.local/state/frame-control/comfort/`, with a private directory
+and 0600 state file. There is no network listener or system service. Cancel a
+session before removing these directories. The iPhone's normal server still
+exits on disconnect; the explicitly started comfort worker is the exception.
+
+## Verification
+
+**Verified 2026-09-28**, SteamOS 0.4.1, build `20260925.6191901`: shipped
+`/opt/steamvr/bin/linuxarm64/vrcmd --notify TEXT` reported success for a custom
+reminder. Steam's CDP `SteamUIStore.Navigate('/library/home')` and
+`SteamClient.OpenVR.VROverlay.ShowDashboard('valve.steam.gamepadui.main')`
+opened Home while the running app ID stayed unchanged. Prior page and dashboard
+visibility were restored. Kernel hot/critical trips and SteamVR activity were
+read from the real device. No temperature or battery fault was induced.
+
+**Verified locally:** deterministic fake-Frame tests cover late warnings,
+failed warnings/Home actions, cancellation, activity gaps, thresholds, duplicate
+suppression, reboot invalidation, shared session state and the exact Home
+JavaScript. `python3 -m unittest discover -s tests` runs them. The iOS Simulator
+build tests notification content and bounds. Physical iPhone delivery and
+wearer-perceived headset notification visibility remain unverified.
+
+**Verified end to end on the same Frame:** a two-minute session with no companion
+connection for 135 seconds emitted its warning, break and check-in, then opened
+Home. The running app ID was unchanged; the test restored the previous page and
+dashboard visibility and confirmed the worker exited. Casting through the Home
+shortcut decoded the existing headset stream at 30 fps.
+
+**Verified on the iOS 26.5 Simulator:** connected to the real Frame, approved the
+notification prompt, and saw the native Frame Control test banner. Seven iOS
+tests passed.
+
+![Native test notification in the iOS Simulator](img/comfort-notification-ios.png)
+
+Desktop and 390-pixel phone layouts had no horizontal overflow.
+On macOS the development Electron app's real notification attempt was denied
+(`UNErrorDomain` 1); the bridge now returns that failure instead of reporting
+success. Successful macOS/Windows/Linux notification display remains unverified.
+
+**Verified on the real Frame:** its naturally discharging 15% battery produced
+one low-battery event during a short session; the test then cancelled the
+session. Overheating alerts use fake sensor samples in tests: the shared
+headset was not deliberately overheated.
+
+**Verified 2026-09-29 on the same Frame:** a fresh one-minute session opened
+Home more than 60 seconds after the successful warning. The test restored the
+previous page and dashboard visibility. Local regression coverage now includes
+slow notification delivery, a total Home-action timeout, failed worker startup,
+unreadable saved state, malformed activity samples and notification UX: 173
+Python tests passed. Desktop and 390-pixel layouts were checked again; system
+notification-denial guidance stayed visible across polls. Initial event history
+did not replay notifications, and only the latest new event was announced.
@@ -15,6 +15,10 @@ depends on SSH working through the `frame` alias from `scripts/connect.sh`.
 | KDE Connect | KDE Connect on both | Guess | There's a macOS build of KDE Connect, but whether it's present or installable on the Frame wasn't confirmed. It would give you clipboard sync, file send, and remote input. Worth checking on-device. |
 | microSD | Physical card | Confirmed that the slot exists ([Wikipedia](https://en.wikipedia.org/wiki/Steam_Frame)) | Offline fallback. |

+Screenshots and Steam clips have their own buttons on **Tools → Capture &
+Live**, and **Settings → Copy to this computer** copies new ones here
+automatically: see [recording.md](recording.md#copying-new-ones-automatically).
+
 ## Clipboard

 `scripts/paste-to-frame.sh` sends the Mac clipboard (or stdin) to the
@@ -36,3 +40,34 @@ desktop has to be running in the headset. It's text only, and pastes over about

 A simpler fallback: `ssh frame 'cat > ~/clip.txt'` < file, then open it in the
 headset.
+
+### In Frame Control
+
+**Tools → Type on the headset** (also on Home, and press `T` anywhere) has the
+clipboard next to the keyboard and trackpad, both ways:
+
+- **Send text** or **Send this computer's clipboard** puts text on the headset's
+  clipboard, as above.
+- **Copy the headset's clipboard** brings it back: copy something in a desktop
+  app on the Frame, then press it, and paste on your computer. It calls
+  Klipper's `getClipboardContents` on the same D-Bus bus, and sends the text
+  back as base64, so accents, emoji and line breaks arrive unchanged. The
+  desktop app and the iPhone app put it on their own clipboard. With the server
+  in a browser, the server sets the clipboard of the computer it runs on:
+  `pbcopy` on macOS, PowerShell's `Set-Clipboard` (or `clip.exe`) on Windows,
+  `wl-copy`, `xclip` or `xsel` on Linux.
+
+It's text only, and copying back takes up to 1 MB; a bigger clipboard is
+refused with a message, so send big things as a file. An empty clipboard, or
+one holding an image, says so. Neither the server nor its log ever records what
+was on the clipboard.
+
+**Verified on a Frame, 2026-10-24:** SteamOS 0.5.6, build
+`20261015.6141207`. With a temporary nested desktop open, Frame Control sent
+a 47-character test marker to Klipper, then **Copy the headset's clipboard**
+read it through `getClipboardContents` and set a Linux PC's X11 clipboard.
+Reading that PC clipboard returned the exact marker. The prior headset
+clipboard was restored and the test desktop was closed. The PC side used a
+temporary X display on lxso2; macOS and Windows clipboard backends, a user's
+copy action in a headset app, and image/size error cases remain untested on a
+real headset. The fake Frame (`tests/fakeframe`) covers the protocol cases.
@@ -17,13 +17,23 @@ python3 ui/server.py         # anywhere: then open http://127.0.0.1:47810

 ## Features

-The window has four tabs: **Home** (headset view, status, screenshots),
-**Games** (installed games, sideloaded titles, getting games), **Android** (apps,
-the catalogue, display settings, reports) and **Tools** (sending files and text,
-Flatpaks, remote and power). Keys 1–4 switch between them. Files can be dropped
-anywhere in the window. When the Frame can't be reached, one banner says why in
-plain words and the app retries every few seconds, filling everything in once it
-answers. Flatpak and Android installs run in the background; the bottom bar
+The window has six tabs: **Home** (what the headset's doing and what you'd do
+next, the headset view, screenshots, family and comfort, the keyboard and
+trackpad, and VR performance, folded away), **Games** (installed games,
+sideloaded titles, getting games), **Android** (apps, the catalogue, display
+settings, reports), **Tools** (in the headset: your computer's windows, the media
+player and the panel switcher; then sending files and text, Flatpaks, remote and
+power), **Devices** (your headsets and their addresses), and **Control** (the
+searchable Control Center, metrics, supported settings and headset guidance).
+Keys 1–6 switch between them, and `T` opens Type on the headset. The gear in the header (key 7) opens **Settings**:
+privacy and updates, library artwork and the assistant. The battery chip opens
+the headset's details: storage, memory, temperature, Wi-Fi, uptime and SteamOS
+build. Files can be dropped anywhere in the window. A
+connection pill in the header always shows which headset and whether it's
+connected, and how (Tailscale, or the network's name). Click it to add, edit or
+reorder the headset's addresses, or to see each step of connecting. When the Frame can't be
+reached, a banner says why in plain words, what was tried, and counts down to
+the next try, filling everything in once it answers. Flatpak and Android installs run in the background; the bottom bar
 counts them while they run.

 - **Headset view**: what the lenses show, as SteamVR composites it (the room,
@@ -39,11 +49,27 @@ counts them while they run.
  gamescope's flat layer instead.
 - **Screenshots** you take in the headset with Steam's shortcut: browse them and
  save them to `~/Pictures/SteamFrame`.
+- **Record** what the headset shows, with its sound, to an `.mp4` in
+  `~/Videos/SteamFrame`, and **Steam clips**: export clips and background
+  recordings from Steam's game recording at any length, without re-encoding.
+  See [recording.md](recording.md). Not tested on a headset yet.
+- **Backup & recovery** (Tools): copy game saves, settings and Android app data
+  to `~/Documents/Frame Control Backups/<time>/` with a checksummed manifest,
+  restore chosen parts after a factory reset, and **Return to Game Mode** when
+  the headset is stuck in Desktop Mode. Sign-in files and keys are never copied.
+  Not tested on a headset yet. See [backup-and-recovery.md](backup-and-recovery.md).
 - **Battery** with charging state: charge rate in watts, time to full or empty,
  charger type and wattage (for example USB-C PD 20 W), and battery temperature.
 - **Status**: storage, memory, temperature, Wi-Fi, uptime, and whether SteamVR,
  the desktop, Lepton and xrdp are running.
 - **Library** shelf with Steam cover art and a Play button (`steam://rungameid`).
+- **Force quit** a game that has stopped responding: Steam games, sideloaded
+  titles, non-Steam shortcuts and Android apps. **Running now** at the top of
+  Games lists what's running; each game's own **Force quit…** (on its library
+  tile, or in a sideloaded title's ⋯ menu) and the Control Center's **Force quit
+  a game** card do the same. Each asks first, because unsaved progress is lost.
+  Steam is asked to quit the game, then its processes get SIGTERM and SIGKILL;
+  see [steam-games.md](steam-games.md#force-quitting-a-running-game).
 - **Get games**: every game you own with its Steam Frame rating (Verified,
  Playable, Unsupported, Unknown). Install on Frame downloads it to the headset
  with live progress. Search the Steam store with prices and Frame ratings; Buy
@@ -56,9 +82,11 @@ counts them while they run.
  Steam library), then launch, stop, test or remove it. **Report an APK** records
  whether any APK worked (F-Droid or not: pick a file, type a package, or use an
  installed app). Your reports are saved on your computer and change the verdicts
-  you see. They aren't uploaded anywhere: the shared database is maintainer-only
-  for now (see [compat-db/README.md](../compat-db/README.md)). Uses the app's bundled
-  `adb`, or yours if you have one.
+  you see. With **Share compatibility results** on (Privacy & updates), they also
+  go to the shared database ([privacy.md](privacy.md),
+  [compat-db/README.md](../compat-db/README.md)). A failed install records
+  itself when the APK was the problem, and after an install the app offers a
+  20-second test. Uses the app's bundled `adb`, or yours if you have one.
 - **Android display**: pick a running Lepton instance (by the app in it) and set
  its resolution (Native 1920×1080, or Sharp 2560×1440 with density scaled to
  match), UI scale (Smaller / Default / Larger, or an exact dpi) and text size
@@ -68,14 +96,95 @@ counts them while they run.
  their own Android app. A game's `.zip`, folder or `.exe` becomes a title in
  the Steam library (Valve's Devkit Game path, with Proton or the Steam Linux
  Runtime picked from the program's header), listed under **Sideloaded titles**
-  with Launch and Remove; see [sideloading.md](sideloading.md). Send typed text, or your computer's clipboard, to the
-  Frame clipboard.
+  with Launch and Remove; see [sideloading.md](sideloading.md).
+- **Type on the headset** (on Home, in Tools, in the Control Center, or press
+  `T`): use your computer's keyboard and mouse, or a phone's touchscreen, in the
+  Frame's apps through KDE Connect, and move clipboard text both ways. Send
+  typed text or your computer's clipboard to the headset, or copy the headset's
+  clipboard back: copy in a desktop app on the Frame, then **Copy the headset's
+  clipboard**. The clipboard needs the desktop panel open in the headset; it's
+  text only, up to 1 MB. See [file-transfer.md](file-transfer.md#clipboard).
+- **Mac in the headset** (macOS): show any Mac window, or a whole screen, as
+  its own panel in the headset. Place it with the SteamVR dashboard, click and
+  scroll with the laser, and type on the Mac. Streams hardware H.264 through
+  an SSH tunnel; see [mac-in-headset.md](mac-in-headset.md).
 - **Flatpaks**: install and remove them (quick picks: Moonlight, Firefox, VLC,
  Remmina).
+- **Devices**: several headsets, each with several addresses (LAN IPs per
+  network, its `.local` mDNS name, its Tailscale IP or MagicDNS name). The app
+  tries them all at once and learns which worked on which network. Add, edit,
+  reorder and test addresses, find a headset on Tailscale or on this network,
+  name your networks, and switch headsets. See [devices.md](devices.md).
 - **One-click tools**: SSH or SFTP in a terminal window, Steam Link, and remote
  desktop (Windows App on macOS, Remote Desktop on Windows, Remmina or FreeRDP on
-  Linux). Sleep, restart and shut down open a terminal window because SteamOS
-  asks for the sudo password over SSH.
+  Linux). Remote desktop first checks that the Frame's xrdp answers on port
+  3389 (Developer Mode turns it on). On Windows it opens a connection file for
+  user `steamos`, because `mstsc /v:` alone offers your Windows account, which
+  xrdp turns away. Accept the warning about the Frame's own certificate, then
+  sign in with the Developer Mode password. Sleep, restart and
+  shut down open a terminal window because SteamOS asks for the sudo password
+  over SSH.
+
+## Control Center
+
+Open **Control** or `#control` to search controls, status and help. Categories
+cover device/audio, VR/comfort, apps/files, connection/remote and Frame Control
+preferences. Search is local: it matches labels, descriptions and familiar terms
+such as “DPI”, “microphone”, “recenter” and “keep awake”. Queries are not sent to
+Steam, app catalogues, analytics or models. Choosing a workflow link opens its
+existing screen; it does not install, cast, open a helper or change a setting.
+Home, Games, Android, Tools and Devices retain their existing features.
+
+The overview and detail cards use the shared `/api/status` sample: charge and
+power, user/system storage, memory, Wi-Fi, IP, OS/build, uptime, detected services
+and VR timings. Values refresh with the normal 30-second poll. A failed status
+request, lost connection, headset switch or sample received more than 90 seconds
+ago clears current readings and disables audio writes. The client receipt time
+and failed-read state are retained through page startup and UI redraws. Only a
+new successful status read clears a failure; reusing the retained sample does not.
+The headset timestamp is shown without assuming
+its clock matches this computer. Missing sensors show **Unavailable**;
+they are not converted to zero. Compositor FPS is separate from application FPS;
+GPU time is render time, not GPU utilisation. The temperature is the hottest
+readable sensor. Thermal trip warnings use per-zone kernel hot/critical limits,
+not a guessed universal temperature threshold.
+
+**Force quit a game** lists what's running only when you choose **Check what's
+running**, and says so when nothing is. Each **Force quit…** names the game and
+warns that unsaved progress will be lost before anything is sent; it uses the
+selected headset, refuses a second quit while one is running, and drops the
+list when you switch headsets.
+
+**Audio** controls the default output sink through the existing `/api/volume`
+endpoint. Moving its slider prepares a value; **Apply volume** sends it. **Mute**
+and **Unmute** send explicit mute states. The UI captures the selected headset,
+blocks duplicate/pending actions, and requests status after command acknowledgment
+for readback, draining an earlier poll first. Mute preserves an unapplied volume
+draft; switching headsets discards it. It reports
+unconfirmed or differing output separately from confirmed state. API validation
+rejects invalid types, out-of-range levels and invalid mute states before SSH.
+
+Other settings link to the existing forms and their validation, progress,
+confirmations and reset behavior: per-instance Android resolution/density/text,
+comfort sessions and alerts, media layouts, Mac streaming quality, privacy,
+sources and artwork. Android display links reveal the **Library** before
+scrolling to the form. Mac streaming remains searchable on all platforms, with
+its action shown only when the server is on macOS.
+
+Hardware guidance is deliberately separate from controls. Microphone/routing,
+brightness, refresh selection, optics, controllers, Bluetooth, passthrough,
+tracking-origin changes, saved spatial layouts, firmware updates and recovery
+have no verified remote setting here. Use the headset's own controls or the
+documented manual procedures. Movement comfort remains per-game; automatic
+sleep is in Steam Settings → Power → Sleep after inactivity. Keep-awake and
+Android data backup/restore remain script/MCP capabilities, not new UI switches.
+
+**Verification:** the Control Center's JavaScript runs under Node with mocked
+API and DOM boundaries; the server tests mock SSH and reject invalid values
+before any device command. Packaging tests cover the shared CSS/JS in desktop
+and iPhone bundles. These checks establish wiring, search, navigation, offline,
+error and device-switch behavior; they do not establish wearer experience,
+physical settings persistence or a new hardware capability.

 ## How it works

@@ -86,13 +195,18 @@ app bundles `ui/`, `scripts/`, `frame/android/`, Valve's `frame/devkit-utils/` a
 ([python-build-standalone](https://github.com/astral-sh/python-build-standalone))
 and `adb` from Google's platform-tools, so there's nothing else to install. It
 also bundles curl's copy of Mozilla's CA list, because Python on Windows only
-trusts root certificates already in the Windows store.
-`app/build/fetch-deps.js` downloads both, pinned by SHA-256.
+trusts root certificates already in the Windows store. And it bundles KDE
+Connect for the Frame (Valve's arm64 build and five libraries, 3.6 MB,
+[`frame/kdeconnect`](../frame/kdeconnect/NOTICE.md)), which it copies to the
+Frame for the keyboard and trackpad.
+`app/build/fetch-deps.js` downloads all of it, pinned by SHA-256.

 The server is Python stdlib only and listens on 127.0.0.1. It rejects requests
 with a non-local `Host` header, and any `/api/` request without a custom
 header, so other websites can't drive it or read captures. Everything reaches
-the headset through the `frame` SSH alias. On macOS and Linux it keeps one
+the headset through its SSH alias (`frame` for the first one), pointed at the
+address that answered with `-o HostName=` (`ui/frame_link.py`, described in
+[devices.md](devices.md)). On macOS and Linux it keeps one
 multiplexed SSH connection open, so status and each capture take about 0.3 s.
 Windows' OpenSSH can't share a connection, so there each request connects on
 its own and the app is a little slower. What differs between the three
@@ -134,8 +248,8 @@ Google publishes no arm64 Linux platform-tools. Set Up Connection runs
 `ui/frame_connect.py` in your terminal emulator (GNOME Terminal, Konsole, xterm
 and others). The log is at
 `~/.config/Frame Control/logs/server.log`. Running `ui/server.py` in a browser
-instead of the app, sending the clipboard needs `wl-clipboard` (Wayland) or
-`xclip` (X11).
+instead of the app, sending the clipboard, or copying the headset's to this
+computer, needs `wl-clipboard` (Wayland), `xclip` or `xsel` (X11).

 ## Building

@@ -148,5 +262,16 @@ npm run dist:win       # Windows: installer and .zip
 npm run dist:linux     # Linux: AppImage and .deb, x64 and arm64
 ```

-Pushing a `v*` tag builds all three in GitHub Actions and attaches them to the
-release (`.github/workflows/release.yml`).
+Pushing a `v*` tag builds all three in GitHub Actions and attaches them to a
+draft release (`.github/workflows/release.yml`). Running copies are offered it
+once you publish it: see [releasing.md](releasing.md).
+
+## AI agents and assistant
+
+**Documented:** [the MCP adapter and assistant panel](agents.md) are Frame
+Control implementations. MCP wraps this HTTP API without API keys. Changes
+require a separate user approval; power also retains its password prompt. The
+assistant uses a user-chosen endpoint and sends nothing until the user opts in
+for a message. Screenshot context is separately opt-in. Model replies cannot
+operate the headset. Settings → Open assistant settings opens the page; the linked guide
+covers putting it in a Chromium panel on the Frame.
@@ -0,0 +1,121 @@
+# Frame doctor
+
+Run `scripts/frame-doctor.sh` from a checkout on macOS or Linux to diagnose
+first-time setup, pairing and common Frame failures without applying repairs.
+It needs Bash, Python 3 and OpenSSH on the computer, and Python 3 on the Frame.
+Windows users can run `python scripts/frame_doctor.py` with Python and OpenSSH
+installed; the desktop app does not bundle this command.
+
+```sh
+scripts/frame-doctor.sh
+scripts/frame-doctor.sh --alias bedroom --lan frame.local
+scripts/frame-doctor.sh --alias frame --lan 192.168.1.40
+```
+
+Use only addresses belonging to your intended headset. The command tries the
+existing `frame` SSH alias first (Tailscale if setup configured it that way),
+then `frame.local`, then USB-C at `10.86.200.233`. `--alias`, `--lan` and `--usb`
+override those defaults. It reports the successful path without printing its
+address. It neither discovers other devices nor enables Tailscale.
+
+The existing alias supplies the user, port, keys and known-hosts file. LAN and
+USB overrides preserve its `HostKeyAlias`, or its configured hostname if no
+alias was set. A first connection therefore needs trusted setup first. The
+doctor never accepts a new host key, deletes a saved key, generates keys or
+sends a pairing request. It uses your trusted OpenSSH configuration; OpenSSH
+can evaluate `Match exec` conditions in that configuration. Proxy commands,
+jumps, local commands, forwarding and connection sharing are disabled for
+these direct diagnostic connections.
+
+## Reading the result
+
+Each of the eleven runbook sections has one line: `OK`, `WARN`, `BROKEN` or
+`UNKNOWN`. `OK` means only that the stated check passed, not that the whole
+feature works. Optional features that are off may produce a warning.
+
+| Section | Read-only evidence and limits |
+|---|---|
+| 1. Boot and crash history | Previous-boot kernel-oops signatures; one day of coredump history. An oops alone does not prove an unclean reset. |
+| 2. Wi-Fi | NetworkManager's wlan0 state, without SSIDs or connection profiles. |
+| 3. Displays and SteamVR | DSI timeout signatures in the recent kernel journal; no display, compositor or game is launched. |
+| 4. Steam and storage | Free space in home and `devkit-game`, including separate mounts; missing Steam install marker. No CRC scan or restart. |
+| 5. Idle sleep and keep-awake | Existing keep-awake unit state; never takes ownership or changes idle timers. |
+| 6. Remote wake | WoWLAN state, without arming it or testing suspend. Wake with the power button. |
+| 7. Lepton | Recent port-clash, binder and data-clearing signatures; no Android/container commands. |
+| 8. Chromium XR | Crash history; no browser profile or environment inspection. |
+| 9. KDE Connect | Process presence; absence is normal when sharing is off. |
+| 10. Streaming and capture | ffmpeg crash history; no capture or encoder is started. |
+| 11. Panels | Always `UNKNOWN`: placement requires the user's visual confirmation. |
+
+Journal samples are capped at 2,000 entries; current-boot samples cover the
+last ten minutes. This can miss older or high-volume failures. Empty,
+inaccessible or permission-limited journals are `UNKNOWN`, never zero errors.
+Commands time out individually, and each SSH probe is bounded to 90 seconds.
+
+Exit codes: **0** if all reported checks are OK; **1** for warnings, broken or
+unknown checks after connecting; **2** if no authenticated SSH path exists or
+local SSH setup is unavailable. Panel verification currently means a connected
+run normally exits 1. Automation should read the labelled checks rather than
+treating a warning as proof the headset is unusable.
+
+## First-time setup and pairing
+
+- **No SSH response:** wake the headset with its power button. Check the cable
+  or Steam Settings → Internet, then enable Developer Mode. Sleep and network
+  failure can look alike; a timeout cannot distinguish them.
+- **Key rejected:** open **Set Up Connection** in Frame Control. In the headset,
+  open Steam Settings → Developer → **Pair new host**, then approve setup's
+  request. The doctor only checks whether devkit port 32000 responds; that
+  does not prove pairing mode is open or the service is healthy.
+- **Unknown or changed identity:** verify the intended headset through setup
+  or Devices → **Check its key**. Do not delete `known_hosts` to bypass this.
+- **No devkit response:** check Developer Mode and the address. See
+  [SSH setup](ssh.md) for the manual fallback. The doctor cannot turn SSH on.
+
+No password is requested. Enter any password required by the separate setup
+flow only in its local terminal; never paste it into an issue or report.
+
+## Restart safety and recovery
+
+Before deeper checks, and again afterwards, the doctor samples Steam and
+SteamVR `NRestarts` and health-check tracker sizes two seconds apart. Rising
+counters stop diagnostics immediately. Nonempty health-check trackers also stop
+deeper checks even when stable, since a short sample could miss a slower loop.
+Missing counters or counters that reset stop deeper checks as `UNKNOWN`. Stable
+counters with empty trackers still cannot rule out a slower restart loop.
+
+The [historical runbook from PR #33](https://github.com/saphid/frame-control/blob/cce3d940302054ab5246c92d740afaefc4e58d38/docs/frame-doctor.md)
+is the source for these section numbers and observed signatures (September
+2026 builds). Its section 4 describes freezing and verifying health-check
+trackers before prolonged recovery during a live loop. That changes headset
+state, so this read-only implementation stops and directs you there instead.
+It does not implement issue #34's automatic fixes, before/after repair runs,
+or full recovery automation. Review build-specific recovery instructions with
+the person using the headset; a repair can remove data.
+
+Never reload `ath12k`, suspend remotely, kill gamescope to diagnose a loop,
+reset or force power off during extraction/repair, or write passwords to disk.
+The doctor never restarts services, signals processes, repairs Steam, changes
+WoWLAN/standby settings, takes keep-awake ownership, installs anything or writes
+a report. Nothing needs restoring after its checks. Manual/device tests must
+follow the [shared-device lock and battery procedure](testing.md#headset-smoke-test).
+
+## Privacy and verification
+
+Only fixed diagnostic summaries leave the remote probe. Raw journals, process
+arguments, environment variables, SSH errors, account IDs, network addresses,
+SSIDs, keys and tokens are not printed. No telemetry or HTTP request is sent;
+the only pairing-service probe is a TCP connection. SSH produces normal access
+logs. Review summaries before sharing them; feature availability can still be
+personal information. Redirecting stdout saves a report only if you choose it.
+
+Unit tests use mocked boundaries. `scripts/e2e.sh test_doctor` covers asleep,
+disk-full, SSH/devkit-off and missing-key faults against the fake Frame. Its
+stable restart counters are explicitly simulated, not proof about SteamOS.
+On 2026-10-24 the script ran read-only against a real Frame, SteamOS build
+`20261015.6141207`. It verified the configured SSH alias and key login,
+then reported `WARN Restart safety`: nonempty health-check trackers. It
+correctly skipped all eleven deeper sections and exited 1. No raw logs,
+addresses or keys were printed, and no repair or restart was attempted.
+This verifies the safety stop on that device; the deeper checks are not
+claimed as headset-tested.
@@ -38,6 +38,11 @@ Lepton (Android 11, podman container "lepton-dev") ← its own panel, app 305600
 | SteamVR settings live in `~/.config/openvr/config/steamvr.vrsettings`, not under `~/.local/share/Steam/config/`. `dashboard.lastAccessedExternalOverlayKey` names the last panel you used. | Settings tweaks |
 | The Steam client's journal (`journalctl --user`) carries SteamVR system UI lines such as `[Overlays] Created: …` and `vroverlay_uid<appid>`. It's the quickest way to see panels come and go. | Debugging |
 | Present: `rsync`, `flatpak`, `python3`, `git`, `qdbus6`, `xrdp`, `xprop`, `xwininfo`, `xterm`, `konsole`, `dolphin`, `gamescopectl`. Missing: `wl-copy`, `xclip`, `xsel`, `kdeconnect-cli`, `tailscale` (installable in `~`, see below), `krfb`, `wayvnc`. | Script design |
+| **SteamOS updates arrive on their own.** The Frame went from 0.3.0 (build 20260922.6101926) to **0.4.1, build 20260925.6191901**, between 2026-09-27 and 2026-09-28 with no action from us; `~` (keys, user Flatpaks, `~/.local/share`) survived. **Verified 2026-09-28.** | Keep changes in `~` |
+| **Valve's package repository has more than the image.** `pacman -Si` / `pacman -Sp` work as `steamos` without root and list Valve's own builds, such as `kdeconnect` 24.02.2 and `python-evdev` 1.7.0 in `extra`. Unpacking those packages into `~` runs them without touching the read-only root. The repository URLs say not to share them, so never write them down; Valve also publishes each build's source package there (`sources/packages/`), which is how Frame Control got the complete source for the KDE Connect it ships. **Verified 2026-09-28**, SteamOS 0.4.1. | [streaming.md](streaming.md#input-type-and-point-in-the-frame-from-the-mac-or-iphone) |
+| **gamescope has its own input injection.** An EIS socket at `/run/user/1000/gamescope-0-ei` (libei 1.4.1 is on the image) offers "Gamescope Virtual Input": relative and absolute pointer, buttons, scroll, keyboard. It drives the panel that has focus in the headset, on either X display. Focus moves only with the controller's laser (or to a new panel when none has it); `gamescopectl focus_info` prints the focus state to the journal. **Verified 2026-09-29.** | [streaming.md](streaming.md#live-view-and-control-watch-a-panel-and-tap-on-it) |
+| **A panel's own pixels:** `ffmpeg -f x11grab -window_id <window> -i :<display>` captures one window (x11grab of the root is black under gamescope). Panels live on `:0` (Steam's UI, windows tagged by `panel-on-frame.sh`) or `:1` (apps Steam starts). **Verified 2026-09-29.** | Frame Control's Desktop view |
+| **gamescope runs two Xwayland displays.** `:0` holds Steam's VR bar and menus (`valve.steam.gamepadui.*`) and ignores XTest pointer motion; `:1` holds apps such as Chromium and takes it. There's also a libei socket, `/run/user/1000/gamescope-0-ei`. **Verified 2026-09-28**, SteamOS 0.4.1. | Keyboard and trackpad |
 | Flathub is a **system** remote. `--user` installs over SSH work and show up in the desktop menu. | `install-apps.sh` |
 | `/` is 10 GB and read-only. `/home` is 929 GB. | Where to put things |
 | Clipboard: Klipper over the nested D-Bus bus (`qdbus6 org.kde.klipper …`). | `paste-to-frame.sh` |
@@ -47,11 +52,15 @@ Lepton (Android 11, podman container "lepton-dev") ← its own panel, app 305600
 | The Steam client runs with `-cef-enable-debugging`, so its UI answers Chrome DevTools on loopback `127.0.0.1:8080`. The `SharedJSContext` page has `appStore` (owned apps), `downloadsStore` and `SteamClient.*`. `steam steam://install/<appid>` over SSH installs an owned game; when the options dialog shows (state 7), `SteamClient.Installs.ContinueInstall()` accepts it. **Verified 2026-09-25** with Balatro and Broforce. The Frame rating is `steam_hw_compat_category_packed >> 8 & 3`. | [steam-games.md](steam-games.md), `ui/frame_steam.py` |
 | Chromium Flatpak 154 has **no immersive WebXR**: `navigator.xr` exists, but `isSessionSupported("immersive-vr")` returns `false`. Web VR180 players (DL8/DeoVR embeds) still play video inline as a flat, pannable view, and their VR button opens a tab on immersiveweb.dev. Forcing it doesn't help. `--enable-features=OpenXR,WebXR --force-webxr-runtime=openxr`, with `/opt/steamvr` and `XR_RUNTIME_JSON` exposed to the Flatpak, still returns `false`. The aarch64 Linux binary has no OpenXR code at all (no `XR_RUNTIME_JSON`, `xrGetInstanceProcAddr` or loader strings), even though `chrome://flags` lists `#webxr-runtime` → OpenXR. **Why (verified against source 2026-09-25):** M154 is the first release that compiles OpenXR on Linux (`enable_openxr` includes `is_linux`, `checkout_openxr` is true in Flathub's tarball, and Flathub's GN args don't turn it off). But `content/services/isolated_xr_device/xr_runtime_provider.cc` only creates an OpenXR device under `ENABLE_OPENXR && IS_WIN`, on 154, 155 and `main`. Nothing on Linux calls the OpenXR code, so the linker drops it. The missing pieces are two unmerged Gerrit CLs (bug 506004811): [8132979](https://chromium-review.googlesource.com/c/chromium/src/+/8132979) wires the provider on Linux (with `kOpenXR` still off by default, so it needs `--enable-features=OpenXR`), and [8441736](https://chromium-review.googlesource.com/c/chromium/src/+/8441736) runs the XR service in a sandbox that allows SteamVR's sockets. The Frame does have an aarch64 runtime: `~/.config/openxr/1/active_runtime.json` → SteamVR `bin/linuxarm64/vrclient.so`. To watch in 3D, use a native player, or a Chromium built with those two CLs ([webxr-chromium.md](webxr-chromium.md)). That build (156.0.8071.0, arm64) reports `immersive-vr` as supported and starts a session that SteamVR takes as its scene app. With the headset on, the WebXR samples scene and three.js's stereo 360 video demo showed in 3D (verified 2026-09-26, seccomp sandbox off). Started with `--remote-debugging-port=9222`, Chromium answers DevTools on loopback. **Verified 2026-09-25**, BUILD_ID 20260922.6101926. | Web video, [panels.md](panels.md) |
 | **DeoVR (Steam app 837380, Windows/Unity) runs immersively** under Proton ARM64 + FEX: Unity's OpenVR XR plugin finds `OpenVR Headset(Steam Frame)` and the `frame_controller`, the GPU shows as Turnip Adreno 750, and AVPro Video decodes through `MF-MediaEngine-Hardware`. It played 7680×3840 and 8192×4096 H.265 VR180 SBS streams in dome/fisheye mode (`FirstFrameReady`). Unity's own `VideoPlayer` (used for grid thumbnails) fails with `0xc00d36bb`, so thumbnail previews stay blank. The first launch takes about 45 s (`ComputeShaders: InitAsync`). Log: `compatdata/837380/pfx/drive_c/users/steamuser/AppData/LocalLow/Deo VR/Deo VR/Player.log`. **Verified 2026-09-25**, BUILD_ID 20260922.6101926. | [vr-video.md](vr-video.md) |
+| **A second Windows program can join a running Proton game.** Copy `WINEPREFIX`, `WINEDLLPATH`, `WINEDLLOVERRIDES`, `PATH` (Proton's `files/bin-arm64` comes first) and `DISPLAY` from `/proc/<game pid>/environ`, keeping your own `HOME` and the rest, and run `wine prog.exe` from an ordinary SSH shell. It shares the game's wineserver (outside the pressure-vessel container). Don't copy `WINESERVERSOCKET`, which is an inherited fd. A game's windows live on gamescope's X display for that app (`:1` for Gravitas), where the full-screen game keeps the pointer and focus: XTest clicks and restacking don't reach other windows, but `XSendEvent` button events sent to a window do. `xwd` captures WPF/DXVK windows but shows Vulkan game windows as black. **Verified 2026-09-29**, BUILD_ID 20260925.6191901, Proton 11.0-2c ARM64. | [mods.md](mods.md) |
 | **Wolvic (VR browser APK) runs in Lepton against SteamVR's OpenXR**, with limits. The stock Lynx build aborts (`Runtime doesn't support selected swapChain color format`: it wants `GL_RGBA8`), and the stock Quest build fails with `XR_ERROR_API_VERSION_UNSUPPORTED`. Patching `DeviceDelegateOpenXR::GetSwapChainCreateInfo` in the Lynx build's `libnative-lib.so` to `GL_SRGB8_ALPHA8` (0x8C43) and re-signing fixes start-up. The Gecko engine then segfaults in `libxul`. The Chromium-engine build (Lynx v1.3-chromium) browses fine as an immersive app. Its page reports `isSessionSupported("immersive-vr") == true`, and `requestSession` succeeds, running about 36 rAF/s, but the headset shows **black** for WebXR content, or Wolvic's loading spinner that never clears, until the session is ended. Video decodes on the software `OMX.google.h264.decoder`. Tapping the URL bar's selection menu crashes it (no clipboard service). Open URLs with `am start -a VIEW -n com.igalia.wolvic/.VRBrowserActivity -d <url>` over the instance's ADB. DevTools is at `localabstract:content_shell_devtools_remote`. **Verified 2026-09-25**, BUILD_ID 20260922.6101926. | Web VR video, [apks.md](apks.md) |
 | Tailscale runs without root as a userspace `tailscaled` user service (static arm64 build in `~/.local/share/tailscale`, lingering on). In userspace mode, inbound tailnet connections reach the Frame's **loopback**, so every port, including DevTools on 8080, is reachable from the tailnet. **Verified 2026-09-25.** | [tailscale.md](tailscale.md), `scripts/tailscale-on-frame.sh` |
 | **T3 Code desktop runs natively.** The stock release `T3-Code-0.0.42-arm64.AppImage` in `~/Applications/T3CodeDesktop/` starts with no extra setup: glibc 2.39, `libfuse.so.2`, GTK 3, NSS and libsecret are on the image. `panel-on-frame.sh --name t3code-desktop -- '~/Applications/T3CodeDesktop/T3-Code.AppImage'` gives it its own panel (`valve.steam.desktopgame.2000281357`, `--ozone-platform=x11`). Its bundled server listens on `127.0.0.1:3773` and shows up in onboarding as the `frame` computer, with `passwordStore: gnome-libsecret`. The image has no agent CLI and no `node`. Agents run through the LAN CLIProxyAPI (`llm-proxy.lan:8317`, which resolves on the Frame). Claude Code 2.1.283 comes from `claude.ai/install.sh`, and Codex 0.157.1 from the `codex-aarch64-unknown-linux-musl` release tarball, both into `~/.local/bin`. `with-cliproxy` and a mode-600 `~/.config/cliproxyapi/secrets.env` are copied from the Mac. The wrappers `claude-cliproxy` and `codex-cliproxy` (a `-c model_provider=cliproxy`, `wire_api="responses"`, `env_key="CLIPROXY_API_KEY"`) are set as `providers.claudeAgent.binaryPath` and `providers.codex.binaryPath` in `~/.t3/userdata/settings.json`, and T3 picked that up without a restart. Through the wrappers, `claude auth status` reports `loggedIn: true` (`oauth_token`), and both CLIs answered a prompt with `kimi-k3`. `gamescopectl screenshot` captured another layer (the Lepton T3 app) rather than this panel. `DISPLAY=:0 xwd -id <win>` piped to `ffmpeg` captures the window itself (1920×1080). **Verified 2026-09-26**, BUILD_ID 20260922.6101926. | Running T3 Code as a host on the Frame |
 | Power actions need `sudo`, which asks for the Developer Mode password over SSH. | Frame Control's power buttons |
 | **SSH server:** OpenSSH 9.7p1. It offers `publickey,password` (keyboard-interactive is off, PAM on) and also asks `userdbctl ssh-authorized-keys` for keys. OpenSSH ≥ 8.8 rejects SHA-1 `ssh-rsa` signatures by default, so a client whose RSA support is SHA-1 only (the Swift library Citadel, for one) can't log in with the RSA key that devkit pairing installs; use ed25519 (**inferred** from OpenSSH defaults). **Verified 2026-09-27**, BUILD_ID 20260922.6101926. | [iphone.md](iphone.md), `ui/frame_connect.py` |
+| **A Chromium app window on gamescope's `:0` with its own `STEAM_GAME` becomes a panel, even when started over SSH.** `chromium-xr/chrome --ozone-platform=x11 --app=URL --window-size=1280,720` got a 1920×1080 window, and tagging it produced `[Overlays] Created: valve.steam.desktopgame.<id>` in `~/.local/share/Steam/logs/vrwebhelper_systemui.txt`. Chromium XR decoded a 1080p H.264 WebCodecs stream from the Mac at about 60 fps. XTest events sent to `:0` (libXtst through Python ctypes) did not reach the page. The Frame has `libXtst`, `xprop`, `xwininfo`, `curl` and the `C.utf8` locale. **Verified 2026-09-28**, BUILD_ID 20260925.6191901. | [mac-in-headset.md](mac-in-headset.md) |
+| **Streaming video into a panel: what the Frame adds.** Chromium XR decodes H.264 in **software** (1920×1290 at about 9 Mbit/s took 4–6 ms per frame); its GL is ANGLE → zink → Turnip on the Adreno 750, and it logs `GetVSyncParametersIfAvailable() failed`. An idle page's `requestAnimationFrame` runs at about 140 Hz; when the headset is worn or woken, SteamVR picks the 4320×2160 @ 144 Hz mode. **An unworn headset throttles panel apps** whatever they draw: a local canvas page ran at 58 fps for about 6 s, then 28, then about 15 once in standby (vrserver logs `entering standby`, with `power.pauseCompositorOnStandby 1`). `vrcmd --handlewakeup` gave 137–144 fps for about 2 s, then 36. So a panel's frame rate and compositor delay can only be measured while it's worn. `tailscaled` runs with `--tun=userspace-networking` and cost about 8% of a core at 9 Mbit/s (0.5% over the LAN address). Wi-Fi power saving is on (`iw … get power_save`). **Verified 2026-09-28**, BUILD_ID 20260925.6191901. | [mac-in-headset.md](mac-in-headset.md#measuring) |
+| **USB-C networking.** Plugged into a Mac, the Frame is a USB network device: macOS names the port "Steam Frame" (here `en9`, 10.86.200.234/29), and the Frame's `usb0` is 10.86.200.233. Ping is about 0.9 ms, and SSH works with the usual host key (`-o HostName=10.86.200.233 -o HostKeyAlias=<tailscale name>`). Steam's Remote Play discovery also broadcasts over it. **Verified 2026-09-28**, BUILD_ID 20260925.6191901. | Frame Control's Mac stream uses it when present ([mac-in-headset.md](mac-in-headset.md)) |
 | **Tools on the image:** Python 3.12.3, `ffmpeg`, `openssl`, `curl`, `rsync`, `zip`/`unzip`, `flatpak`, `wpctl`, `podman`. **No `adb`.** `steamos` is uid 1000, in `wheel`, and sudoers has `%wheel ALL=(ALL) ALL`, so `sudo -S` takes the Developer Mode password on stdin. **Verified 2026-09-27.** | Running Frame Control's server on the Frame (`FRAME_LOCAL=1`, [iphone.md](iphone.md)) |
 | **Each Lepton instance is a podman container** named `lepton-steamlaunch-<instance id>`, labelled with its ADB port (`podman ps --format '{{.Names}} {{.Labels.adb_port}}'`). `podman exec <container> /system/bin/sh -c '…'` runs Android's shell inside it with no adb at all (used for `pidof` and `logcat` by the app tester). Running `wm size`/`wm density` that way is untested. **Verified 2026-09-27.** | `ui/frame_android.py`, the iPhone app's display settings |
 | **Asleep means off the network.** In standby the Frame stops answering on its LAN address, `frame.local` and Tailscale alike (`Host is down`, `No route to host`, timeouts), and ping fails. It was unreachable for about 2.5 hours until woken. Nothing over SSH can wake it. **Verified 2026-09-27.** | Frame Control's offline banner and retries |
@@ -59,7 +68,7 @@ Lepton (Android 11, podman container "lepton-dev") ← its own panel, app 305600
 | **Battery at full on a charger** can read `Discharging` at about 0 W (for example 99 %, 0.0 W, USB-C PD 18 W). Treat under 0.5 W on a charger as "not charging", not "draining". **Verified 2026-09-27.** | Frame Control's battery card |
 | **The OS image is downloadable.** Valve's recovery images for the Frame are at `https://steamdeck-images.steamos.cloud/recovery/`. The root filesystem inside is btrfs, and it runs as an SSH test target on ARM64 Linux without the headset (`tests/frame-container/frame-image.sh`). **Verified 2026-09-27.** | [recovery-and-images.md](recovery-and-images.md) |
 | **Boot / recovery menu.** Hold Power ~10 s until the LED goes off, then power on while holding the **AUX button on top of the Power button** (not the volume keys) until a text menu appears. Entries: `Current` (SteamOS-A/B + build), `Previous` (the other A/B slot), `Boot from USB`, `Repair Steam Installation`, `Erase User Data` (factory reset), `ADB mode`, `Battery Ship Mode`. It auto-boots `Current` after a ~15 s countdown. **Volume Up/Down (left side) move, AUX (right side) selects.** For a boot loop, Valve says pick `Previous` (keeps user data); then `Repair Steam Installation`; `Erase User Data` wipes `~` (SSH keys, Tailscale, Flatpaks, T3 setup). Last resort is a full re-image, two ways: (1) USB: write `steamframe-oobe-repair-<build>.img.bz2` to an 8 GB+ USB-C stick (Balena Etcher on the Mac), pick `Boot from USB`, then use "Wipe Device & Install SteamOS" / "Repair SteamOS" (keeps games and personal content) from the recovery desktop; (2) cable/EDL: `steamframe-oobe-repair-qdl-<build>.tar.gz`, run `flash.sh` (Linux) or `flash.cmd` (Windows), then with the Frame off for 10 s hold Power + Vol Up + Vol Down for 10 s and plug it in; it reflashes and reboots. Both images: `https://steamdeck-images.steamos.cloud/recovery/` (build 20260922.5153644, 0.3.0, 3.8 GiB each, no published checksums); local copies in `~/Downloads/steam-frame-recovery/`. File names, checksums and what's inside: [recovery-and-images.md](recovery-and-images.md). Source: Valve's [SteamOS Recovery FAQ](https://help.steampowered.com/en/faqs/view/1B71-EDF2-EB6D-2BB3) and [Installation and Repair FAQ](https://help.steampowered.com/en/faqs/view/65B4-2AA3-5F37-4227), plus a menu photo in [EloiStree/HelloSteamFrame#9](https://github.com/EloiStree/HelloSteamFrame/issues/9). **Inferred** (Valve docs, 2026-09-26); not yet tried on our Frame. | Recovering from a boot loop |
-| **Boot loop cause: the SteamVR health check.** `steamvr.service` runs `/usr/share/deckard/steamvr-health-check`, which appends `frog:glasses:` to `$XDG_RUNTIME_DIR/steamvr-short-session-tracker` on every failed or <10 s SteamVR run. At 3 it runs `steam-health-check --repair-now`, which **deletes all of `~/.local/share/Steam` (games, login, Developer Mode) and `~/.steam`**, keeping only `registry.vdf`. At 4 it also tries `steamos-bootconf set-mode reboot-other` (fails as the user: `bootenv: Permission denied`). SteamVR normally fails 1–2 times per boot while it waits for the Steam client (`SteamAPI_InitEx failed … Steam is probably not running`, then `fatal stalled cross-thread pipe`). Once Steam has been wiped, it has to re-download a ~210 MB client on every boot, so SteamVR keeps failing, Steam keeps getting wiped and the Frame reboots, in a loop. Also, the Steam updater can deadlock at `Installing update...` (main process blocked writing to the `-child-update-ui` process, which is stuck in `drm_syncobj_array_wait_timeout`). Killing only the `-child-update-ui` process lets the install finish (`package/*.installed` appears). **Fix without sudo:** over USB-C ADB (`adb -s frame shell` works as `steamos` while the Frame is looping; SSH is refused once Developer Mode is lost), truncate both `/run/user/1000/steam{,vr}-short-session-tracker` files and `chmod 444` them (the health check then logs `Permission denied` and does nothing; this is tmpfs, so it resets on reboot). Unstick the updater if needed, let Steam finish installing, then hold Power 10 s and start the Frame normally. `systemctl reboot` over ADB needs interactive auth. After the fix, sign in to Steam and turn Developer Mode back on. **Verified 2026-09-26**, BUILD_ID 20260922.6101926, slot B (clean boot: 0 SteamVR failures, SSH and Tailscale back). | Diagnosing a boot loop |
+| **Boot loop cause: the SteamVR health check.** `steamvr.service` runs `/usr/share/deckard/steamvr-health-check`, which appends `frog:glasses:` to `$XDG_RUNTIME_DIR/steamvr-short-session-tracker` on every failed or <10 s SteamVR run. At 3 it runs `steam-health-check --repair-now`, which **deletes all of `~/.local/share/Steam` (games, login, Developer Mode) and `~/.steam`**, keeping only `registry.vdf`. At 4 it also tries `steamos-bootconf set-mode reboot-other` (fails as the user: `bootenv: Permission denied`). SteamVR normally fails 1–2 times per boot while it waits for the Steam client (`SteamAPI_InitEx failed … Steam is probably not running`, then `fatal stalled cross-thread pipe`). Once Steam has been wiped, it has to re-download a ~210 MB client on every boot, so SteamVR keeps failing, Steam keeps getting wiped and the Frame reboots, in a loop. Also, the Steam updater can deadlock at `Installing update...` (main process blocked writing to the `-child-update-ui` process, which is stuck in `drm_syncobj_array_wait_timeout`). Killing only the `-child-update-ui` process lets the install finish (`package/*.installed` appears). **Fix without sudo:** over USB-C ADB (`adb -s frame shell` works as `steamos` while the Frame is looping; SSH is refused once Developer Mode is lost), truncate both `/run/user/1000/steam{,vr}-short-session-tracker` files and `chmod 444` them (the health check then logs `Permission denied` and does nothing; this is tmpfs, so it resets on reboot). Unstick the updater if needed, let Steam finish installing, then hold Power 10 s and start the Frame normally. `systemctl reboot` over ADB needs interactive auth. After the fix, sign in to Steam and turn Developer Mode back on. **Verified 2026-09-26**, BUILD_ID 20260922.6101926, slot B (clean boot: 0 SteamVR failures, SSH and Tailscale back). **Seen again 2026-09-28** on BUILD_ID 20260925.6191901, beta client 1790377368. The Frame rebooted by itself while off the network. At 21:13 the check tried `reboot-other` (`bootenv: Permission denied`) and reset the unpacked Steam install, and the tracker had 15 entries. The tracker fix above, applied over SSH, stopped the resets (the checks then log `Permission denied`), but Steam itself kept exiting about 17 s after each start (34 restarts). Cause not found yet. | Diagnosing a boot loop |

 ## Debug recipes

@@ -87,5 +96,6 @@ ssh frame 'cat /opt/steamvr/resources/webinterface/dashboard/localization/dashbo
 - Remote access from anywhere: [tailscale.md](tailscale.md)
 - Floating windows in space: [panels.md](panels.md)
 - Recovery images, what's in them, testing without the headset: [recovery-and-images.md](recovery-and-images.md)
+- Backing up saves and settings, restoring after a reset, leaving Desktop Mode: [backup-and-recovery.md](backup-and-recovery.md)
 - Frame Control on iPhone (the server running on the Frame itself): [iphone.md](iphone.md)
 - What's still unverified: [open-questions.md](open-questions.md)
@@ -12,7 +12,8 @@ An iPhone can't run Python or `ssh`, but the Frame can. So the app:
 1. connects to the Frame over SSH itself (the [Citadel](https://github.com/orlandos-nl/Citadel)
   Swift SSH library), with its own ed25519 key from the Keychain;
 2. copies Frame Control's server and helpers (`ios/scripts/make_frame_bundle.py`,
-   under 1 MB) to `~/.cache/frame-control/<version>` on the Frame, once per version;
+   4.6 MB, 3.6 MB of it the KDE Connect the keyboard and trackpad use) to
+   `~/.cache/frame-control/<version>` on the Frame, once per version;
 3. starts `ui/server.py` there with `FRAME_LOCAL=1`. It listens only on the
   Frame's own 127.0.0.1, and it stops when the phone disconnects (`--exit-on-eof`);
 4. tunnels to it through the SSH session and shows the same page as the desktop
@@ -25,7 +26,10 @@ as its transport too), so the desktop and phone share one code path. Android
 display settings use `podman exec` into each Lepton container instead of adb,
 which the Frame doesn't have.

-Nothing is left running on the Frame after the phone disconnects; the copied
+The app server stops after the phone disconnects. An explicitly started
+[comfort session](family-comfort.md) keeps its timer and headset reminders running
+until the session ends or is cancelled; phone notifications require the app to
+remain connected and running. The copied
 files stay in `~/.cache/frame-control` (delete it any time).

 ## Pairing
@@ -98,8 +102,9 @@ app installed opens its App Store page.
 Things iOS asks the first time: **Local Network** (tap Allow, or the app can't
 see the Frame), and **Paste** when you send the iPhone's clipboard (tap Allow
 Paste, or set Settings → Apps → Frame Control → Paste from Other Apps → Allow).
-Sending text to the Frame's clipboard needs the desktop panel open in the
-headset, as on the desktop app.
+Sending text to the Frame's clipboard, or copying the Frame's to the phone
+(**Type on the headset → Copy the headset's clipboard**), needs the desktop
+panel open in the headset, as on the desktop app.

 Not yet exercised: Android display changes through podman (no Android app was
 running), a real sleep/restart/shut down on the Frame, and a physical iPhone.
@@ -107,3 +112,11 @@ running), a real sleep/restart/shut down on the Frame, and a physical iPhone.
 Debug builds have Simulator test hooks (`FRAME_TEST_HOST`, `FRAME_TEST_PAGE`,
 `FRAME_TEST_JS`, and the tunnel URL in the app's Caches folder); release builds
 don't.
+
+## Family and comfort
+
+The shared Home card sets session limits, breaks and check-ins; **Cast to this
+screen** is in Home's **Right now** card. **Enable / test notifications** requests iOS notification
+permission and sends a local test. These are local notifications, not APNs push;
+iOS background suspension can interrupt phone alerts. The headset timer still
+runs. See [the behavior and verification limits](family-comfort.md).
@@ -0,0 +1,222 @@
+# PC VR streaming from Linux
+
+**Recommendation, 2026-09-28:** test Valve's current SteamVR/Steam Link path
+on a Linux gaming PC before building another streamer. Valve now documents
+Linux streaming fixes and USB support. We have no Linux host attached, so
+Linux-to-Frame VR streaming remains **unverified here**.
+
+This is the feasibility and options report for
+[#24](https://github.com/saphid/frame-control/issues/24), not a shipped streaming
+feature. Frame Control's features must use our own implementation or standard
+platform components. WiVRn and ALVR are research comparisons, not dependencies.
+An optional install shortcut is the most we would offer for a third-party app.
+Our own streamer requires Alex's choice before implementation.
+
+## What was checked on the Frame
+
+**Verified** on 2026-09-28: aarch64, SteamOS **0.4.1**, BUILD_ID
+`20260925.6191901`, SteamVR **2.18.1**. Version and build are recorded separately;
+earlier docs associate this build with other SteamOS version labels.
+
+| Client | Installation | Runtime result |
+|---|---|---|
+| WiVRn **26.9**, upstream `WiVRn-release.apk` | API 29, arm64-v8a; installed in its own immersive Lepton instance | OpenXR instance creation fails: missing `XR_KHR_convert_timespec_time`. Both 1.1.58 and 1.0.58 attempts return `XR_ERROR_EXTENSION_NOT_PRESENT` |
+| ALVR **20.14.1**, upstream `alvr_client_android.apk` | API 26, arm64-v8a; installed in its own immersive Lepton instance | Same missing extension. Client panics at `client_openxr/src/lib.rs:220` with `ERROR_EXTENSION_NOT_PRESENT` |
+
+The [evidence excerpt](evidence/linux-vr/2026-09-28.txt) includes APK SHA-256s,
+upstream release links, loader errors and cleanup results. These are failures
+before an OpenXR session, not successful VR clients. WiVRn was launched twice;
+ALVR's container remained up despite its client panic. Container liveness alone
+does not establish VR compatibility.
+
+Both APKs already declare `MAIN` and `LAUNCHER`. They were installed unmodified
+using this branch's existing `python3 ui/frame_android.py install APK --vr`,
+then launched through their Steam shortcuts. Logs came from the instance's
+`podman exec … /system/bin/logcat`; the user journal also retained WiVRn's errors
+after its container exited. No headset was worn and no host was connected.
+
+All test app files, compatdata, shortcuts and containers were removed afterwards.
+SteamVR's original process remained running. No global settings changed.
+
+### Locked repeat, 2026-09-29 (verified)
+
+Acquired `/tmp/frame-test.lock` before installing or launching anything and
+released it after cleanup. Battery was 44% and charging at preflight, 46–47%
+during the launches, and 47% at cleanup. The SteamOS version/build was unchanged.
+
+Reinstalled and launched both original APKs in immersive Lepton instances.
+WiVRn again failed at OpenXR 1.1 and 1.0 with the missing timespec extension;
+ALVR again panicked on `ERROR_EXTENSION_NOT_PRESENT`. This repeats the
+unmodified-client test, not the newer installer's automatic compatibility-layer
+path. Neither reached a session that could be paired or exercised further.
+Fresh [journal excerpts and cleanup evidence](evidence/linux-vr/2026-09-29.txt)
+record the failures.
+
+SteamVR's screenshot API returned a 1920×1080 headset capture after each
+launch. Both are uniformly dark: [WiVRn](evidence/linux-vr/2026-09-29-wivrn.png)
+and [ALVR](evidence/linux-vr/2026-09-29-alvr.png). These images do **not** prove
+rendering or a working client. The headset was unworn; visibility, controllers,
+frame rate and motion-to-photon latency could not be judged. The explicit
+OpenXR errors, rather than the dark captures, establish the client blocker.
+
+Both test installs, app data, shader caches, shortcuts, containers and temporary
+capture files were removed. The original Steam and SteamVR process IDs were
+unchanged. No reboot, power action or global setting change was used. Native
+clients remain untested, and no Linux gaming host was available for Valve's
+streaming path. The recommendation below is unchanged.
+
+### Relation to the VR APK branch
+
+**Documented from source:** [PR #20](https://github.com/saphid/frame-control/pull/20)
+was read, not edited (branch inspected at
+[`038dcd4`](https://github.com/saphid/frame-control/commit/038dcd48cd75336f6a86c63c7878bfc9c52deec9)).
+Its compatibility layer handles OpenXR version negotiation, some controller
+profiles and refresh-rate requests. It does **not** implement
+`XR_KHR_convert_timespec_time`. Its launcher fix is unnecessary for these APKs.
+This report has **no unmerged code dependency** on that PR, and neither APK was
+tested with its layer injected.
+
+**Documented from upstream source:** WiVRn requests the extension in
+[`application.cpp`](https://github.com/WiVRn/WiVRn/blob/bbc6e4cc36c355fa6180980abd231673dc15115d/client/application.cpp#L1286)
+and uses it to convert `CLOCK_MONOTONIC` into `XrTime` in
+[`instance::now()`](https://github.com/WiVRn/WiVRn/blob/bbc6e4cc36c355fa6180980abd231673dc15115d/client/xr/instance.cpp#L335).
+ALVR also [requests it unconditionally](https://github.com/alvr-org/ALVR/blob/a9f6542fa507a841f40ab4f3fcb531427cd02550/alvr/client_openxr/src/lib.rs#L188).
+Simply deleting the extension request or returning made-up timestamps would
+not prove correct tracking or timing. A real fix needs a valid clock mapping
+and further runtime tests. No such patch was made.
+
+### Native SteamOS aarch64 clients
+
+**Verified:** the Frame has a native OpenXR runtime manifest at
+`~/.config/openxr/1/active_runtime.json`, pointing to SteamVR's
+`bin/linuxarm64/vrclient.so`.
+
+**Documented:** WiVRn's [26.9 README](https://github.com/WiVRn/WiVRn/blob/bbc6e4cc36c355fa6180980abd231673dc15115d/README.md)
+describes its Linux client as debugging-only, without audio or hardware decode.
+ALVR 20.14.1's [non-Android decoder](https://github.com/alvr-org/ALVR/blob/a9f6542fa507a841f40ab4f3fcb531427cd02550/alvr/client_core/src/video_decoder/mod.rs)
+returns no decoded frames. The inspected releases ship Android clients, not a
+ready-to-run native Frame client.
+
+**Inferred:** a native port is possible research, but neither release offers a
+demonstrated native alternative to the blocked APKs. Native builds, native
+extension enumeration, hardware decoding and audio were **not tested**. The
+Android extension failure does not establish that the native runtime lacks it.
+
+## (a) Valve's own path — recommended first
+
+**Documented**, from Valve's release notes rather than launch-window reports:
+
+- [SteamVR 2.17.8 beta](https://steamstore-a.akamaihd.net/news/externalpost/steam_community_announcements/1842212951314598)
+  says “Fix crash using Steam Link on Linux when games submit invalid textures”
+  and “Improve streaming recovery when using Steam Link on Linux.” It also
+  adds initial USB streaming, with Steam Client Beta required to use USB
+  without Wi-Fi.
+- [SteamVR 2.17 release](https://steamstore-a.akamaihd.net/news/externalpost/steam_community_announcements/1843481262693486)
+  repeats Linux streaming fixes and initial USB support. USB is no longer
+  solely a claim about an old beta, but version/channel requirements still
+  need checking on the actual host.
+- [SteamVR 2.18.1 beta](https://steamstore-a.akamaihd.net/news/externalpost/steam_community_announcements/1844751498219787)
+  adds USB-tethered **Quest** support with Steam Link Beta. That entry is not
+  proof of a Frame/Linux combination.
+- The [Steam Link page](https://store.steampowered.com/app/353380/Steam_Link/)
+  lists Linux desktop clients, while its Quest VR requirements still say
+  Windows 10 or newer. Desktop Steam Link support is not equivalent to VR host
+  support, and the Quest requirements are not a Frame support matrix.
+
+**Inferred:** Valve has a Linux VR streaming path worth testing. The old blanket
+claim “Linux cannot stream VR” is no longer justified by the evidence. These
+release notes do not establish which Linux GPU/driver/Frame combinations work.
+USB changes the transport; it does not by itself prove host encoder support.
+
+**Not verified:** Linux host discovery, pairing, wireless or USB streaming,
+stereo rendering, controllers, haptics, audio, latency, or a game. Frame-only
+inspection cannot establish any of these. Flat Remote Play and a desktop shown
+on a panel are not substitutes for this test.
+
+Next test, once a Linux gaming PC is available: record distro, GPU/driver,
+Steam client channel/version and SteamVR version; use the Frame's built-in
+Steam connection flow, first wirelessly and then over a data-capable USB cable.
+Launch a free OpenXR sample or developer-consented VR game. Verify stereo,
+head/controller tracking, haptics and audio while worn; retain both ends' logs
+and measure latency and recovery after a link interruption. Restore any test
+channel changes. Do not change the shared headset's channel just for this report.
+
+If that works, Frame Control can provide our own host checks, setup guidance
+and session controls around Valve's existing platform. First establish which
+controls have a usable interface; no stable automated pairing API has been
+verified. **Estimate (inferred):** 2–5 engineer-days for the hardware feasibility
+pass; another 1–2 weeks for a small integration if those interfaces exist.
+
+## (b) Our own streaming — proposal only
+
+This is a new VR transport and device integration, not a desktop capture feature.
+A plausible first target is **one Linux GPU family, one host, one Frame**, using
+SteamVR on both ends. Our host driver would expose a remote HMD/controllers,
+receive poses and inputs, and obtain stereo textures for hardware encoding.
+Our Frame OpenXR app would decode, submit the correct eye views and render poses
+at predicted display times, and return tracking/input. SteamVR/OpenXR, bundled
+codec/transport libraries and platform GPU APIs fit the ownership rule; a
+WiVRn/ALVR/Monado server dependency would not.
+
+**Inferred design risks:** Linux SteamVR texture-sharing/driver interfaces and
+Frame decode-to-GPU interoperability need a spike before committing to this
+architecture. Sending an already-composited desktop mirror loses the stereo,
+pose and timing information we need. Late reprojection, clock conversion,
+backpressure, controller bindings, audio sync and reconnects are substantial
+work. A runtime shim must not assume `XrTime` equals monotonic nanoseconds.
+
+### Reuse from `mac-in-headset`
+
+**Documented from our code**, read-only at
+[`1b90c64`](https://github.com/saphid/frame-control/commit/1b90c64b73bace54c63a3aae154c5d29a9448d72):
+
+- Reuse the ideas for low-latency encoding without B-frames, dropping work
+  before encoding, bounded queues, keyframe recovery, adaptive bitrate,
+  per-frame timing and authenticated session setup.
+- Its VideoToolbox encoder and ScreenCaptureKit capture are macOS-specific.
+  Linux needs a new GPU encoder path (for example VA-API or NVENC via bundled
+  libraries) and VR texture capture, not a port of window capture.
+- Its WebSocket over SSH is useful for a first controlled transport experiment
+  and control messages. Reliable TCP can stall behind lost packets; a VR media
+  path needs measured deadline behaviour, likely datagrams with loss recovery
+  using an ordinary bundled transport library. Do not invent cryptography.
+- Its Chromium/WebCodecs panel viewer is not a VR client. That branch reports
+  software H.264 decoding and occasional long Wi-Fi stalls on the Frame.
+  Its desktop latency measurements are not motion-to-photon measurements or
+  evidence that a 90/120 Hz stereo stream will work.
+
+**Size/effort estimate (inferred, one experienced full-time engineer, hardware
+available):**
+
+| Phase | Deliverable / stop condition | Effort |
+|---|---|---|
+| Feasibility | Linux driver texture access, Frame hardware decode into OpenXR, pose/clock loop; stop if any cannot meet frame deadlines | 2–4 weeks |
+| First end-to-end prototype | One GPU/codec, stereo sample over a controlled LAN, head/controllers, logs and teardown | 4–8 additional weeks |
+| Usable limited beta | Audio/haptics, pairing, recovery, bitrate/loss handling, installer, worn testing and latency work | 6–12 additional weeks |
+| Wider support | Multiple GPU vendors/distros, USB and Wi-Fi variation, long-session stability | 2–4 additional months |
+
+Planning range: **12–24 engineer-weeks for a limited beta**, roughly
+**10–25k lines of our code plus tests/tooling**, excluding bundled libraries.
+This is a low-confidence scope estimate, not a delivery promise; an unsupported
+driver or decode interface could block it entirely. Foveated streaming,
+eye tracking and parity with Valve are excluded. A Linux gaming PC and repeatable
+worn-headset testing are prerequisites. **Do not build this until Alex chooses.**
+
+## (c) Optional “install WiVRn” shortcut only
+
+Allowed as a clearly optional convenience, never a prerequisite for a Frame
+Control feature. **Documented:** WiVRn's server Flatpak ID is
+`io.github.wivrn.wivrn`; its client/server versions must match, and its Flatpak
+includes xrizer/OpenComposite. Those are properties of an independently
+installed third-party stack, not components of our implementation.
+
+**Recommendation:** defer the shortcut while the current client fails before
+session creation. If offered later, label that compatibility result and let
+the user choose the install; do not present “install” as “streaming works.”
+**Estimate (inferred):** 1–2 engineer-days for an optional host-side shortcut
+with package/version detection and honest status, excluding third-party fixes.
+No shortcut, host install, pairing automation or streaming UI was built here.
+
+Choose **(a)** for the next hardware test. Keep **(b)** as a separately approved
+project if Valve's path fails or lacks a required capability. **(c)** does not
+solve the verified client blocker and should not be the product's foundation.
@@ -0,0 +1,486 @@
+# Mac in the headset
+
+Frame Control can show any Mac window, or a whole Mac screen, as its own panel
+in the Steam Frame. You place each panel anywhere in the room with the SteamVR
+dashboard. The laser clicks and drags, the thumbstick scrolls, and you type on
+the Mac's own keyboard. Find it under **Tools → Mac in the headset** (macOS
+only). On Windows there's an experimental port,
+[PC in the headset](pc-in-headset.md).
+
+The confidence labels are the same as in [ssh.md](ssh.md).
+
+## Why this design
+
+First-party options come first, as the repo's rule asks, with the reason
+each one was or wasn't chosen. The full list for every device is in
+[streaming.md](streaming.md#first-party-options-and-why-they-do-or-dont-fit).
+Checked 2026-09-28.
+
+| Goal | First-party option | Chosen? | Why |
+|---|---|---|---|
+| One Mac screen in the headset | **Apple Screen Sharing** (VNC) → Remmina (Remmina 1.4.43 is already installed on this Frame) | Kept as the fallback (`panel-on-frame.sh mac-screen`) | It's the closest to first-party and needs nothing new. But VNC sends compressed tiles rather than video, so moving content is slow: noticeable lag even on a good 5 GHz link (**verified** 2026-09-27, see [streaming.md](streaming.md)), and the Mac's pointer isn't in the picture without a helper. It shows only whole screens |
+| One Mac screen | **Steam Remote Play**, Mac as host (Valve) | For Mac games only; see [Steam's own streaming](#steams-own-streaming) | The Mac's and the Frame's Steam clients already find each other (**verified**). But Remote Play streams a game (the whole desktop only while the game is out of focus, untested from a Mac), never single windows, and a Mac can't host the Frame's VR streaming |
+| One Mac screen | **AirPlay** (Apple) | No | Apple licenses AirPlay receivers only to TV and speaker makers, and nothing official runs on Linux. UxPlay is an unofficial receiver, and it mirrors a whole screen, not single windows |
+| One Mac screen | **Sidecar / Mac Virtual Display** (Apple) | No | These work only with an iPad or Apple Vision Pro |
+| **Each Mac window as its own panel** | None | – | No first-party way does this: Apple's per-app streaming is only for Vision Pro, and Valve's desktop streaming needs a Windows SteamVR host. So Frame Control does it itself |
+| Mac keyboard and trackpad driving the headset | **Bluetooth HID** | No | macOS can't act as a Bluetooth keyboard or mouse. A real Bluetooth keyboard paired with the Frame still works |
+| Mac keyboard and trackpad | **KDE Connect** (KDE) | No | The Frame has no `kdeconnectd` and it isn't on Flathub. Its Mac app has no keyboard or mouse sharing (**inferred**), and on Wayland it can only reach the desktop panel |
+| Mac keyboard and trackpad | **xrdp** (Valve, Developer Mode) | No | It runs a separate Linux session that you view on the Mac. It isn't the headset's view, and it doesn't carry input the other way |
+
+What that leaves is our own stream: nothing to install on the Mac or the
+Frame, and whole screens or single windows. Here the Mac's own keyboard and
+trackpad need no forwarding, because the windows are still on the Mac. The
+laser is the only input that has to be sent back.
+
+Other routes that were compared:
+
+| Option | One screen | Each window | Speed | Verdict |
+|---|---|---|---|---|
+| Sunshine → Moonlight | ✓ | – | Good | Sunshine's macOS support is still experimental ([discussion #777](https://github.com/orgs/LizardByte/discussions/777)), and it captures whole screens only |
+| Virtual Desktop, Immersed | – | – | – | No Frame client as of September 2026 |
+| **Frame Control's own stream** | ✓ | ✓ | Hardware H.264, sending only changed frames | **Built** |
+
+To type into VR surfaces other than these panels (SteamVR's dashboard,
+games), the Frame supports a uinput keyboard and mouse without sudo
+(verified 2026-09-27: `steamos` is in `input`, and `/dev/uinput` is
+`root:input 660`). That's a separate feature, not part of this one.
+
+## Steam's own streaming
+
+The Frame is built around Steam streaming, so this was checked first
+(2026-09-28). It fits Mac games, not Mac windows.
+
+- **VR streaming from the Mac: no.** The Frame streams VR from a PC running
+  SteamVR ("Steam Link" with foveated streaming). SteamVR dropped macOS in
+  2020, and Valve lists PCs, laptops, Steam Deck and Steam Machine as
+  hosts, never a Mac (**documented**:
+  [UploadVR](https://www.uploadvr.com/steamvr-drops-mac-support/),
+  [Road to VR](https://roadtovr.com/steam-frame-game-certification-specs/)).
+- **Flat Remote Play from the Mac: probably, for Steam games.**
+  - Steam on this Mac has streaming on, and the two Steam clients already
+    see each other. The Frame's `remote_connections.txt` shows it
+    connecting directly to "Alexs-MacBook-Pro-7" at 192.168.1.211:27036,
+    and the Mac's shows the Frame connecting over Wi-Fi and over the USB-C
+    link (**verified** in both clients' logs).
+  - Whether a stream then starts, and how a flat game looks in the headset
+    (reviews describe a theater screen), is **not tested yet**. The Frame's
+    Steam was crash-looping during this session (below).
+  - Mac-hosted Remote Play has a long-standing report of the stream
+    closing as the game loads
+    ([Steam forum](https://steamcommunity.com/groups/homestream/discussions/1/574921459914429988/),
+    **reported**).
+- **The Mac desktop through Steam: untested; single windows: no.** Valve
+  says Remote Play shows the host's desktop when the game loses focus
+  ([Steam Remote Play FAQ](https://help.steampowered.com/en/faqs/view/0689-74B8-92AC-10F2),
+  **documented**), so a whole Mac screen may be reachable by starting a
+  game, then switching away from it. Nobody has tried that from a Mac
+  host. It would still be one screen in one panel: Remote Play has
+  nothing like one panel per Mac window, so Frame Control's own stream
+  stays the way to see separate windows.
+- **Steam has a "stream desktop" call, and it pairs with a Mac
+  (verified 2026-09-28).** In the Remote Play device list, the Frame's
+  Steam UI calls `SteamClient.RemotePlay.StartDesktopStream(<client id>)`
+  for a connected device. Called over CDP with the Mac's client ID, it made
+  the Mac's Steam show "Authorize Device" and ask for a 4-digit code shown
+  on the Frame. Once the code was entered, the Mac logged
+  `k_ERemoteDeviceAuthorizationSuccess`. No stream started in that attempt,
+  and a second attempt, now that the device is authorized, is the next
+  test. If it streams the Mac's desktop, that's a whole-screen option built
+  into Steam: one panel, Valve's encoder and transport. It still wouldn't
+  give each window its own panel.
+- **What Steam's work did give us: the USB-C link.** Plugged into the Mac,
+  the Frame appears as a network port called "Steam Frame". Steam's Remote
+  Play discovery uses it, and so does Frame Control's stream now (see
+  "USB-C, when it's plugged in" below).
+
+Next steps, once Steam on the Frame is healthy:
+
+1. Call `StartDesktopStream` again now that the Frame is authorized, and
+   compare its latency and sharpness with Frame Control's stream of the same
+   screen.
+2. Stream the one Mac game installed here (Fortune Mill) from the Frame's
+   library, over Wi-Fi and over USB-C.
+3. Record whether it starts, how it's shown, and its latency. Steam's
+   streaming overlay shows this; our benchmark can't measure it.
+4. While streaming, switch away from the game on the Mac, and see whether
+   the Mac's desktop appears in the headset, and whether its keyboard and
+   pointer work.
+5. If it works, Frame Control's Games page could offer "Stream from the
+   Mac" for Mac-installed games.
+
+## How it works
+
+```
+Mac                                              Frame
+ScreenCaptureKit (one window or display)
+  → VideoToolbox H.264 (hardware, low-latency,
+    no B-frames)
+  → frame-mac-view, 127.0.0.1 ──ssh -R──→ 127.0.0.1:479xx
+                                             → Chromium app window per stream
+                                               (WebCodecs decode), on gamescope's
+                                               X display, tagged STEAM_GAME
+                                             → its own SteamVR panel
+  ← CGEvent (clicks, drags, wheel, keys) ←──── pointer, wheel and key events
+```
+
+- **The agent** is `mac/bin/frame-mac-view`, built from `mac/frame-mac-view`
+  (Swift, no dependencies; `build.sh`). Frame Control's server starts it on
+  first use and stops it on quit.
+  - It captures with ScreenCaptureKit, which sends frames only when something
+    changes, so idle windows cost nothing.
+  - It encodes in hardware with VideoToolbox's low-latency rate control (plain
+    real-time mode where that's unavailable).
+  - While anyone is watching, it keeps the Mac's display awake. A sleeping
+    display isn't drawn, so there would be nothing to capture.
+- **The link** is an `ssh -R` tunnel on its own connection. It's encrypted and
+  works anywhere `ssh frame` works, Tailscale included, with no firewall
+  changes on the Mac. If the headset sleeps or the network drops, Frame
+  Control reopens the tunnel on the same port, and open viewers reconnect by
+  themselves.
+- **USB-C, when it's plugged in.** Connected to the Mac by cable, the Frame
+  is also a USB network device: macOS lists a network port called "Steam
+  Frame", and the Frame's `usb0` answers in under 1 ms. Frame Control
+  checks for it each time it opens the tunnel and uses it when it's there,
+  with the Frame's usual SSH host key. Otherwise it uses the normal path.
+  `FRAME_MACVIEW_USB=0` turns this off. **Verified** 2026-09-28, in two
+  interleaved pairs of runs:
+
+  | | USB-C | Wi-Fi (Tailscale) |
+  |---|---|---|
+  | test: content p50 / p95 | 6.9–7.3 / 8.4–8.8 ms | 9.8–10.1 / 12.1–12.3 ms |
+  | test: click to drawn p50 | 16.6–16.9 ms | 26.4–27.8 ms |
+  | scroll: content p95 | 23.0–23.5 ms | 31.4–36.9 ms |
+  | scroll: late frames | 3.2–3.3% | 4.7–7.0% |
+
+  (`bench/results/2026-09-28-*-usb1.json`, `-usb2`, `-wifi1`, `-wifi2`.)
+- **Access.**
+  - Frame Control's own key never leaves the Mac.
+  - Each viewer is opened with a **single-use ticket**. It's tied to one
+    window or display and expires after a minute. It's spent as soon as the
+    viewer confirms it has received its reconnect key. Until then, a retry
+    gets the same key, so a connection lost at that moment doesn't strand
+    the viewer. Stop revokes tickets that haven't been used yet.
+  - After that, the viewer holds a reconnect key for that one source, in
+    memory only. **Stop** revokes it.
+  - Remaining risk: a program running as `steamos` on the Frame could read a
+    ticket from Chromium's command line in the first second or so and use it
+    first. That gets it the one source being opened, not the Mac, and the
+    real viewer would then fail to connect. Android apps in Lepton run in
+    their own podman container, so they shouldn't see the Frame's process
+    list (inferred, not checked).
+- **The viewer** is `ui/mac-view.html`, served by the agent. It opens on the
+  Frame as a Chromium app window, preferring Chromium XR (`~/chromium-xr`,
+  built with H.264) over Flathub Chromium.
+  - The page puts `[fcNNNNN]` in its title. The launcher finds the window by
+    that tag and sets `STEAM_GAME` to a stable id per source, which gives it
+    its own panel (see [panels.md](panels.md)). The same Mac window gets the
+    same panel id each time.
+  - It decodes with WebCodecs. If it falls behind, it skips to the next
+    keyframe instead of showing old frames late.
+  - It falls back to JPEG stills (**Compatible** quality) where H.264 isn't
+    available.
+- **Flow control.** The agent never lets frames queue up anywhere on the
+  way. It skips capture frames *before* encoding, so no reference frame goes
+  missing, and it lowers the bitrate, then the frame rate, then the size, to
+  fit the link (see [Adapting to the network](#adapting-to-the-network)).
+- **Input.**
+  - A click on a window's panel brings that Mac window to the front
+    (Accessibility API), then clicks at the same point. Double clicks, right
+    clicks, drags and the wheel work too.
+  - Keys typed into the panel are sent as Mac key codes. Any keys or buttons
+    still held down are released if the viewer loses focus or disconnects, or
+    when the stream stops. Characters the key
+    table doesn't know, such as those from other keyboard layouts, are typed
+    as text.
+  - The Mac's own keyboard and trackpad keep working as normal. Click a panel
+    with the laser, then type on the Mac.
+
+## Permissions (Mac)
+
+- **Screen Recording**, to see windows. Without it, the card asks for it.
+- **Accessibility**, so input from the headset reaches the Mac. Without it the
+  stream still works, and the viewer says clicks won't go through.
+
+Both are granted to Frame Control. After granting, press **Refresh**, which
+restarts the helper so it picks them up. The app is ad-hoc signed, so macOS
+may ask again after an update.
+
+## Quality settings
+
+| Setting | Long side | fps | Codec | Use |
+|---|---|---|---|---|
+| Sharp | 2560 | 60 | H.264, ~0.14 bits/pixel | Text-heavy windows on a strong link |
+| Balanced (default) | 1920 | 60 | H.264, ~0.1 bits/pixel | Most things |
+| Light | 1280 | 30 | H.264 | Weak Wi-Fi or Tailscale off the LAN |
+| Compatible | 1280 | 20 | JPEG | A Frame browser without H.264 |
+
+## Measuring
+
+Every frame carries a sequence number, and the agent records its journey on
+the Mac's clock (`Sources/Stats.swift`):
+
+| Stage | From → to |
+|---|---|
+| capture | the Mac composited it (ScreenCaptureKit's display time) → the agent got it |
+| queue, encode | → encoding started → the encoder finished |
+| network | → the viewer received it |
+| decode, draw | → WebCodecs decoded it → it was drawn on the page's canvas |
+| present | → the page's next animation frame |
+
+- **Clock sync.** The viewer syncs its clock to the Mac's the way NTP does:
+  it pings over the stream's own WebSocket and keeps the sample with the
+  shortest round trip. It then reports, in Mac time, when each frame arrived
+  (right away, so the agent can pace itself) and when it was decoded and
+  drawn (in batches every 250 ms).
+- **Input.** The first frame captured after a click or key carries that
+  event's id. So input latency is the viewer's event → injected on the Mac →
+  the first frame after it → drawn in the headset.
+- **Where to see it.**
+  - `GET /stats` (key required) returns every frame and input record.
+  - `/status` includes a two-second summary, which Frame Control's card
+    shows next to each live stream.
+  - In the headset, add `?stats=1` to the viewer or press
+    Ctrl+Alt+Shift+S for an overlay.
+- **The benchmark.** `scripts/macview-bench.py` runs fixed scenarios on the
+  real Frame, from the Mac, with nobody wearing the headset:
+  - **test** is the moving test pattern.
+  - **scroll** is a Chrome page on its own display scrolling at 240 pt/s,
+    which gives about 9 Mbit/s of real 1920×1290 video.
+  - **type** types into a Chrome text box, first fast and then with pauses.
+
+  It writes `bench/results/<date>-<commit>-<label>.json`, compares two
+  results, and runs interleaved A/B tests between agent settings (`ab`).
+  Wi-Fi changes from minute to minute, so single runs at different times
+  aren't comparable. Throttled links come from a shaping relay on the Mac,
+  which needs no sudo (`--net 50@0,3@8,50@16` means 50 Mbit/s, then 3 from
+  8 s, then 50 from 16 s).
+- **What's graded.** "Content" runs from when the Mac composited a frame
+  (or when ScreenCaptureKit delivered it, if that was earlier) to when it was
+  drawn in the viewer. The Frame's compositor adds its own delay after that.
+  That part is reported, but not graded: an unworn Frame throttles panels to
+  about 36 fps after a few seconds, and to 15 fps in standby, whatever they
+  draw. This was **verified** with a local canvas page that ran on the Frame
+  with no network involved (`bench/pages/present.html`). The compositor's
+  share needs a run with the headset worn.
+
+Targets: content p50 ≤ 25 ms (p95 ≤ 40), click to photon p50 ≤ 50 ms
+(p95 ≤ 70), 60 fps with ≤ 1% late frames, no stall over 100 ms, and adapting
+to a new link rate within 1 s.
+
+Baseline on 2026-09-28 (**verified**, home Wi-Fi, Tailscale, Balanced,
+`bench/results/2026-09-28-a3c6e5c-dirty-baseline-fixed.json`; ms p50/p95):
+
+| Scenario | Content | Input to drawn | fps drawn | Notes |
+|---|---|---|---|---|
+| test (1280×720) | 9.9 | 28.8 / 39.0 | 59 | encode 4.0, network 4.0, decode 1.3 |
+| scroll (1920×1290) | 14.7 | – | 50.4 | encode 6.7, decode 6.4 (software), 9.4 Mbit/s, 16% late |
+| type (1920×1290) | 16.7 | 51.2 / 73.2 | – | most of the input time is the Mac app reacting |
+
+After this work, with the controller on (**verified**, same setup,
+`bench/results/2026-09-28-9e4dcdd-final.json`; ms p50/p95). Content is
+now measured from the earlier of display time and delivery, which adds
+about 5 ms to scroll compared with the baseline's way of measuring:
+
+| Scenario | Content | Input to drawn | fps drawn | Grades |
+|---|---|---|---|---|
+| test | 10.5 / 16.7 | 29.6 / 36.3 | 60 | all within target |
+| scroll | 19.8 / 27.4 | – | 55.9 | fps, late frames (4.8%) and worst gap (222 ms) only "acceptable": Wi-Fi stalls (the Mac captured 57 fps in this run; earlier runs got 46–53 from virtual displays) |
+| type | 15.0 / 21.4 | 43.0 / 60.5 | – | all within target |
+
+Of the targets, click to photon is met without the Frame's compositor (the
+headset has to be worn to measure its share), and so is content latency.
+The frame-rate and no-stall targets aren't yet met while scrolling.
+
+**Worn** (**verified** 2026-09-28 22:00, `vrcmd --stats` activity level 1,
+home Wi-Fi over Tailscale, `bench/results/2026-09-28-39afb23-worn.json`;
+ms p50/p95):
+
+| Scenario | Content | Input to drawn | fps drawn | What happened |
+|---|---|---|---|---|
+| test | 10.8 / 15.3 | 25.7 / 34.4 | 58 | as unworn |
+| scroll | 22.2 / 141 | – | 37.7 | Wi-Fi queued 56–364 ms and held frames for 220–270 ms; the controller went to 2.6 Mbit/s and 45 fps |
+| type | 13.5 / 24.3 | 56.8 / 106 | – | slower replies than unworn (43 / 60) |
+
+The Frame's CPU wasn't the limit: about 27% in total, and the viewer took
+45% of one core. The link to a headset on someone's head is much rougher
+than to one lying still. The adaptation keeps latency bounded there, but
+the frame rate drops. The USB-C cable avoids Wi-Fi entirely. Frames
+actually shown were still about 39 fps for the test pattern while worn, so
+the unworn throttling isn't the whole story. The cause is **unknown**.
+
+What was learned (all **verified**, unless marked):
+
+- The biggest costs are encoding (4–7 ms), network (4–6 ms), and decoding
+  on the Frame. Chromium XR on the Frame decodes H.264 in **software**.
+- Wi-Fi alone stalls for 240–580 ms now and then, over Tailscale and over
+  the LAN alike. Over the LAN (`--host 192.168.1.237`) latency was no
+  better, but the Frame used about 8% less CPU, because tailscaled runs in
+  userspace there.
+- With no controller, a link that slows down queues without limit. In one
+  run, frames arrived 1.9 s late, and at worst 9.4 s late.
+- Tried, and no help, so not kept: VideoToolbox options (require hardware,
+  no frame delay, prioritise speed, a hard data-rate cap), Chromium flags
+  (`--disable-gpu-vsync`, `--disable-frame-rate-limit`,
+  `--use-angle=vulkan`), and a 120 Hz virtual display.
+- Inconclusive, so off by default: keeping the Frame's Wi-Fi awake during
+  typing (`FRAME_MAC_VIEW_WARM=40`, a tiny message every 40 ms for 5 s
+  after input). Over three interleaved runs each, input p50 went 44 → 47 ms
+  and p95 92 → 71 ms, and the ranges overlapped widely
+  (`…-ab-keepwarm.json`). In that run, and in the baseline's typing, the
+  harness typed spaces as "+" (a URL-encoding bug, since fixed), so they
+  went through the text path rather than as space keys.
+- Pointer moves now go out on an 8 ms timer, not on the page's next
+  animation frame, which an unworn Frame slows to 15–36 Hz. This is
+  **inferred** to help dragging; the benchmark has no drag scenario yet.
+- Kept: the encoder's timestamps never jump more than two frame intervals.
+  Before this, the first frame after a pause got a quarter of a second's
+  bit budget, and one P-frame reached 204 KB.
+
+## Adapting to the network
+
+`Sources/Controller.swift`, per stream, latency first:
+
+- **The gate.** The viewer acknowledges every frame as it arrives. A new
+  frame is sent only while the oldest unacknowledged one is younger than
+  the path's usual round trip, plus one frame interval, plus room for this
+  link's normal jitter (1.5 times its recent spread, 25–80 ms). So frames
+  never queue in SSH, TCP or the Wi-Fi driver. While the link is stuck, the
+  newest picture waits and goes out as soon as it moves.
+- **The bitrate.** The link counts as congested when, for two checks in a
+  row (100 ms apart), round trips grow by more than 40 ms while the stream
+  uses much of its budget, or the gate holds frames back, or a frame is
+  stuck for 100 ms. Then the bitrate drops to a bit under what actually got
+  through: at least a fifth off, and at most half. Once the link has been
+  clear for a second, it rises by 10% steps, never above the quality
+  setting's bitrate.
+- **The tier.** When the bitrate stays low, and the stream is really
+  limited by the link rather than having little to send, it steps down:
+  60 → 45 → 30 fps, then 75%, then 50% of the pixels. It goes straight to
+  the tier the bitrate supports after half a second, and steps back up one
+  tier at a time after two seconds with room to spare.
+- `FRAME_MAC_VIEW_ADAPT=0` turns it off, for comparison.
+
+Measured on the real Frame, 2026-09-28 (**verified**; interleaved A/B, off
+versus on, medians of the runs, ms):
+
+| Link | Scenario | Content p95, off → on | fps drawn, off → on | Result file |
+|---|---|---|---|---|
+| Clean Wi-Fi (3 runs each) | scroll | 32.3 → 33.6 | 57 → 56.2 | `…-ab-adapt-clean2.json` |
+| Clean Wi-Fi | test | 15 → 14.2 | 60 → 59.7 | same |
+| 50 → 3 → 50 Mbit/s at 8 s and 16 s (2 runs each) | scroll | **4670 → 72** | 45 → 42 | `…-ab-adapt-step3.json` |
+| 50 → 3 → 50 Mbit/s | test | 66 → 16 | 60 → 60 | same |
+
+- **Clean link.** On a clean link it costs nothing measurable. An earlier
+  version with a fixed gate slack lost 9 fps to Wi-Fi jitter while
+  scrolling (47 → 38 fps), and that's why the slack now follows the link's
+  jitter.
+- **Throttled link.** Without the controller, frames queued for up to 5.6 s
+  and never caught up while the link was slow (p95 1.8–5.6 s, second by
+  second). With it, in the two runs:
+
+  | | Run 1 | Run 2 |
+  |---|---|---|
+  | Worst second's p95 just after the drop | 219 ms | 428 ms |
+  | p95 back under 100 ms for 3 s in a row | after 1 s | after 4 s |
+  | Stepped down to 1440 px at 30 fps | 1.8 s after the drop | 3.5 s after |
+  | p95 per second after that, on the 3 Mbit/s link | 55–94 ms | 60–148 ms |
+
+  Sending one frame takes about 27 ms on that link by itself. After the
+  link recovered, the stream was back at full size and 60 fps in about
+  7.5 s. It steps up one tier every two seconds, on purpose, so it doesn't
+  bounce. The 1 s adaptation target was met in one run of two.
+- **A hiccup on a small stream.** In one clean run, a Wi-Fi hiccup made an
+  earlier version halve the test pattern's bitrate five times and drop it
+  to half size. That cut couldn't help: the stream only sends
+  0.47 Mbit/s. Now the controller estimates what a stream wants (captures
+  per second × average frame size). While a stream wants about half its
+  budget or less, no cut takes it below twice what it wants, so it doesn't change
+  tier.
+
+## Checked so far (2026-09-28)
+
+- **Mac (checked by hand, macOS 26.5.2).**
+  - The agent builds, and lists windows and displays.
+  - In a browser, the test pattern decoded at about 60 fps (H.264) and 30 fps
+    (JPEG, about 22 Mbps).
+  - A click in the viewer arrived at the same point in the source.
+  - `pmset -g assertions` showed the display-awake assertion only while a
+    stream was being watched.
+- **Automated** (`tests/test_macview.py`, on CI's macOS runner): the agent
+  builds (a build failure fails the job).
+  - `/ping` and the page are open; everything else needs the key.
+  - Tickets work once and only for their own source, and Stop revokes
+    reconnect keys.
+  - A WebSocket frame claiming 2^63 bytes closes that socket, and the agent
+    keeps running.
+  - A stream sends an SPS-led H.264 keyframe, and sends another when asked.
+  - Stop ends the stream on the Mac even if the viewer ignores it.
+  - The test doesn't decode video, time it, or check where clicks land.
+- **Linux aarch64 (verified in a stand-in, not on the Frame).** In an Arch
+  Linux ARM container with sshd, Xvfb as `:0` and Chromium 153:
+  - The real `show` path worked in 1–2.3 s: tunnel, ticket, launcher,
+    window found and tagged `STEAM_GAME`.
+  - Frame Control's key didn't appear anywhere in the stand-in's process
+    list.
+  - After the tunnel was killed, it came back on the same port within 4 s,
+    and the viewer reconnected by itself.
+  - Chromium decoded the stream in software with the GPU off.
+  - Stop closed the window, and Chromium exited.
+  - This test found and fixed a bug: in a C locale, `xwininfo` can't print a
+    title with non-ASCII characters, so the launcher reads `_NET_WM_NAME`
+    with `xprop`.
+- **On the Frame (verified 2026-09-28, SteamOS build 20260925.6191901,
+  from the Mac, nobody wearing the headset).** Frame Control's **Show** with
+  the test pattern:
+  - The panel was ready in 1.45 s. SteamVR logged `[Overlays] Created:
+    valve.steam.desktopgame.2001639889` (in `vrwebhelper_systemui.txt`).
+  - The window was tagged `STEAM_GAME`, and gamescope sized it to 1920×1080
+    although 1280×720 was asked for.
+  - Chromium XR decoded it live at about 60 fps: the frame counter advanced
+    62 in 1.04 s.
+  - Over home Wi-Fi, the frames in the viewer window had been drawn on the
+    Mac about 11–17 ms earlier, plus `xwd`'s own time. That's measured
+    against the two clocks, which were 37–39 ms apart (±4 ms, measured over
+    one SSH session). SteamVR's compositor and the display come on top.
+  - Clicks and keys injected with XTest into gamescope's Xwayland didn't
+    reach the page. That's inconclusive, not a failure: XTest on gamescope
+    isn't how real input arrives. The laser should arrive as a left mouse
+    button and the thumbstick as a wheel, because the window has an app id
+    (from gamescope's source; see [panels.md](panels.md)).
+- **Not yet checked:**
+  - Clicking, dragging and scrolling with the laser while wearing the
+    headset.
+  - Real window capture and input on a Mac with both permissions granted.
+  - Keys from SteamVR's on-screen keyboard.
+  - Whether Flathub Chromium has H.264. Chromium XR is used when it's
+    installed, as it is on this Frame.
+  - Latency with real, busy windows at Sharp.
+  - A benchmark run while wearing the headset. Only then does the Frame show
+    panels at full rate, so only then can the compositor's share of the
+    latency, and the frame rate you actually see, be measured.
+
+## Limits
+
+- Only windows on the Mac's current desktop (Space) are listed, and
+  minimised windows can't be captured.
+- A window's panel shows only that window. Its menus and sheets are separate
+  windows on the Mac, so open them from the Mac or use **Whole screen**.
+- Keys go to whichever Mac window is in front. Clicking a panel brings its
+  window to the front first.
+- Ctrl stays Ctrl. On the Mac, copy is ⌘C, so use Meta+C on a keyboard paired
+  with the Frame.
+
+## Switching panels and workspace limits
+
+**Tools → Panel switcher** lists open SteamVR panels, including Mac viewers.
+Use **Show** to request focus or **Open in headset** for Frame Control's own
+switcher panel. It uses SteamVR/gamescope and Chromium, with no third-party
+overlay app. [Device checks and limits](panels.md#frame-controls-panel-switcher)
+include the difference between a panel surviving a scene launch and staying
+visible over it.
+
+Saved spatial layouts are blocked on this build: the public OpenVR transform
+setter denies access to gamescope-owned panels. Reconnecting an existing viewer
+is supported; restoring its room position after a reboot is not. We do not
+save short-lived Mac window IDs or viewer access keys as if they were a durable
+workspace. See [the feasibility evidence](panels.md#saved-spatial-layouts-blocked-on-the-current-panel-route).
@@ -0,0 +1,200 @@
+# Real separation of Mac windows in the headset
+
+Research and experiments on 2026-09-28 (macOS 26.5.2, Apple M5 Pro), for
+making each streamed Mac window truly independent. Builds on
+[mac-in-headset.md](mac-in-headset.md). Labels: **verified** (tried here),
+**documented**, **source** (read in someone's code) or **reported**.
+
+## What "separate" lacks today
+
+Today each window is captured on its own
+(`SCContentFilter(desktopIndependentWindow:)`), but it still lives on the
+Mac's one desktop:
+
+- **Clicks.** A click has to raise the window first, so it reorders the
+  Mac's windows, steals focus and moves the real cursor.
+- **Child windows.** A window's menus, popovers, sheets and tooltips are
+  separate windows, so they aren't in its panel. Apple documents that the
+  single-window filter leaves them out.
+- **Size.** A panel's size is tied to the window's size on the Mac screen.
+- **Hidden windows.** Minimised windows, and windows on other Spaces, can't
+  be shown live.
+
+## How the Mac draws, and where pixels can be read
+
+Apps draw with Core Animation into IOSurfaces. WindowServer's compositor
+stacks every window onto each display, including virtual ones, and sends
+the result to the screen. Pixels can be read at three points:
+
+| Where | How | Gets | Doesn't get |
+|---|---|---|---|
+| One window's content | ScreenCaptureKit `desktopIndependentWindow` (public, what we use) | Live, zero-copy, even when covered by other windows (**documented**) | Menus, popovers, sheets. It pauses while the window is minimised (**reported**) |
+| One window plus its children | `SCStreamConfiguration.includeChildWindows` (macOS 14.2+, **reported**) | Menus, popovers and sheets attached to the window | Anything the app puts outside the window's bounds |
+| A whole display | ScreenCaptureKit display filter, with apps or windows included or excluded | Everything on that display, cursor included | Only what's on that display |
+| A snapshot of any window | Private `CGSHWCaptureWindowList`, which AltTab uses for minimised windows and other Spaces (**source**: `alt-tab-macos` `PrivateApis.swift`) | Minimised windows and other Spaces | It's one still image, not a live stream |
+| Another app's layer tree | Private `CALayerHost` with a context id | A live, zero-copy picture | It only works when the other app cooperates, so it's no good for arbitrary windows (**reported**) |
+
+`CGWindowListCreateImage` and `CGDisplayStream` are obsolete in the macOS 15
+SDK (**reported** by MacPorts and JUCE). Nothing reads another app's pixels
+without the Screen Recording permission.
+
+## The idea: each window gets its own virtual display
+
+A virtual display (the private `CGVirtualDisplay`, as used by BetterDisplay,
+DeskPad and quest-display) is a compositor target with no physical screen.
+Put one streamed window alone on its own virtual display, sized to its
+panel, and capture the whole display:
+
+- **Nothing can overlap it.** A plain click at that point always lands on
+  that window, so input doesn't need raising or background-event tricks.
+- **Menus, sheets, popovers, tooltips and context menus appear on the same
+  display, so they're in the panel.** With "Displays have separate Spaces"
+  on (it is on this Mac), each display also has its own menu bar, so the
+  app's menu bar can be part of the panel.
+- **The panel's size is the display's size,** in HiDPI. Resizing the panel
+  means changing the display mode and resizing the window to fill it
+  (Accessibility API).
+- **Minimised windows and other Spaces stop being a problem,** because
+  streamed windows live on their own displays.
+- **The cursor goes where the laser points.** That display's panel is the
+  one being used, much as Vision Pro's Mac Virtual Display works. Keys from
+  the Mac's keyboard go to the window last clicked.
+
+### Verified here (no permission needed)
+
+- **Creating one.** It needs no permission or entitlement. 1920×1080 HiDPI
+  gives a 3840×2160-pixel, 60 Hz display, placed next to the built-in
+  screen, and `NSScreen.screensHaveSeparateSpaces` was true.
+- **Many at once.** 16 were created at once with no error.
+- **Placement.** `CGConfigureDisplayOrigin` far away failed with error
+  1014: macOS keeps displays edge to edge. Disabling one with the private
+  `CGSConfigureDisplayEnabled` from a command-line tool also failed with 1014.
+- **Removal is deferred while the physical screen sleeps.** With the
+  built-in display asleep, virtual displays were not removed when released,
+  or when their process exited, even from their own `.app`. A second display
+  with the same vendor, product and serial couldn't be created. All of them
+  disappeared as soon as the screen woke (`caffeinate -u`). The helper must
+  therefore:
+  - keep a fixed pool of displays and reuse them rather than create new ones;
+  - keep the screen awake while they exist, which it already does while
+    streaming.
+
+### Built: "Give each window its own display"
+
+Frame Control's helper now does this (`mac/frame-mac-view/Sources/Separate.swift`),
+and it's the default in the Tools card. Streams of this kind are named
+`separate:<window id>`.
+
+- For each window shown, it creates a HiDPI virtual display. The display is
+  sized to the window plus a menu bar, with a fresh serial each time, so a
+  display left over while the screen slept can't block a new one.
+- The window is moved onto the display and resized to fill it (Accessibility
+  API), and the whole display is captured.
+- On **Stop** the window goes back where it was, and the display is released.
+- Plain per-window capture is still there: untick "Give each window its own
+  display". Separate mode needs Accessibility (to move the window), and
+  without it, Show says so rather than quietly falling back.
+
+Verified 2026-09-28 (macOS 26.5.2, M5 Pro), using a TextEdit test document
+and the benchmark's Chrome windows:
+
+- **Separation works.** The window moved onto its own display and streamed,
+  with its first frames in 3.8 s. The display showed TextEdit's own menu bar.
+- **Child windows come along.** A context menu and the Page Setup sheet
+  opened on the same display and were in the capture.
+- **Input.** Typing through the stream reached the window. The benchmark's
+  typing scenario does this on every run.
+- **Stop.** Stop put the window back at exactly its old size (656×422), and
+  the display was removed.
+- **The helper must run a real Cocoa event loop.** Earlier, it ran only a
+  `RunLoop`. With that, AppKit never learned about new displays:
+  - `NSScreen.screens` never listed them, so placing the window always waited
+    3 s and then gave up;
+  - the display's HiDPI mode never applied, so windows were captured at 1×
+    (1280×860 instead of 1920×1290 pixels) and text was soft.
+
+  Running `NSApplication` (with no Dock icon) fixed both. The screen now
+  appears within 100 ms, and captures are at 2× (**verified** in
+  `bench/results/*-baseline.json` against `*-baseline-fixed.json`).
+- **Frame rate.** Chrome drew at 60–61 fps on the virtual display, but
+  ScreenCaptureKit delivered only about 46–53 fps from it, whereas the
+  built-in ProMotion screen gave 121 fps (**verified**). Neither a 120 Hz
+  virtual display (`FRAME_MAC_VIEW_VD_HZ=120`) nor a looser
+  `minimumFrameInterval` changed that. Cause unknown.
+- **Windows that keep their own size** (Calculator, 230×408). The window
+  moved and streamed, but stayed small in the display's corner, so the panel
+  was mostly wallpaper. Now only that corner is captured
+  (`SCStreamConfiguration.sourceRect`): the menu bar and the window, at least
+  480×360 points so menus fit. Clicks map to the cropped area. The first
+  frame arrived in 0.6–1.1 s, and on Stop the window went back and the
+  display was removed. A display's menu bar names the active app, so it
+  shows the window's own menus only once the panel has been clicked.
+- **Several windows at once** (on the Mac, with a local stand-in viewer that
+  acknowledges frames). Three and four Chrome windows, each scrolling on its
+  own display at 1920×1290 pixels, streamed at 53–56 fps and about
+  9 Mbit/s each. The helper used 20% (three) or 24% (four) of one CPU core.
+  The hardware encoder is shared: its time per frame went from 6.7 ms for
+  one stream to 7–15 ms for three and 11–25 ms for four, so each extra
+  moving window adds latency to the others. Once in four runs, before a fix,
+  one window left its display when several displays appeared at the same
+  moment, and its stream stopped: nothing on that display was changing any
+  more. The helper now checks every second and puts the window back. Three
+  further runs with four windows were clean, and every display was gone
+  afterwards (checked with `CGGetOnlineDisplayList`).
+- **Quitting.** On SIGTERM, or when Frame Control goes away, the helper ends
+  every stream first, so windows go back before their displays disappear.
+  Verified: a 900×600 window was back at 900×600 after SIGTERM. Closing a
+  viewer 0.05–1.5 s into startup left the window at its old size and no
+  extra display.
+- **A consent prompt.** macOS 26 asks whether to let ScreenCaptureKit apps
+  "bypass the system private window picker". The prompt appeared *on the
+  virtual display*, so it would show up inside the headset panel. Allow it
+  once on the Mac.
+
+### Still to check
+
+1. That a context menu opens over the text being clicked, not just somewhere
+   on the display. This needs a retest after the consent prompt above is
+   allowed.
+2. Stage Manager, which is reported to undo Accessibility resizes.
+3. Where the Dock and the cursor go on a virtual display.
+4. Closing the lid with virtual displays attached (clamshell).
+5. Many moving windows at once in the headset: whether to lower the bitrate
+   or frame rate of panels you aren't using, so the one you are using keeps
+   the encoder to itself.
+6. All of it in the headset, with the laser.
+
+## Input without disturbing the Mac (a finer option)
+
+For windows that stay on the Mac's own screen, input can go to a window
+behind others without raising it:
+
+- **Background click.** `CGEventPostToPid` with the CGEvent fields 91 and
+  92, which say which window a click is for (`kCGMouseEventWindowUnderMousePointer`
+  and `...ThatCanHandleThisEvent`), plus a window-relative location
+  (**reported**, reverse-engineered, needs testing).
+- **Focus without raise.** yabai makes a window key without raising it by
+  posting event records with the private `SLPSPostEventRecordTo` (**source**:
+  `yabai/src/window_manager.c`).
+- **Known failures.**
+  - Chromium and Electron ignore background clicks unless they're built
+    with the 2026 `acceptsFirstMouse` fix (electron/electron#54493).
+  - Games and canvas apps often need real activation.
+  - Password fields (Secure Event Input) drop synthetic keys.
+  - IME composition, as for Chinese or Japanese input, isn't reliable.
+
+With a virtual display per window, most of this isn't needed. It's worth
+having for hovering over one panel while typing in another.
+
+## Encoding and transport
+
+- **Codec.** Keep H.264 4:2:0. Apple's High Performance Screen Sharing uses
+  4:4:4 over two virtual displays (**documented**), but the Frame's Chromium
+  decodes H.264 in software, and no 4:4:4 decode path is confirmed there.
+- **Sharper static text.** When a panel has been still for a moment, send a
+  high-quality refresh, either a keyframe at low quantisation or a lossless
+  WebP overlay, so static text is sharp. Moving content stays as video.
+- **Transport.** Keep WebSocket over SSH for now, as it works everywhere.
+  WebRTC (UDP, congestion control) is the proven step up for Wi-Fi.
+  WebTransport over QUIC is promising, but its server side on macOS is
+  unverified.
@@ -0,0 +1,371 @@
+# My Meta library (Experimental)
+
+Install Quest games your Meta account owns straight from Meta, test whether
+they run on the Steam Frame, and add the result to the shared compatibility
+reports. It is off until you connect, and it lives in **Android → My Meta
+library**.
+
+**Read this first.** My Meta library uses Meta's undocumented internal API
+with your own Meta access token, as the open-source tools QAVS, DownQuest and
+BSManager do. It may stop working whenever Meta changes things, and using it
+may be against Meta's terms. Your account is your responsibility. Frame
+Control is unofficial and not made or endorsed by Meta or Valve.
+
+What it will and won't do:
+
+- It lists and downloads only games your account owns (its active
+  entitlements). It never downloads anything else.
+- It installs games as Meta ships them. There is **no licence-check bypass**:
+  no patching, no stub libraries, no OVRPort or FramePort. A game that
+  enforces Meta's licence check may quit or hang on the Frame, because that
+  check needs Meta's services, which the Frame doesn't have.
+- Game files go only to a cache folder on your computer and then to your
+  Frame. They are deleted from the cache after each install, and never
+  uploaded or shared. Reports carry results only (below), never files or your
+  token.
+
+## Your token
+
+Frame Control never reads your browser's cookies. There are two ways to give
+it your token.
+
+**Sign in with Meta** (desktop app only; a plain browser doesn't show it).
+Tick **I understand…**, then choose **Sign in with Meta…**. Meta's own login
+page (`https://secure.oculus.com`, the address BSManager's "Log in with Meta"
+opens) appears in a Frame Control window. Sign in there as usual. When Meta's
+site sets its access token cookie (`oc_ac_at`), Frame Control takes it from
+that window, connects with it exactly as if you had pasted it (below), and
+closes the window. The window:
+
+- has its own storage (the Electron session `persist:meta-signin`), separate
+  from Frame Control's main window and from every browser on your computer;
+  the cookie is read only from there, and only for `oculus.com`;
+- runs Meta's page sandboxed, with no access to Frame Control (no preload, no
+  Node, context isolation on, no developer tools) and every permission request
+  or check (camera, notifications…) refused;
+- only goes to `https://` pages on `oculus.com`, `meta.com`, `facebook.com` and
+  `instagram.com` (and their subdomains; Meta's login passes through Instagram).
+  A link anywhere else opens in your system browser; a redirect anywhere else
+  is stopped, and closing the window then says where it went. Pop-ups on those sites open in the same
+  window. Frames inside the page may load any `https://` page (Meta's login
+  can embed captcha and CDN frames from other sites, such as Arkose Labs,
+  hCaptcha, reCAPTCHA or fbcdn) or an empty frame (`about:blank`,
+  `about:srcdoc`); a frame going to any other scheme (`http:`, `data:`,
+  `blob:`, `javascript:`, `file:`…) is just stopped. Redirects follow the same
+  rule for the frame they happen in. Frames can't reach the
+  token: the cookie is read by Frame Control's main process, and the window
+  has no IPC;
+- sends a cookie value on only if it looks like a user token (`OC` or `FRL`,
+  mixed case, no `%`, `|` or `:`), and each value once. If the server refuses
+  it, the window stays open; close it to see why;
+- gives up after 10 minutes, as BSManager does;
+- when it is closed, times out or is signed out of, stops reading cookies
+  first, then is destroyed (a page's "leave this page?" can't keep it open),
+  and, if it had sent a token, tells the server to drop that sign-in: each
+  sign-in has a random attempt id. The cancel is sent only once the connect
+  under way has finished (saved, refused or failed), so the server always
+  gets it after the connect; with nothing sent, there is nothing to cancel.
+  The server lists an attempt only while its
+  connect is being checked: cancelled then, it isn't saved; cancelled just
+  after its token was saved (even before the connect has finished tidying
+  up), that token is deleted (as a disconnect). A cancel
+  never removes an account a later sign-in or a pasted token saved, and one
+  for an unknown attempt changes nothing.
+
+The page never sees the token: the desktop app hands it straight to the local
+server's connect path and tells the page only the answer (connected, and your
+Meta display name). It isn't logged anywhere.
+
+The window keeps you signed in to Meta (in `persist:meta-signin` only) until
+you **Disconnect**, which also signs out of Meta in Frame Control: it clears
+that window's cookies, storage and cache, so the next sign-in starts fresh.
+**Sign out of Meta in Frame Control** next to the sign-in button does just
+that part, for a sign-in that didn't connect.
+
+**Paste a token.** You copy your own token and paste it in. The two ways
+BSManager's guide describes
+([How to obtain your Oculus Token](https://github.com/Zagrios/bs-manager/wiki/How-to-obtain-your-Oculus-Token)):
+
+1. **From the Meta website.** Sign in at `https://secure.oculus.com`. Open the
+   browser's developer tools, then **Application** (Chrome, Edge) or
+   **Storage** (Firefox) → **Cookies** → `https://secure.oculus.com`. Copy the
+   value of `oc_ac_at`. It starts with `OC`.
+2. **From the Meta Quest PC app** (Windows). Press **Ctrl+Shift+I**, open
+   **Network**, press **Ctrl+R**, type `graph` in the filter, open the first
+   request, and copy the access token from its **Payload**. It starts with
+   `FRL`.
+
+Treat it like a password: anyone with it can act as your Meta account. Signing
+out of the Meta website or app usually ends it; then connect again with a new
+one.
+
+Frame Control refuses values that can't be a user token before sending them
+anywhere: anything not starting with `OC` or `FRL`, URL-encoded copies (with
+`%`), app tokens (with `|` or `:`), cut-off values (under 20 characters), and
+anything with characters other than letters, digits and `+ / = - _` (the
+base64 and URL-safe base64 sets), or with no lower-case letter (real tokens are
+mixed case). The public tools only describe tokens made
+of letters and digits; if a real token needs another character, it has to be
+added in one place (`TOKEN_CHARS` in `ui/frame_meta.py`), which also makes
+the redaction look for it.
+
+## Using it
+
+1. Tick **I understand…**, then either choose **Sign in with Meta…** (desktop
+   app) or paste your token and choose **Connect**. Frame Control asks Meta
+   "who am I" with it (a read-only query); only a token Meta accepts is kept.
+2. Your Quest games are listed with their version, size, game data, and the
+   community verdict from the compatibility database when there is one.
+3. **Install** downloads the game and installs it on the Frame, as [Import
+   from my Quest](quest-import.md) does. A game with OBB game data then needs
+   a second **Install** while you have it open on the Frame: Frame Control
+   downloads the data, copies it in and closes the game.
+4. **Install and test** does the same, then starts the game itself (twice if
+   it has game data to add: once to add it, once for the test), watches it for
+   about 30 seconds, and records what happened. **Test again** repeats only
+   the test.
+5. Choose **Install** on as many games as you like: they wait in the
+   **Install queue** and install one after another, with Quest imports and
+   Android store installs. **Remove** takes one out before it starts.
+6. **Disconnect** deletes the token from your computer and, in the desktop
+   app, signs out of Meta in Frame Control (clears the sign-in window's
+   storage). Games already on the Frame stay.
+
+From a terminal (the token is read from standard input, never an argument):
+
+```sh
+python3 ui/frame_meta.py connect --token-stdin      # paste the token, press Enter
+python3 ui/frame_meta.py status
+python3 ui/frame_meta.py list [--refresh]           # owned Quest games, with community verdicts
+python3 ui/frame_meta.py download APP_ID            # into the local install cache (a later download clears it after 6 hours)
+python3 ui/frame_meta.py install APP_ID [--test] [--allow-downgrade]
+python3 ui/frame_meta.py test com.example.game [--wait 30] [--no-report]
+python3 ui/frame_meta.py disconnect
+```
+
+The local API: `GET /api/meta` (connected or not, never the token),
+`GET /api/meta?library=1[&refresh=1]` (the games), and `POST /api/meta` with
+`{"action": "connect", "token": "...", "acknowledged": true}`,
+`{"action": "disconnect"}`, `{"action": "cancel-signin", "attempt": "..."}`, `{"action": "install", "app": "APP_ID", "test": false,
+"allowDowngrade": false}` or `{"action": "test", "package": "..."}`. Install and
+test return a background job (`/api/job`); an install waits its turn in the
+install queue (`/api/queue`, see [Import from my Quest](quest-import.md#using-it)). Sign in with Meta uses the same
+`connect` action: the desktop app's main process (`app/meta-signin.js`) posts
+it with the cookie's value and an `"attempt"` id (8 to 64 letters, digits or
+`-`) once the page's acknowledgement box is ticked. A connect whose attempt
+was cancelled while it was checked answers 409 and saves nothing;
+`cancel-signin` for an attempt whose token is the saved account deletes it.
+
+## What an install does
+
+1. Checks the game is in your library (by its Meta app id), and that there's
+   room on this computer for the APK, a patched copy of it and the game data.
+2. Downloads the APK from Meta's download server, within the size Meta
+   declared (and at most 2 GB), checking its SHA-256 if Meta gives one, and
+   that it is an APK of the package and version Meta said.
+3. Runs Import from my Quest's checks. Refused: 32-bit-only builds (no
+   `arm64-v8a`), and apps that need an Android newer than Lepton's (API 30).
+   Installed with a warning: games that use Meta's licence check
+   (`libovrplatformloader.so`), and VrApi-only games ("may not run").
+4. Installs it with `frame_android.install` (VR detection, the launcher fix,
+   the OpenXR compatibility layer, artwork, a Steam library entry). An older
+   copy never silently replaces a newer one on the Frame unless you tick
+   **Allow replacing a newer version**.
+5. Game data: the app's OBB (saved as Meta names it when that is a standard
+   `main|patch.<version>.<package>.obb` name, otherwise as
+   `main.<version code>.<package>.obb`) and any *required* asset files with
+   standard OBB names, at most 8 GB each. One install downloads at most 16 GB
+   in all, APK included, counted as the bytes arrive (so files whose size Meta
+   doesn't give count too); a file that goes past it is stopped and removed.
+   The data is copied into the
+   running game by the same two-step helper Import from my Quest uses. Other
+   asset files (DLC packs, `.pak` files a game fetches through Meta's
+   Platform SDK) aren't downloaded; the result lists them.
+6. Deletes everything it downloaded, whether the install worked or not.
+   Downloads left by a crash are removed after six hours.
+
+## The test, and the report
+
+The test starts the game, waits for its Lepton instance, then looks every 10
+seconds for 30 seconds: is the game's process alive, and what does its Android
+log (main, system and crash buffers) say. It recognises:
+
+| In the log | Cause given |
+|---|---|
+| "entitlement check failed", "not entitled", `ovr_Entitlement…` or `IsUserEntitledToApplication` reporting a failure (a call that passes doesn't count) | Meta's licence check failed |
+| Platform SDK initialisation errors | Meta's Platform SDK couldn't start |
+| `Failed to launch SystemActivities`, "device not supported" | The game checks it is on a Meta Quest |
+| `vrapi_… was called before…`, VrApi load errors | Legacy VrApi |
+| `XR_ERROR_EXTENSION_NOT_PRESENT` (with the extension's name) | An OpenXR feature the Frame lacks |
+| `XR_ERROR_API_VERSION_UNSUPPORTED`, runtime/initialisation failures | OpenXR version or session problems |
+| `VK_ERROR_DEVICE_LOST`, `zink`, `libgallium_dri.so`, `GL_INVALID_…`, shader failures | Graphics driver errors |
+| `ClipboardManager` | Lepton's missing clipboard service |
+| `NoSuchMethodError`, `UnsatisfiedLinkError`, `dlopen failed` | Android or library features Lepton lacks |
+| `FATAL EXCEPTION`, `Fatal signal`, `SIGSEGV` | A crash |
+| `ANR in …`, "Input dispatching timed out" | Not responding |
+| `XR_SESSION_STATE_SYNCHRONIZED/VISIBLE/FOCUSED`, `xrEndFrame` | The VR scene started |
+
+The outcome is one of **runs** or **runs with issues** (its VR scene
+started), **quits** (closed by itself), **crashes**, **hangs** (Android
+reported it not responding; Frame Control then closes it, and says so if it
+couldn't), **didn't start** (its instance never came up), or **no clear
+result**: it stayed open but its log never showed its VR scene starting. That
+can be a game still loading, or one stuck after a failed check; Frame Control
+can't tell, so it doesn't record it. A game already open is closed first, so
+the test starts clean; if it can't be closed, nothing is tested. If the Frame
+fails while the test reads the game or its log, the test says only that it
+couldn't read the log: a failed command's output can be the game's log, so it
+is never shown. Signatures look at the first 4 KB of each log line and are
+bounded, so a huge or hostile log can't make the test slow. They
+are written for Frame Control from the research in [vr-apks.md](vr-apks.md);
+they are not copied from other tools.
+
+The result goes through the same path as **Test** and **Report**
+(`frame_catalog.add_report`): `via` probe, source `meta-store`, the result
+(`runs`, `crashes` or `instance_failed`, the values the database accepts), and
+notes with the outcome, the cause, the names of the matched signals and the
+version code, all fixed texts. Nothing else: no token, account name, file or
+log text. The page may show a short name from the log next to the cause (an
+OpenXR extension, a library, an exception class or a signal), taken from the
+log line after the token is removed from it, and left out unless it looks like
+a real name (64 characters at most, no long mixed letter-and-digit runs);
+reports never carry it. It is always kept on this
+computer; it reaches the shared database only if **Share compatibility
+results** is on in Privacy settings, as for every report. The panel offers to
+turn that on.
+
+## Your token and privacy
+
+- Kept in `meta-account.json` in Frame Control's per-user data folder
+  (`~/Library/Application Support/Frame Control` on macOS,
+  `%APPDATA%\Frame Control` on Windows, `~/.local/share/frame-control` on
+  Linux), written owner-only (0600) through a temporary file. Never on the
+  Frame.
+- Sent only to `graph.oculus.com` (in the POST body) and
+  `securecdn.oculus.com` (in the download address, which is how that server
+  takes it), over HTTPS. A download that redirects elsewhere (Meta's CDN) is
+  followed only to a plain HTTPS address on port 443 that doesn't carry the
+  token in any form, anywhere (host, path, query or fragment; literal,
+  recased, percent-encoded, or 8+ characters of it), nor an `access_token`
+  parameter. That is checked before any lookup or connection, and a redirect
+  that fails it is refused. API answers that redirect are refused.
+- Every connection goes to an address checked to be public (an allow-list:
+  no private, local, shared, reserved, multicast, NAT64, 6to4 or Teredo
+  addresses), and to exactly that address, with TLS checked for the host
+  name, so a second DNS answer can't swap in another. System proxy settings
+  aren't used. This is the same design as the itch.io purchases branch.
+- Never in logs, error messages, API answers, the page or reports: every
+  message is cleaned first, matching the token (or any start of it of 8+
+  characters) in any letter case, with each character written literally or
+  percent-encoded once or twice (%XX, %25XX) in either case, and any other
+  token-shaped value, before it is shortened.
+- Disconnect deletes the file and the library Frame Control holds in memory,
+  and stops a download under way. A connect still being checked when you
+  disconnect is not saved. In the desktop app, before the token is deleted, it
+  also stops any sign-in under way, destroys the sign-in window, waits for a
+  connect still under way to finish and cancels its attempt (deleting its
+  token if it was just saved), and only then clears the `persist:meta-signin` session
+  (cookies, storage, cache); no other session is touched. Sign in with Meta
+  stays off from the moment you press Disconnect until the token is deleted,
+  and the app refuses a sign-in while it is signing out.
+- With Sign in with Meta, a second copy of your Meta sign-in lives in that
+  session's storage, in Electron's per-user data folder (`Partitions/meta-signin`),
+  until you disconnect or sign out of Meta in Frame Control.
+
+## Where the protocol comes from
+
+Frame Control's code is its own (MIT). These public sources describe the
+protocol; Frame Control takes only facts from them (addresses, query ids,
+field names), not code. QAVS and BSManager are GPL-3.0; OculusGraphQLApiLib,
+DownQuest and Oculus-downgrader state no licence.
+
+| Fact | Source |
+|---|---|
+| GraphQL at `https://graph.oculus.com/graphql`, POST, form fields `access_token`, `doc_id`, `variables`; `OAuthException` / code 190 for a bad token | [OculusGraphQLApiLib `GraphQLClient.cs`](https://github.com/ComputerElite/OculusGraphQLApiLib/blob/866081e2/OculusGraphQLApiLib/GraphQL/GraphQLClient.cs), [DownQuest `DownQuest.js`](https://github.com/ptrpaws/DownQuest/blob/50f5d558/DownQuest.js) |
+| "Who am I": `doc_id` 4149322231793299, `data.viewer.user {id, alias}` | OculusGraphQLApiLib `GraphQLClient.cs` and `Results/User.cs` |
+| Owned apps: `doc_id` 4850747515044496, `data.viewer.user.active_entitlements.nodes[].item` | OculusGraphQLApiLib `GraphQLClient.cs`, `Results/Entitlement.cs`, `Results/Application.cs`; used by [QAVS `GameDownloadManager.cs`](https://github.com/ComputerElite/QuestAppVersionSwitcher/blob/36bebd5a/QuestAppVersionSwitcher/GameDownloadManager.cs) |
+| One binary: `doc_id` 4734929166632773 with `binaryID`; fields `package_name`, `version`, `version_code`, `size`, `obb_binary`, `asset_files` (`file_name`, `size`, `is_required`) | OculusGraphQLApiLib `GraphQLClient.cs`, `Results/OculusBinary.cs`, `Results/AssetFile.cs` |
+| Downloads: `https://securecdn.oculus.com/binaries/download/?id=<id>&access_token=<token>`, the same for the APK, `obb_binary` and asset files; QAVS takes the OBB and required asset files | OculusGraphQLApiLib `Game/GameDownloader.cs`, [QAVS `DownloadManager.cs`](https://github.com/ComputerElite/QuestAppVersionSwitcher/blob/36bebd5a/QuestAppVersionSwitcher/DownloadManager.cs), DownQuest `DownQuest.js` |
+| Token checks (`OC`/`FRL`, no `%`, `|`, `:`, not `OC` + 15 digits) | OculusGraphQLApiLib `GraphQL/TokenTools.cs`, [BSManager `oculus.helpers.ts`](https://github.com/Zagrios/bs-manager/blob/00380c79/src/shared/helpers/oculus.helpers.ts) |
+| How users get a token (`oc_ac_at` cookie, or the PC app's `FRL` token) | [BSManager wiki](https://github.com/Zagrios/bs-manager/wiki/How-to-obtain-your-Oculus-Token) |
+| Sign in with Meta: open `https://secure.oculus.com` in an app window, and once it reaches `https://secure.oculus.com/my/profile` read the `oc_ac_at` cookie from that window's session (checked on `did-navigate`, `did-navigate-in-page`, `did-stop-loading`); 10-minute timeout | [BSManager `bs-oculus-downloader.service.ts`](https://github.com/Zagrios/bs-manager/blob/00380c79/src/main/services/bs-version-download/bs-oculus-downloader.service.ts) (`getUserTokenFromMetaAuth`) |
+
+**Assumed, not seen in those sources:** that the entitlement and binary
+answers carry the fields above in the shapes the libraries' models describe
+(the stored queries are opaque); that the list isn't paginated (none of the
+tools page it); that `ANDROID_6DOF` marks Quest apps; where downloads redirect
+to (the tools let their HTTP client follow redirects; Frame Control follows
+up to five, to any public HTTPS host, and refuses one carrying the token); and
+a `sha256`
+field (none was seen, so in practice only sizes are checked). The log
+signatures for the test are inferred from the compatibility research, not
+from a real Frame.
+
+## Evidence and limits
+
+**Verified locally** with a fake Meta and a fake Frame (`tests/test_meta.py`):
+token checks; storage owner-only through a temporary file; redaction (the
+stored token recased and percent-encoded in either case, starts of it, other
+token-shaped values, and redaction before cutting); HTTPS-only URLs; the
+public-address allow-list, resolving once and connecting only to a checked
+address within one deadline; redirects off Meta's hosts refused when they
+carry the token anywhere, before any lookup; the token never sent to the
+CDN; GraphQL redirects refused; disconnect beating a connect in flight and
+stopping a download; library parsing (Quest apps only, nothing expired or
+inactive, details from the binary query); downloads with an OBB; size, length
+and checksum limits and the whole-install budget with nothing left behind;
+refusing apps not in the
+library; the handoff to Import from my Quest's checks, install and data step;
+the licence-check warning instead of a refusal; test outcomes from log
+fixtures (a passing licence check isn't a failure; no VR scene is no clear
+result, not recorded); closing before a test and saying when a close failed;
+reports carrying only fixed texts, no token, log text or file names. The
+real TLS connection code is the same as the itch.io branch's and is exercised
+here only with a fake connection.
+
+**Sign in with Meta, verified locally** with fake Electron windows, sessions
+and cookies (`app/test/meta-signin.test.js`): the window's options (its own
+partition, sandbox, context isolation, no Node, no preload) and Meta's login
+address, no developer tools; permission requests and checks refused; the
+navigation allow-list (links elsewhere open in the system browser, redirects
+elsewhere are stopped, pop-ups never become windows); frames allowed any
+HTTPS page but no other scheme, the window itself only Meta's sites; a
+page's beforeunload not keeping the window open; the cookie read only from
+`persist:meta-signin` and only for `oculus.com`, even when the default
+session and another partition hold one; signed-out values ignored; a value
+sent once, a refused one keeping the window open; the window closing after a
+connect, with the page getting the answer and never the token; nothing
+logged, and a server error that echoed the token scrubbed; closing the window
+or signing out ending the wait; closing, timing out or signing out during a
+connect cancelling that attempt (with its own id) after cookie intake stops
+and only after that connect settles (saved, refused or failed), and no cancel
+when nothing was sent; sign-out (Disconnect's first step) waiting for a
+connect under way, even from a window closed earlier, before it finishes;
+sign-out destroying the window before clearing only that partition; no
+sign-in starting while a sign-out runs (it is refused: "try again in a
+moment"), a second sign-out joining the first, and a sign-out waiting for
+every window and connect started before it. The
+frame redirects held to the frame rule, the window's to Meta's sites. The
+server side (`tests/test_meta.py`): an attempt cancelled while Meta is
+answering not saved, even after 300 other cancels; a cancel just after the
+save deleting that token; a cancel never removing a later sign-in's or a
+pasted token's account; unknown attempts changing nothing; no attempt left
+listed after a connect, refused or not; attempt ids checked.
+
+**Differences from BSManager:** BSManager uses its app's default session (and
+clears only `.oculus.com` cookies afterwards); Frame Control uses a session of
+its own and clears all of it on disconnect. BSManager waits for
+`/my/profile`; Frame Control also reacts to the cookie changing, and to any
+`oculus.com` page, and lets the server decide whether the value is a
+working token.
+
+**Not verified:** Sign in with Meta against Meta's real login page (no real
+Meta account was used, and nothing was sent to Meta): whether Meta's login
+still goes only through `oculus.com`, `meta.com` and `facebook.com`, whether
+it uses pop-ups (they open in the same window here, which would break a
+pop-up that reports back to its opener), whether `oc_ac_at` is still set on
+`oculus.com` after login, and whether Meta accepts the window (Frame Control
+removes Electron's name from the user agent as a precaution). A real Meta token and Meta's real answers (none was
+available; nothing was sent to Meta), real downloads and their redirects,
+real Quest games installed or tested on a real Frame, and whether the log
+signatures match what real games print in Lepton.
@@ -0,0 +1,171 @@
+# Flat-to-VR mods and Beat Saber songs
+
+**Status: UEVR works from Frame Control for Unreal Engine games; Beat Saber
+songs are blocked.** Installed games have a **VR mod** button. It installs
+UEVR, starts the game with UEVR injected, and removes UEVR again. That whole
+loop was run on a real Frame on 2026-09-29 with Gravitas, from the app's own
+UI ([evidence](evidence/mods-2026-09-29.md)). The headset was unworn, so
+the VR image, head tracking and controls haven't been seen yet. Beat Saber
+custom songs need an owned Beat Saber on the Frame, which there isn't
+([below](#beat-saber-songs-first)).
+[Issue #26](https://github.com/saphid/frame-control/issues/26) stays open for those.
+
+The mod manager must be Frame Control's own implementation. Mods and songs
+are permitted third-party content; BSManager, ModsBeforeFriday, MO2 and other
+managers must not be dependencies. Steam, Proton and SteamVR remain platform
+dependencies. No game purchases, entitlement bypasses, withdrawn builds or
+unofficial mod mirrors are part of this work.
+
+## Per-game support
+
+Checked 2026-09-28 and 2026-09-29 on SteamOS **0.4.1**, BUILD_ID
+**20260925.6191901**, aarch64, with **Proton 11.0-2c ARM64** and SteamVR **2.18.1**. “Verified” describes only
+the observation stated, not a promise that the game is playable. “Documented”
+means an upstream source describes it; “inferred” means it still needs a test.
+
+| Game / build | Mod or content | Evidence and support status | Next check |
+|---|---|---|---|
+| Half-Life 2: VR Mod – Episode One, Steam 2177750, build 25413453 | Official Steam community mod, shared base depot 658920 build 25413418 | **Verified: startup only.** Already installed; launched through Proton ARM64. The stereo headset capture showed its first-time setup, and SteamVR loaded `bindings_frame.json`. Gameplay, controller interaction, fresh installation and removal are unverified. | Complete first-time setup and play a level before offering a tested install shortcut. |
+| Gravitas, Steam 1067310, Windows, Unreal Engine 4 | UEVR 1.05, from Frame Control | **Verified 2026-09-29: installs, injects, SteamVR receives frames, removes cleanly.** Twelve cold **Play in VR** runs (33–35 s each), and the full cycle from the app's UI, all loaded `UEVRBackend.dll` into the game, with UEVR logging `Hooked DirectX 11` and `Requested runtime: openvr_api.dll`. SteamVR's compositor counted frame submits for `steam.app.1067310`. Remove restored the prefix to its pre-install file listing. **Not seen:** the stereo image, head tracking and controls (headset unworn). | Wear the headset: check the image, tracking, UEVR's in-headset menu and controller input. |
+| Beat Saber, Steam 620980, Windows / Proton | Basic custom songs; later SongCore and version-matched mods | **Verified: absent from the 868-game library returned by this Frame.** Store metadata lists Windows, not Linux. **Documented:** the PC game reads basic maps from `Beat Saber_Data/CustomLevels` without a mod manager. Playback on Frame is unverified. | An already-owned, legitimately installed copy is required. Do not buy it as part of this task. |
+| Beat Saber, claimed native ARM64 build | Custom songs / native mods | **Inferred: unverified.** The research mentions this build but supplies no verified official distributable or tested layout. CPU architecture alone does not identify Android versus Linux, the game version or the mod ABI. | Establish official provenance, ownership, binary type and version before touching files. Do not apply Quest patches to an unidentified build. |
+| Beat Saber, Alex's Quest 2 copy | Custom songs / Android mods | **Documented: owner-reported copy on Quest 2.** Not reachable on 2026-09-29 (no device on `adb` from the Mac). No APK, version or installed mods inspected; no Frame playback verified. A Quest copy is a Meta store purchase; copying it to another headset would need its entitlement check to pass there, which this work won't work around. This does not establish ownership of the Steam build. | When the Quest is available, inspect the owned copy's version and supported transfer path, then test Lepton/OpenXR compatibility without bypassing entitlement checks. |
+| Hogwarts Legacy, Steam 990080 | R.E.A.L. | **Verified: listed in this Frame's owned library, not installed.** Official release access and redistribution permission were not established; the referenced author Patreon page returned HTTP 403. No archive downloaded or game tested. | Obtain a current free release from the author and confirm its terms before any test. A news report saying “free” is not a redistribution grant. |
+| Horizon Zero Dawn, Steam 1151640; Horizon Forbidden West, Steam 2420110 | R.E.A.L. | **Verified: both listed as owned, neither installed.** Same source/permission blocker as above; runtime support is unverified. | Check each game's supported version against an accessible official release. |
+| Half-Life 2 VR / other OpenVR games | OpenComposite, per-game replacement | **Documented:** forwards OpenVR calls to OpenXR. **Inferred: Frame compatibility unknown.** Not installed or tested. HL2 VR reached setup with the shipped OpenVR path already. | Test a specific game and replacement DLL only if needed; preserve its original DLL. Never switch the shared headset's runtime globally. |
+| Doom / Quake / Half-Life Team Beef ports | Author's VR ports plus separately owned or free game data | **Inferred: untested.** Android ARM64 support does not establish OpenXR extension or controller compatibility on Lepton. | Choose an official release and legally usable data set, then test that exact port. |
+| Skyrim VR, Steam 611670 | SKSEVR / HIGGS / PLANCK stack | **Verified: Skyrim VR is absent from this library.** Owning flat Skyrim or Special Edition is not the VR game's entitlement. Runtime and mod support are unverified. | An already-owned VR copy and version-matched official mod releases are required. |
+
+The test records ([2026-09-28](evidence/mods-2026-09-28.md),
+[2026-09-29](evidence/mods-2026-09-29.md)) distinguish process startup,
+visible output, injection and failures.
+
+## UEVR from Frame Control
+
+![The VR mod dialog while UEVR runs in Gravitas](img/uevr-dialog.png)
+
+**Get games → an installed game → VR mod.** The dialog says whether the game
+can take UEVR, then offers **Install UEVR**, **▶ Play in VR** and **Remove
+UEVR**. The server route is `POST /api/mods {"action": "status" | "install" |
+"start" | "uninstall", "appid": …}`, and the work happens on the Frame in
+[`ui/frame_mods.py`](../ui/frame_mods.py). It uses only the standard library,
+like `frame_steam.py`.
+
+- **Which games.** Unreal Engine games, recognised by their
+  `…/Binaries/Win64/*-Win64-Shipping.exe` (inferred as the general rule; checked
+  with Gravitas, and HL2 VR is correctly refused). The game needs a Proton
+  prefix, so it must have been played once.
+- **Install.** Downloads praydog's [UEVR 1.05 release](https://github.com/praydog/UEVR/releases/tag/1.05)
+  and python.org's [Windows x64 embeddable Python 3.14.7](https://www.python.org/downloads/release/python-3147/).
+  Each is checked against a pinned SHA-256; Python's comes from python.org's
+  release-file API. Archives are unpacked into a staging folder, rejecting
+  entries that escape it, symlinks and oversized contents, then moved into
+  `<prefix>/drive_c/frame-control/` (about 47 MB). Nothing goes into the
+  game's own folder. A receipt in `~/.local/share/frame-control/mods/` lists
+  what to remove, including UEVR's own settings folders only if they didn't
+  exist before. Install and remove refuse while the game is running, and only
+  one mod action runs at a time.
+- **Play in VR.** Launches the game through Steam if it isn't running and
+  waits for its window. It then runs a short script under that Python, with
+  the game's Proton, prefix and Wine settings read from the game process, so
+  both share one Wine session. The script repeats what UEVR's own injector does
+  when you press Inject with OpenVR chosen: load `UEVRPluginNullifier.dll` and
+  call its `nullify`, load `openvr_api.dll`, set `Frontend_RequestedRuntime` in
+  UEVR's per-game `config.txt`, then load `UEVRBackend.dll`. Each load is a
+  `LoadLibraryW` remote thread, as in the frontend's `Injector.cs`. `start`
+  then checks that `UEVRBackend.dll` is mapped into the game and reads
+  SteamVR's frame count for the app (`vrcmd --stats`).
+- **Remove.** Deletes what the receipt lists, then the download cache once
+  no game uses it. Verified: the prefix's file listing matched its
+  pre-install state afterwards.
+
+Why not UEVR's own injector (verified 2026-09-29):
+
+- It's a .NET 6 WPF app, so it needs Microsoft's .NET runtimes (about 70 MB
+  of downloads) in the prefix.
+- Under gamescope, the full-screen game keeps the pointer and focus. XTest
+  clicks, raising or refocusing the injector window don't change that, so in
+  the headset you can't click it. Sending X `ButtonPress`/`ButtonRelease`
+  events straight to its window does work.
+- `--attach=<process>` didn't inject under Wine, and it didn't restore its
+  saved OpenVR choice.
+- Driven that way, it injected in 6 of 8 cold starts. In a bad session every
+  new injector ignored input or never showed its window, even after
+  relaunching. Its process list reads each process's main window title, a
+  cross-process window message; a window that isn't answering would block its
+  UI thread (inferred from the frontend's source, not proven). Without a
+  `HOME` it also died in .NET at startup.
+- Frame Control's own script injected in 12 of 12 cold starts through the API, in 33–35 s, and again from the UI.
+- UEVR's DirectX 12 probing logs errors before it settles on DirectX 11 for
+  Gravitas, and some of its Unreal engine scans fail on this older UE4 game.
+  Neither stopped the injection.
+
+## Beat Saber: songs first
+
+**Documented:** the [BSMG PC guide](https://bsmg.wiki/pc-modding.html)
+describes extracting each map into its own directory below
+`Beat Saber/Beat Saber_Data/CustomLevels`. Basic custom songs do not require
+SongCore; maps that require mod features need their matching dependencies.
+This is a candidate for our own file manager, not a verified Frame feature.
+
+Alex's Quest 2 copy is a separate Android candidate. Its ownership does not
+make the PC `CustomLevels` layout applicable. Until the actual build is
+inspected, neither a direct song-copy recipe nor APK patching is justified.
+
+Only maps whose music and chart are permitted for distribution may be used
+as test fixtures or bundled content. A public download alone does not establish
+those rights. Start with an original or explicitly licensed basic map.
+
+**Documented:** [ModsBeforeFriday](https://github.com/Lauriethefish/ModsBeforeFriday)
+targets Quest Beat Saber over WebUSB/ADB. It is not a generic native ARM64
+modding protocol. [BSManager](https://github.com/Zagrios/bs-manager/releases/tag/v1.6.0)
+publishes an aarch64 Flatpak, but its architecture says nothing about Beat
+Saber or its plugins running on Frame. Neither app is an installation step
+or dependency for Frame Control.
+
+## Requirements for our manager
+
+For UEVR, items 1–5 are implemented in `frame_mods.py` and covered by
+fake-library tests (`tests/test_mods.py`), apart from checking ownership
+through Steam; item 6 is met for Gravitas. Beat Saber songs would need the
+same, and none of it is built for them:
+
+1. Resolve the selected Steam game's real library, installed build, executable
+   architecture and Proton prefix. Confirm ownership through Steam; a directory
+   or app manifest alone is not proof. Keep downloading, installed and playable
+   as separate states.
+2. Download a pinned mod version from the author's official release. Record
+   the URL, version, license and digest. Verify the published digest when
+   available; an upstream SHA-256 detects corruption but is not a signature.
+   Do not treat “free to download” as permission to redistribute.
+3. Stage and validate archives before writing into the game or prefix. Reject
+   path traversal, links escaping the destination, archive bombs and unexpected
+   executable content in song packs. Check song metadata and its referenced
+   files, not just the `.zip` suffix.
+4. Refuse changes while the game is running. Back up originals and journal
+   every managed file and digest. Apply changes atomically where possible and
+   roll back partial failures. Keep runtime prerequisites scoped to this game.
+5. Uninstall only files still matching our receipt; restore originals without
+   overwriting later user edits. Preserve saves, unrelated mods and songs.
+   Song removal must target one managed map, never the whole CustomLevels tree.
+6. Expose one-click actions beside the game only after real-Frame install,
+   playback and uninstall pass. Test filesystem and download failure handling
+   with fake-Frame fixtures; those cannot prove FEX injection or VR rendering.
+
+## Sources
+
+- [UEVR 1.05 official release](https://github.com/praydog/UEVR/releases/tag/1.05)
+  and [author's usage instructions](https://github.com/praydog/UEVR#getting-started).
+- [Microsoft .NET 6 release metadata](https://builds.dotnet.microsoft.com/dotnet/release-metadata/6.0/releases.json),
+  including the SHA-512 hashes used for the injector tests.
+- [UEVR frontend source](https://github.com/praydog/UEVR-Frontend): `UEVR/MainWindow.xaml.cs`
+  (`Inject_Clicked`) and `UEVR/Injector.cs`, the steps `frame_inject.py` follows.
+- [python.org release-file API](https://www.python.org/api/v2/downloads/release_file/) for the
+  embeddable Python's `sha256_sum`.
+- [OpenComposite's OpenXR branch](https://gitlab.com/znixian/OpenOVR/-/tree/openxr),
+  including per-game installation and the need to preserve original DLLs.
+- [R.E.A.L. author post referenced by the research](https://www.patreon.com/realvr/posts/but-wheres-link-165840151)
+  (HTTP 403 from this environment; contents not verified).
+- [Half-Life 2 VR official site](https://halflife2vr.com/) and
+  [Episode One on Steam](https://store.steampowered.com/app/2177750/).
+- [Beat Saber store metadata](https://store.steampowered.com/api/appdetails?appids=620980).
@@ -102,10 +102,11 @@ Still open: 4, 6, 7, 12–15, 16 (off-LAN and after a reboot), 17–21.
 20. **F-Droid 2.0** (Compose 1.12): does it run? If so, the catalogue can
    install it instead of 1.17.2.

-21. **DeoVR local files:** does DeoVR's file browser show `Videos → VR`
-    (the symlink from `push-vr-video.sh`) or `Z:\home\steamos\Videos\VR`, and do
-    the colour-coded test clips play in 3D (red left eye, cyan right) for both
-    H.264 and H.265? Does the DLNA browser find a server on the Mac?
+21. **Owned media player:** worn-headset comfort, audio quality/lip sync, long
+    movies and 4K/8K decoding remain to check. Native spatial-photo container
+    extraction, large/immersive splats and existing-panel theatre docking need
+    further implementation. Remote eye isolation, short hardware decode and
+    owned-overlay cleanup are verified; see [vr-video.md](vr-video.md).

 ## Verified 2026-09-27

@@ -129,6 +130,28 @@ Still open: 4, 6, 7, 12–15, 16 (off-LAN and after a reboot), 17–21.
 - The Mac EDL flashing script in `~/Downloads/steam-frame-recovery/` hasn't
  been run against a Frame.

+## Mac in the headset (2026-09-28)
+
+The test pattern streams to the Frame as its own panel at about 60 fps
+(verified, build 20260925.6191901; see
+[mac-in-headset.md](mac-in-headset.md#checked-so-far-2026-09-28)). Still to
+check in the headset:
+
+- Laser clicks, drags and thumbstick scrolling in a viewer panel.
+- Real window capture and input once Screen Recording and Accessibility are
+  granted to Frame Control.
+- Keys from SteamVR's on-screen keyboard.
+- Whether Flathub Chromium decodes H.264 (otherwise use Compatible).
+
+## Battery charge limit (2026-10-23)
+
+Can an 80% charge limit be set over SSH or adb? **Not safely yet.** SteamOS
+Manager's `BatteryChargeLimit1` (the interface Steam's own setting uses) has
+no backend in the Frame's device config, and the battery node is a
+`max1720x` fuel gauge with no `charge_control_end_threshold`. Research and
+read-only headset checks are in [charge-limit.md](charge-limit.md). Not tested
+on a headset.
+
 ## Unconfirmed claims made in these docs

 - `/home` and `/etc` persist across Frame OS updates. This is inferred from
@@ -121,5 +121,171 @@ a limit on the number of floating panels.
  script needed.
 - **Inside the desktop panel**: KWin tiling (Meta+arrow keys with a Bluetooth
  keyboard) or virtual desktops arrange windows within the 1280×800 rectangle.
- **Windows-only overlay tools** (Desktop+, OVR Toolkit, OVRdrop) do this for a
-  PC's desktop in SteamVR. They don't run on the Frame's standalone Linux.
+- **Optional overlay tools:** Desktop+, OVR Toolkit and similar software are
+  separate from Frame Control. Public reports describe some Proton support;
+  Windows-only does not by itself prove a Frame app cannot run. Local status
+  and sources are in [VR utilities](vr-utilities.md).
+- **Our performance HUD:** Home → VR performance (unfold it) → Open HUD in
+  headset creates its own gamescope panel using built-in tools. It needs no
+  third-party overlay app. [Metrics and verification](vr-utilities.md).
+
+## Frame Control's panel switcher
+
+**Verified 2026-09-28**, SteamOS 0.4.1, BUILD_ID `20260925.6191901`,
+SteamVR 2.18.1: **Tools → Panel switcher** lists SteamVR's open main panels,
+including panels that are currently hidden. **Show** asks SteamVR to bring one
+forward. **Open in headset** opens the same switcher as its own panel; choose
+it again from Steam's dashboard after switching away. Refresh updates the list.
+This is a list, not thumbnail Exposé.
+
+![Frame Control's switcher rendered on the Frame](img/panel-switcher.png)
+
+This is our own Python/HTML implementation (`ui/frame_panels.py`), using the
+Frame's shipped `vrcmd` OpenVR client and gamescope. The headset page uses
+Chromium (Chromium XR when present, then system Chromium, then the existing
+Chromium Flatpak). No XSOverlay, OVR Toolkit, WayVR or other overlay application
+is needed. This dependency boundary also applies to future layout and panel
+persistence work: platform APIs and bundled libraries are fine; another app
+must not implement the feature for us.
+
+The companion runs the helper over SSH. Opening it in the headset installs a
+copy under `~/.local/share/frame-control/panels/` and starts a loopback HTTP
+server and an isolated Chromium profile. There is no startup service or global
+setting change. Close the switcher to stop its server and browser. Other
+Chromium profiles, Steam and SteamVR are left alone. If the window or runtime
+closes, use **Open in headset** again.
+
+The page carries a random, per-process access key in its URL fragment, removes
+it from the address bar, keeps it in tab session storage for page reloads, and
+sends it in a header. Panel lists and actions need
+that key; Host and Origin checks reject other sites. The key permits only
+listing panels, requesting focus and closing this switcher. Like Mac viewer
+launch tickets, it is initially readable by another process running as the
+same Frame user. Panel titles are rendered as text, never HTML. The companion
+retains its existing request guards. No Mac capture credentials cross this API.
+
+**Verified:** the real headset page rendered its panel list (image above), its
+HTTP focus request changed `GAMESCOPE_FOCUSED_APP` to `2000999030`, a request
+without the key returned HTTP 403, and Close stopped the helper and its browser.
+Opening an already running switcher requests its focus rather than creating a
+second one. The companion uses the same list/focus helper. **Unverified:** laser
+selection while wearing the headset, physical placement, and non-XR Chromium.
+The API reports that focus was *requested*: another action can take focus before
+we observe the result. Closed panels are rejected after re-enumeration.
+
+### Shared-device recheck, 2026-09-29
+
+**Verified:** the follow-up's atomic `mkdir /tmp/frame-test.lock` attempts
+failed because another thread held the lock. The existing lock was left alone;
+no applications were installed, launched or stopped in this follow-up. The last
+read-only battery check showed 62%, charging. The 180 Python and 8 website tests
+passed again locally.
+
+**Unverified in this follow-up:** the prepared browser-button test (Refresh,
+selection, reload and Close) and repeated OpenXR transition could not run under
+the shared lock. The device results elsewhere in this page are the earlier
+2026-09-28 observations, not results from this blocked recheck. In particular,
+HTTP focus is not evidence of worn-headset laser input. Follow the
+[shared-device test procedure](testing.md#headset-smoke-test) for the next run.
+
+## Saved spatial layouts: blocked on the current panel route
+
+**Verified 2026-09-28**, same build, using a temporary xterm panel with
+`STEAM_GAME=2000999031` and `FnTable:IVROverlay_028` from
+`/opt/steamvr/bin/linuxarm64/libopenvr_api.so`:
+
+| OpenVR call | Result |
+|---|---|
+| `FindOverlay("valve.steam.desktopgame.2000999031")` | Success |
+| `GetOverlayWidthInMeters` | Success, 2.67 m |
+| `SetOverlayWidthInMeters` (same width) | Success |
+| `GetOverlayTransformType` | Success, type 5 (`VROverlayTransform_DashboardTab`) |
+| `GetOverlayTransformAbsolute` | 18 (`WrongTransformType`) |
+| `SetOverlayTransformAbsolute` (identity rotation, 1.2 m up, 1.5 m forward) | 12 (`PermissionDenied`); type remained 5 |
+
+The public interface names type 5 **DashboardTab**; SteamVR's dashboard code
+places these panels through its scene graph. It owns the frame/docking
+transforms. A successful width setter does not grant permission to restore the
+position. `vrcmd --dock-overlay world <key>` dispatched a docking request but
+the dashboard logged `Failed to get SGTransform in setInitialTransformForLocation.
+Invalid transform ID`. This does not establish working world placement.
+
+**Inferred:** saving X11 pixel rectangles or Mac window IDs would not restore
+this spatial arrangement. Mac window IDs also change when an application
+reopens; viewer tickets and reconnect keys must not go into a layout file.
+The base Mac stream reconnects after a network break, but that is different
+from recreating windows and their room positions after a reboot.
+
+There is consequently no Save/Restore control yet. A durable layout needs a
+working transform restore path, stable source identity, and a fresh capture
+permission/ticket flow. The tested gamescope-owned overlay route denies that
+transform operation. A future Frame Control-owned overlay renderer, or a
+supported platform API for dashboard frame transforms, needs its own device
+proof before building layout UI. This is a blocker for the current approach,
+not a claim that all possible implementations are impossible. Reboot recovery
+was not tested: the shared headset was not rebooted.
+
+## Panels during an immersive session
+
+**Verified 2026-09-28**, same build: our Chromium switcher panel remained in
+OpenVR's overlay list before, during and after the Frame's shipped `helloxr -g
+Vulkan` sample. During the test `vrcmd --stats` identified
+`system.generated.openxr.helloxr.helloxr`, with 242 frame submissions. The test
+ended only its own sample process; no SteamVR, Steam, power or global settings
+were changed. The switcher was still selectable afterwards.
+
+This proves survival of that panel across an OpenXR scene session, **not** that
+it stayed visibly composited over the scene: OpenVR reported it `not_visible`
+before, during and after. **Verified:** calling `ShowOverlay` on our
+*gamescope-owned* switcher overlay returns 12 (`PermissionDenied`). A helper
+cannot force that panel visible using the public overlay call. Use the
+switcher/dashboard to request access to it; we do not fight the runtime with a
+repeated force-focus loop.
+
+**Verified in a second controlled run:** a live H.264 test-pattern stream from
+this checkout's Mac helper, through its own SSH tunnel and a temporary Chromium
+profile, survived the same OpenXR sample (257 scene-frame submissions). Its
+panel `2000999032` changed from `visible` before launch to `not_visible` during
+and after the scene. The Mac helper still reported the same `test` stream;
+captured frames increased from 35 to 232, with 29.5 decoded/drawn fps afterwards.
+The test did not capture personal Mac windows or inject Mac input. The sample,
+viewer, temporary profile, tunnel and Mac helper were cleaned up. This proves
+stream survival, and also shows why it must not be advertised as always visible.
+
+**Unverified:** persistent visible placement while playing a Steam-launched VR
+game, Plasma desktop and real Mac-window behavior during that launch, and worn
+headset input. Other threads were launching games and changing the runtime on
+the shared device, so those transitions were not treated as controlled evidence.
+A runtime/X-server restart can destroy the viewer windows; a network reconnect
+cannot recreate them. No “always visible during games” guarantee is shipped.
+
+## Keyboard passthrough feasibility
+
+**Verified 2026-09-28**, same build, using `FnTable:IVRTrackedCamera_006`:
+`HasCamera(0)` returned success and true. `GetCameraFrameSize` returned 100
+(`OperationFailed`), with zero dimensions, for all three public frame types
+(distorted, undistorted and maximum-undistorted), including after acquiring the
+video service. Acquisition returned success and a handle; release returned 101
+(`InvalidHandle`). The probe shut down its OpenVR client afterwards. No camera
+frames were captured and no camera settings were changed.
+
+**Documented:** the public OpenVR camera interface provides camera frame sizes,
+intrinsics, projections and streaming handles; these are prerequisites for a
+spatially aligned camera cutout. See Valve's
+[OpenVR C API](https://github.com/ValveSoftware/openvr/blob/master/headers/openvr_capi.h).
+
+**Inferred:** camera presence alone does not establish access to camera pixels.
+The failed frame-size path blocks a keyboard cutout in our current panel
+implementation. We have not established a keyboard detector or a calibrated
+camera-to-panel mapping. Built-in full-room passthrough is not proof of a
+public, selectively masked camera stream. No keyboard cutout is offered, and
+no third-party camera/overlay app is substituted for it.
+
+## Frame Control's media theatre
+
+[The owned media player](vr-video.md) can show its video or stereo image on a
+larger, head-relative screen with its own dark surround. **Verified remotely
+2026-09-28**, SteamOS 0.4.1 / BUILD_ID 20260925.6191901: screen, eye isolation,
+surround and cleanup. It does not alter panel docking or global settings.
+Its `Overlay` RGBA rendering hook is available to stream producers; applying
+SteamVR theatre docking to existing Mac/PC panels is still unverified here.
@@ -0,0 +1,137 @@
+# PC in the headset (Experimental)
+
+**Tools → PC in the headset**, on Windows only. It's the Windows version of
+[Mac in the headset](mac-in-headset.md): a window or a whole screen on your PC
+becomes its own SteamVR panel in the Frame. Place it with the dashboard's Float
+in World, Move and Size, point and click with the laser, scroll with the
+thumbstick, and type on the PC's keyboard.
+
+**It's experimental and not yet tried on a real Windows desktop or with a
+headset.** It builds and passes its protocol tests on Windows in CI, but desktop
+capture, Media Foundation encoding and laser input have never run on a real
+PC. Expect it not to work yet. Reports are welcome through the
+[feedback form](https://frame-control.pages.dev/feedback/).
+
+## Turning it on
+
+It's off by default. Until you tick **Turn on PC in the headset** on the card,
+Frame Control starts no helper, loads none of the streaming libraries and lists
+no windows. The choice is remembered on this computer. Turning it off ends
+every stream, closes the viewers in the headset and stops the helper.
+
+The card and its Tools link only appear in the Windows app. On macOS the card
+is Mac in the headset, as before; on Linux and the iPhone app there's no card.
+
+## How it works
+
+The ported pieces are from [PR #38](https://github.com/saphid/frame-control/pull/38).
+Frame Control owns the host helper, input, authentication, streaming protocol,
+panel launcher and adaptation. It doesn't need Sunshine, OBS or any other
+streaming app. GStreamer is bundled as ordinary libraries.
+
+| | Windows |
+|---|---|
+| Capture | Windows.Graphics.Capture, through GStreamer's `d3d11screencapturesrc capture-api=wgc`, by window (HWND) or screen (HMONITOR) |
+| Encoding | Media Foundation's H.264 encoder (`mfh264enc`), low latency, no B-frames. Whether it uses the GPU depends on the PC; **not verified** |
+| Input | `SendInput`, mapped through the source's bounds, with per-monitor DPI awareness |
+
+API choices are documented, not device-verified:
+[Windows capture](https://learn.microsoft.com/en-us/windows/uwp/audio-video-camera/screen-capture),
+[GStreamer WGC](https://gstreamer.freedesktop.org/documentation/d3d11/d3d11screencapturesrc.html),
+[Media Foundation encoder](https://gstreamer.freedesktop.org/documentation/mediafoundation/mfh264enc.html).
+
+It needs Windows 10 or 11 with Windows.Graphics.Capture and a Media Foundation
+H.264 encoder (Windows "N" editions need the Media Feature Pack). Clicks and
+keys can't reach administrator apps or the sign-in screen, which an ordinary
+process can't drive. If Windows won't bring the chosen window to the front,
+input is refused rather than sent to the app covering it: bring the window
+forward, then Stop and Show again. Minimised or closed windows stop sending
+pictures. Protected video isn't supported. Mac in the headset's "own display
+per window" is Mac-only.
+
+The pieces it shares with Mac in the headset:
+
+- `ui/frame_macview.py` owns the SSH tunnel, reconnects, quality presets and
+  panel launch. `ui/frame_pcview.py` picks the Windows helper
+  (`ui/frame_pc_agent.py`, `ui/frame_pc_capture.py`, `desktop/capture.c`) and
+  holds the on/off switch. `/api/macview` is the endpoint for both.
+- `ui/mac-view.html` is the one viewer in the headset. Its frame header, H.264
+  and JPEG payloads, reconnect handshake, clock sync, timing reports and input
+  messages are unchanged.
+- `desktop/controller.c` is a C port of the Mac helper's Swift rate controller,
+  for the Windows helper. The Mac keeps its Swift original;
+  `tests/test_pc_controller.py` checks the port against a frozen 600-state
+  trace from it. Capture is gated before encoding, so encoded frames are never
+  dropped, and an idle window's last picture still goes out once the link
+  clears.
+- Stream records use the Mac helper's JSON schema (`ui/frame_stream_stats.py`),
+  so the benchmark's analysis is shared.
+
+## Packaging
+
+The Windows installer carries `resources/desktop/bundle`: `pc-host.dll` (our
+adapter and the controller), the GStreamer plugins it uses (`coreelements`,
+`videotestsrc`, `videoconvertscale`, `app`, `jpeg`, `videoparsersbad`,
+`d3d11`, `mediafoundation`) and the DLLs they depend on, all from MSYS2's
+UCRT64 packages. `libraries.json` records the GStreamer version, the plugins,
+each copied DLL and the exact package versions; licence texts are in
+`licenses/`.
+
+The x264 encoder is **not** in the installer. It's GPL, and Windows encodes
+with Media Foundation, so only CI's test builds include it
+(`desktop/build.py --with-x264`). The macOS and Linux installers are unchanged.
+
+From a source checkout on Windows, build the bundle from an MSYS2 UCRT64 shell
+with the packages listed in `.github/workflows/pc-host.yml`:
+
+```sh
+python desktop/build.py              # what the installer ships
+python desktop/build.py --with-x264  # for the native tests
+```
+
+A missing bundle is reported on the card; nothing is downloaded on first use.
+
+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.
+
+## Measuring
+
+The Mac benchmark accepts the PC host:
+
+```sh
+python scripts/macview-bench.py run --pc --scenario test --label pc-test
+python scripts/macview-bench.py run --pc --scenario capture --source window:12345 --label windows-window
+```
+
+`capture` needs a window or screen id from the helper's `/windows` or
+`/displays`. CPU sampling isn't available on PC hosts. See
+[Mac in the headset](mac-in-headset.md) for the targets and shaping options.
+
+## What's verified
+
+- **Verified in CI, this PR:** on `windows-latest`, the native adapter builds
+  and a real x264 synthetic stream passes the ticket, reconnect, timing and
+  input-echo tests through the helper. The Windows installer builds with the
+  bundle, and its bundled Python loads the bundle, starts the helper and
+  confirms x264 isn't in it. Unit tests check the Windows-only, off-by-default
+  gating, the card and the Tools link on macOS, Windows and Linux.
+- **Verified, #38, 2026-09-28, real Frame on SteamOS BUILD_ID 20260925.6191901:**
+  the PC helper (ARM64, on the Frame itself, x264 test pattern) streamed
+  through a Mac SSH relay into the shared viewer: 400 frames at 39.3 fps,
+  content latency p50/p95 15.1/28.6 ms, synthetic input p50 76.6 ms, at
+  revision `cd20243`. The frame-rate and late-frame targets failed. This
+  proves the helper and the viewer work together; it isn't Windows capture or
+  a real PC link.
+  [Raw result](../bench/results/2026-09-28-cd20243-pc-agent-frame-arm64-final.json),
+  [earlier probe](../bench/results/2026-09-28-c5fc552-dirty-pc-agent-frame-arm64-hairpin.json).
+- **Verified, #38, on a Mac:** the C controller matched all 600 states of the
+  Swift controller's congestion and recovery trace.
+- **Not verified:** anything on a real Windows desktop. Windows.Graphics.Capture,
+  Media Foundation (hardware or software), encoder restarts when the bitrate
+  changes, mixed-DPI and multi-monitor input alignment, window resize and
+  minimise, non-US keyboard layouts, physical laser input, and how it performs
+  in a worn headset. No numbers are claimed for these. Not tested with a
+  headset from this branch.
+- **Not reviewed independently.** #38 tried an independent review that never
+  completed.
@@ -0,0 +1,298 @@
+# Privacy and analytics
+
+Frame Control sends anonymous analytics to [PostHog](https://posthog.com)
+(US cloud) so the maintainer can see how many people use it, which features
+matter and where installs fail. You choose how much in **Privacy & updates**,
+the last panel on the page. `ui/frame_telemetry.py` is the whole
+implementation.
+
+## The three levels
+
+| Level | Default | What it sends |
+|---|---|---|
+| Anonymous usage statistics | On, after a notice on first run | The events in the table below |
+| Share compatibility results | Off | Your Android compatibility reports and tests |
+| Send error details | Off | Scrubbed error messages and tracebacks |
+
+Nothing is sent until the first-run notice has been shown. The notice's
+**Share more to help fix problems** button turns on the second and third
+levels together. Either can be turned off later. Turning a level off
+deletes that level's events that haven't been sent yet.
+
+**Show what's been sent** in the panel lists the last 50 events that left your
+computer, exactly as they were sent.
+
+## Anonymous
+
+- Events carry a random id, made when Frame Control first runs and kept in
+  its data folder (`telemetry/settings.json`). It isn't derived from your
+  computer, account or network. To get a new one, delete that file.
+- Events are sent without person profiles (`$process_person_profile: false`)
+  and without location lookup (`$geoip_disable: true`). Each carries a
+  placeholder address (`$ip: 0.0.0.0`), so PostHog stores that instead of
+  yours.
+- Every event includes the app version, OS name (macOS, Windows or Linux),
+  CPU architecture and Python version.
+- `app_installed`, `app_updated` and `app_opened` carry `country`: a
+  two-letter code such as `DE`, worked out on your computer from its time zone
+  setting (`Europe/Berlin` becomes `DE`), so we can see roughly where Frame
+  Control is used without knowing your address. The time zone itself isn't
+  sent, and a setting that isn't a country's zone (such as UTC) sends nothing.
+  Copies that showed the first-run notice before it mentioned this (before
+  0.4.6) show it again, and don't send a country until it has been seen.
+- `frame_connected` carries `headset_id`: the first 16 hex digits of a
+  SHA-256 hash of the headset's `/etc/machine-id`, worked out on the headset,
+  so the machine-id itself never leaves it. It is the same for one headset on
+  every computer, which lets us count headsets rather than installs; it can't
+  be turned back into the machine-id or a serial number. `headsets_saved` is how
+  many headsets this copy of Frame Control has set up.
+
+## Usage events
+
+| Event | When | Properties besides the common ones |
+|---|---|---|
+| `app_installed` | First run | `country` |
+| `app_updated` | First run of a new version | `from_version`, `country` |
+| `app_opened` | At most once a day | `country` |
+| `frame_connected` | The first time a SteamOS build is seen on each headset | `steamos_build`, `steamos_version`, `headset_id`, `headsets_saved` |
+| `tab_viewed` | The first click on each tab in a session | `tab` |
+| `install_finished` | Any install finishes, working or not | `kind` (apk, flatpak, steam, title, web), `ok`, `seconds`, `error_category` (only when `ok` is false), `installer_code`, `xr_layer_missing`, and see below |
+| `update_offered`, `update_started`, `update_failed` | The update banner | `to_version`, `error_category` |
+
+`install_finished` never includes a file name, path or error message. An
+error becomes one category from this fixed list, plus Android's own
+`INSTALL_FAILED_…` code (`installer_code`) when there is one:
+`android_installer`, `apk_needs_newer_android`, `apk_wrong_abi`,
+`layer_missing`, `apk_repack_failed`, `tool_missing`, `apk_unreadable`,
+`cant_run_on_frame`, `steam_shortcut`, `steam_not_running`,
+`steam_not_responding`, `steam_js_error`, `artwork_failed`, `copy_timeout`,
+`frame_not_set_up`, `frame_auth`, `frame_unreachable`, `frame_disk_full`,
+`copy_failed`, `download_failed`, `flatpak`, `cancelled`, `lepton`,
+`internal_error` (a mistake in Frame Control's own code) or `other`. A VR
+APK that installed without Frame Control's OpenXR compatibility layer,
+because this copy of the app is missing it, carries `xr_layer_missing: true`;
+otherwise that field is left out. It names what was installed only when
+that's already public:
+
+- F-Droid catalogue apps: `package`. Never the version, since a local build can reuse a
+  catalogue app's package name
+- Flathub apps: `flatpak_id`
+- Steam games: `steam_appid`
+- A sideloaded title: only its runtime (Proton or Linux)
+
+Any other APK is sent as `catalog: false`, with no name.
+
+## Compatibility results (opt-in)
+
+Each report becomes a `compat_report` event with the fields the Report dialog
+shows: package, version, result or rating, your notes, how it was run, and the
+SteamOS and Lepton builds. Before sending:
+
+- the notes, app name and version are scrubbed like error messages (see
+  below)
+- the APK's source is kept only if it's `F-Droid`, `meta-store` (a test
+  from [My Meta library](meta-library.md)) or the public host name of a
+  download link (`https://example.com/…`). File names, user names,
+  passwords, ports, paths, IP addresses and local host names are dropped
+
+When you turn this on, reports you made earlier on this computer are shared
+too.
+
+My Meta library (Experimental) keeps your Meta access token in
+`meta-account.json` in this data folder, owner-only, and sends it only to
+Meta's API and download hosts. It is never in analytics, reports, error
+details or logs; its test reports carry the package, version, version code,
+title, result and the reason, nothing from your account. See
+[meta-library.md](meta-library.md#your-token-and-privacy).
+
+The maintainer's `python3 ui/frame_compat_db.py sync` copies these events
+into the compatibility database, marked `via=community…`. It takes at most
+30 per reporter per day.
+
+## Error details (opt-in)
+
+`$exception` events carry an error message, the Frame Control file, line and
+function it came from, and the request that failed (for example
+`POST /api/android install`). Before anything is sent, the message is
+scrubbed:
+
+- your home folder becomes `~`, and any user name becomes `<user>`
+- IP and MAC addresses, email addresses, `.local`, `.lan` and Tailscale host
+  names, Steam ids, SSH and PEM keys, API tokens and long hex strings are
+  replaced
+- URLs are cut down to their scheme and a public host name, or `<url>`. User
+  names, passwords, ports, paths and queries are dropped
+- `token=`, `key=`, `password=` and similar values are replaced
+
+The same error is sent at most once every 10 minutes. When ssh reports that
+it couldn't reach the Frame (asleep, away, or not set up yet), or that ssh and
+the Frame refused each other's keys (a changed host key, or this computer's
+key not accepted), that is sent at most once per kind each time Frame Control
+runs.
+
+## Report a problem
+
+**Report a problem** is the speech-bubble button in the header, also in the
+Settings page (**Privacy & updates**) and under **Help → Report a Problem…**. It sends the report
+privately to Frame Control's PostHog project as a `problem_report` event, the
+same way as the analytics above, so only the maintainer can read it and
+nothing is published. It works whatever the analytics settings are, because
+the person sends it deliberately. The report has the kind, title and text you
+wrote, a short reference shown after sending, and the diagnostics below. Your
+email address goes with it only if you tick **The maintainer may contact me
+with follow-up questions** (the report then carries `contact_followup: true`);
+it's filled in from **Contact email** below when you've agreed there. It has its own random id, so it isn't linked to
+your analytics events. With that box ticked, the address also becomes your
+**Contact email** below with follow-up questions ticked, so you remove it there
+like any other. If it's a different address from the one saved there, it
+replaces it, and update notices stop until you turn them on again (they were
+agreed for the old address); the form says so before you send. The report then also
+carries this copy's contact id and change number (`contact_id`, `contact_rev`,
+see below), so removing or changing the address later takes back the
+follow-up permission given with the report too.
+
+With **Include diagnostics** ticked (the default), the report adds:
+
+- the app version and whether it's a built app
+- the OS, its release and CPU, and the Python version
+- the Frame's SteamOS build, if it has connected since the app started
+- which analytics levels are on
+- a short connection summary, so "the app can't find the headset" can be
+  diagnosed. It is made of fixed words and counts only; no text from an
+  error message or from ssh, and no name you chose, goes in it:
+  - the connector's state (idle, connecting, connected or failed), the stage
+    it failed at (network, find, ssh, identity or login), the attempt number
+    and why it started (such as "Starting up" or "Trying again")
+  - how many headsets are saved; whether the active one's ssh alias is the
+    default `frame` or a custom one (a custom alias itself is never named);
+    whether it's saved or a bare ssh alias; and its number of addresses
+  - for the current error and the last failure: the stage and an error
+    category (the same fixed names as error details, such as
+    `frame_unreachable`, `frame_not_set_up`, `frame_auth` or `other`), and
+    how long ago the last failure was. The category is decided when the
+    failure happens
+  - for each address tried, only its kind (`.local`, `ipv4`, `ipv6`,
+    `ipv6 link-local`, `tailscale`, `hostname`, `alias` for one read from
+    `~/.ssh/config`, and what a name resolved to, such as
+    `.local->ipv6 link-local`) and how the try went (answered, unresolved,
+    timeout, refused, unreachable, sshfailed). Never the address itself
+  - when connected, the kind of address used and its round trip in ms
+  - whether this computer has a network gateway, and whether Tailscale is on,
+    off or not installed
+  - which kind of `ssh` the app runs (Windows OpenSSH in System32, OpenSSH in
+    Program Files, Git for Windows, MSYS2/Cygwin, Homebrew or /usr/local,
+    Nix, the system one, or "other"), never its path; its version, rebuilt
+    from the numbers in the banner `ssh -V` prints (such as
+    `OpenSSH for Windows 9.5p1, LibreSSL 3.8.2`; anything that isn't a
+    plain OpenSSH banner is reported as `unknown`); and the kinds of any
+    other `ssh` on the PATH.
+    This is looked up once in the background after the app starts, so a
+    report sent straight away may say "still being checked"
+  - whether `~/.ssh/config` exists, whether it has Set Up Connection's
+    managed block for the active alias, how many managed blocks it has, and
+    whether a hand-written `Host` line also names the alias (yes or no; the
+    alias isn't named)
+
+**Also include recent activity and the server log** is off by default,
+because those lines can name files and apps. When ticked, it adds the newest
+Activity lines and server log lines, without the request lines. The server
+log has a line for each failed connection attempt: its stage, the message
+shown on the connection pill, ssh's last line about it, and the address kinds
+above. That free text is scrubbed when the line is written: the addresses, ssh alias
+and display name of the headset being tried (whole, whatever their case), and
+then any name ssh gives after "hostname", "host" or "to", become `<host>`; a Windows home folder's
+whole name (spaces and apostrophes included) becomes `<user>`, and ssh's
+whole `user@host:` field (spaces, `DOMAIN\user` and full domain names
+included) becomes `<user>@<host>:`; then the scrubbing below. A host name ssh mentions
+in some other wording can still get through, so check the log lines in **Show
+exactly what's included** before sending. A failure that repeats on every
+retry is written at most once every 5 minutes.
+
+Everything is scrubbed like error details and limited to what fits in the
+report. Environment details are kept first, then the newest lines. **Show
+exactly what's included** shows the snapshot that will be sent, and later
+activity isn't added to it. If PostHog can't be reached, **Copy report** puts
+the whole report on the clipboard.
+
+The maintainer reads reports on the Frame Control dashboard in PostHog, or
+with `python3 ui/frame_report.py inbox [days]`, which uses the same personal
+API key as `frame_compat_db.py sync`.
+
+## Contact email (optional)
+
+Frame Control never needs an email address. If you'd like to leave one, there
+are two separate choices, both off until you tick them:
+
+| Choice | What it's for |
+|---|---|
+| **Email me about Frame Control updates** | Occasional notices about new releases and updates |
+| **The maintainer may contact me with follow-up questions** | Questions about problem reports you send, mostly |
+
+You're asked once, in a bar at the top of the page, after the Frame has
+connected for the first time, and never in the same visit as the first-run
+privacy notice. **No thanks** hides it for good, and it isn't
+shown again even if you ignore it. **Contact email** in **Privacy & updates**
+is where you add, change or remove the address and either choice at any time.
+
+**What's sent, and where.** The address and the two choices go privately to
+Frame Control's PostHog project, the same place as problem reports, as a
+`contact_consent` event with `email`, `updates`, `followup`, `action` (`set`
+or `withdraw`) and the common properties above. Only the maintainer can read
+that project, and nothing in it is published or shared. It's sent only when
+you save, or when you send a problem report with follow-up questions ticked,
+whatever the analytics settings are, because you chose to. With a report, the
+address and choices are saved before the report is sent and stay saved if it
+fails; like any change, they're sent as soon as PostHog can be reached. It
+carries its own random contact id, not the analytics id, so it isn't linked
+to your usage events, and a `rev` number that goes up with each change, so
+the newest choice always wins. Like everything else sent, it's listed under
+**Show what's been sent**. On this computer the address and choices are kept in
+`contact/contact.json` in Frame Control's data folder. An address is only
+kept with at least one choice ticked.
+
+**Removing it.** **Remove my email** (or clearing the address and saving)
+deletes it from this computer, including from the **Show what's been sent**
+log (in earlier contact events and problem reports), and sends a `withdraw`
+event with no address in it. The maintainer's list only uses the newest event from each copy, so from
+then on the address isn't listed for either choice. Unticking one choice
+works the same way for that choice. This also covers problem reports you sent
+from this copy with follow-up questions ticked: if your newest choice since the
+report (by change number, not the clock) no longer agrees to follow-up
+questions at that address, the maintainer's inbox shows the permission as
+withdrawn and leaves the address out. If you're offline, the change waits on
+this computer and is sent when PostHog can be reached. The earlier event
+stays in PostHog until its data retention removes it; to have it deleted
+sooner, ask the maintainer (for example in a problem report).
+
+Nothing sends email yet: this only records who agreed to what. The
+maintainer lists the addresses with
+`python3 ui/frame_report.py contacts [updates|followup]`, which uses the same
+personal API key as `inbox`.
+
+## Turning it all off
+
+Untick the boxes, or set `DO_NOT_TRACK=1` or `FRAME_CONTROL_TELEMETRY=0` in
+the environment that starts Frame Control. A copy run from a source checkout
+never sends analytics unless `FRAME_CONTROL_TELEMETRY=1` is set.
+
+These switches cover the analytics above. A problem report or a contact email
+is sent only because you pressed its Send or Save button, so those still go
+when you choose to send them (a contact change saved while offline is sent
+by itself once PostHog can be reached); if you don't, nothing is sent.
+
+## Update checks
+
+The desktop app asks GitHub for the latest release shortly after starting,
+then every 6 hours: the latest release's `update.json` on GitHub, or
+`api.github.com/repos/saphid/frame-control/releases/latest` if that fails.
+Those requests carry no id. To stop it, set
+`FRAME_CONTROL_NO_UPDATE_CHECK=1`. See [releasing.md](releasing.md).
+
+## Screenshots and clips
+
+Copying screenshots and Steam clips to your computer, by hand or with
+**Settings → Copy to this computer**, goes straight from the headset to your
+computer over the same SSH connection as everything else. Nothing is uploaded
+anywhere. What automatic copying has copied is listed, by file id, in
+`auto-sync.json` in Frame Control's data folder; see
+[recording.md](recording.md#copying-new-ones-automatically).
@@ -0,0 +1,219 @@
+# Import from my Quest
+
+Move games you own from your own Meta Quest onto the Steam Frame. Plug the
+Quest into your computer by USB; Frame Control copies the game off it with
+adb, checks it can run in Lepton, and installs it on the Frame like any other
+APK, with its own Lepton instance and Steam library entry
+([apks.md](apks.md), [vr-apks.md](vr-apks.md)).
+
+There is no licence-check bypass, and there won't be. A game that uses Meta's
+licence check installs with a warning: on the Frame that check can't reach
+Meta, so the game may quit, hang, or run without online features. See
+[What won't run](#what-wont-run-and-why) below. To install games you own
+straight from Meta instead, without a Quest, see [My Meta library
+(Experimental)](meta-library.md).
+
+## Before you start: developer mode on the Quest
+
+adb only reaches a Quest with developer mode on:
+
+1. Make a free Meta developer organisation at
+   [developers.meta.com](https://developers.meta.com/horizon/) on a **verified**
+   Meta account (Meta asks for a phone number or payment card to verify).
+2. In the Meta Horizon phone app, open the headset's settings and switch
+   **Developer Mode** on. Restart the Quest.
+3. Plug the Quest into the computer that runs Frame Control with a USB cable
+   that carries data (some charging cables don't).
+4. Put the headset on. Accept **Allow USB debugging**, and tick **Always allow
+   from this computer**.
+
+Frame Control uses the adb it already uses for the Frame (an adb you
+installed, else the one it bundles). On Linux, adb also needs permission to
+open the Quest: add a udev rule for Meta's USB vendor id `2833`, for example
+`SUBSYSTEM=="usb", ATTR{idVendor}=="2833", MODE="0660", GROUP="plugdev"`, and
+be in that group. Then unplug and replug.
+
+The Frame's own adb connections (Frame Control's SSH tunnels to Lepton,
+`127.0.0.1:<port>`) are never touched. Only a device that reports itself as
+Oculus or Meta, or a Quest model, counts as a Quest.
+
+## Using it
+
+In the app: **Android → On your Frame → Import from my Quest**. Choose **Look
+for my Quest**, then **Import** next to a game. Tick **Also copy each game's
+files in Android/data** to bring the game's external files too (some games
+keep saves there). Progress shows in the panel; the result and any reason it
+stopped appear in the message log.
+
+You can choose **Import** on several games at once: they wait in the **Install
+queue** (above the Quest panel) and import one after another, together with
+installs from the Android store and My Meta library. **Remove** takes a game
+out of the queue before it starts. Each queued import runs on the headset that
+was selected when you queued it; if you switch headsets first, it stops with a
+message instead.
+
+A game with game data (an OBB folder, or Android/data when ticked) takes two
+imports, because Android can only take that data while the game's Lepton
+instance is running, and there is no way to start the instance without the
+game. Frame Control never starts the game itself:
+
+1. **Import** installs the game and says it has data to add.
+2. Start the game from the Steam library and leave it on its first screen.
+3. **Import** again. Frame Control sees the game open with the same version,
+   copies the data in, and closes the game. Start it again to play.
+
+A game that quits straight away without its data can't take it this way.
+
+From a terminal:
+
+```sh
+python3 ui/frame_quest.py devices                      # what adb sees, and which is the Quest
+python3 ui/frame_quest.py list                         # apps on the Quest: split or not, OBB and data sizes
+python3 ui/frame_quest.py import com.example.game      # copy one to the Frame
+python3 ui/frame_quest.py import com.example.game --data --serial 1WMHH000XXXXXX
+```
+
+The API is `GET /api/quest` (the Quest, or guidance when there isn't one, and
+its apps) and `POST /api/quest` with
+`{"action": "import", "package": "...", "serial": "...", "data": false}`,
+which returns a background job (`/api/job`) with a stage and percentage. The
+job waits in the install queue first (`"queued": true`, stage `Queued: N
+installs before it`). `GET /api/queue` lists the running install and those
+waiting; `POST /api/queue` with `{"action": "remove", "job": "..."}` removes a
+waiting one (409 once it has started).
+
+## What an import does
+
+1. Finds the Quest (`adb devices -l`, then `ro.product.manufacturer` and
+   `ro.product.model`). If none is ready it says why: nothing connected, not
+   authorised yet (the **Allow USB debugging** prompt), no USB permission
+   (Linux), offline, another Android device, or several Quests.
+2. Lists third-party apps (`pm list packages -3`), each with `pm path` (one
+   line is a single APK, several a split APK) and the size of
+   `/sdcard/Android/obb/<package>` and `/sdcard/Android/data/<package>`.
+   Labels aren't shown before import: reading them means copying the APK.
+3. Checks there is room on the computer (the APKs, a merged and a patched copy
+   of them counted uncompressed, the OBB folder and the Android/data folder
+   once, since that is streamed to the Frame), then copies the APK and its splits
+   into a per-user cache folder (`Frame Control/quest-import` under the
+   system's cache folder).
+4. **Checks the APK before anything else is copied**, refusing:
+   - 32-bit-only builds (native code but no `arm64-v8a`);
+   - apps that need an Android newer than Lepton's (API 30);
+   - split APKs Frame Control can't merge (below).
+   Installed with a warning:
+   - games that use Meta's licence check (Meta's Platform SDK,
+     `libovrplatformloader.so`, in the base or any split): "uses Meta's
+     licence check; it may quit, hang, or run without online features on the
+     Frame. Frame Control won't remove the check." Linking the SDK doesn't
+     mean a game enforces the check: Meta's store requirement for it
+     (VRC.Quest.Security.1) has only been *Recommended* since 2021, and what
+     a game does when the check fails is up to the game;
+   - VrApi-only builds (`libvrapi.so`, no OpenXR loader), flagged as for any
+     APK: the Frame has no VrApi, so they may not run.
+5. If the game isn't open on the Frame: installs it through the normal APK
+   install (`frame_android.install`: VR detection, the launcher fix, the
+   OpenXR compatibility layer, artwork), and says whether it has game data
+   to add. An older Quest copy never replaces a newer version on the Frame
+   (compared by version code, which installs now record; for installs made
+   before that, it is read from the installed APK on the Frame) unless you
+   tick **Allow replacing a newer version** (`--allow-downgrade`,
+   `"allowDowngrade": true`). If the installed version can't be read at all,
+   that permission is needed too.
+6. If the game is open on the Frame (the second import): refuses unless
+   Frame Control's install record is for this package, in this package's own
+   instance, at the Quest's exact version code (an open game is never
+   replaced), otherwise copies the OBB folder (and `Android/data` if ticked)
+   from the Quest, waits for Android to have finished booting, copies standard
+   `main|patch.<version>.<package>.obb` files with the existing OBB helper
+   (checked by SHA-256 inside Android, [sidequest.md](sidequest.md#obb-files))
+   once Android's own `dumpsys package` in that instance shows the same
+   package at the same version code,
+   streams the `Android/data` files in as a tar written straight into the SSH
+   transfer, then closes the game so its next start reads them, and checks
+   its instance has stopped; if it hasn't, or the check itself fails (a
+   failed `podman ps` never reads as "stopped"), the result says so. The game
+   is closed whatever happens during the copy, including a local read error
+   or Ctrl-C, and a failed close is reported without hiding the original
+   error. Everything
+   from the second check of the install record (after the pulls, which can
+   be long) to closing the game holds the same lock as installs and removals.
+   Other files in the OBB folder are listed as not copied.
+7. Deletes the cache folder, whether the import worked or not. A folder left
+   by a crash is removed by a later import after six hours.
+
+Only one import runs at a time. If copying game data fails, the game is
+closed and the message says so; start it and import again.
+
+### Split APKs
+
+Lepton runs one APK per instance, so the shared install path now merges a
+split APK into one (`frame_android.install(base, splits=[...])`, or
+`python3 ui/frame_android.py install base.apk split1.apk ...`). Merging copies
+each split's native libraries and assets into the base, drops the base
+manifest's `isSplitRequired` flag, and re-signs with Frame Control's key, as
+the launcher fix already does. Only splits that hold nothing but
+`lib/<abi>/*.so`, `assets/` and their own signature, manifest and an empty
+resource table are merged. Refused, with the reason, before anything reaches
+the Frame: splits with resources (screen density or language), code
+(`classes*.dex`), any other file, a manifest that declares components or
+libraries (`<service>`, `<uses-library>`...), feature splits, and splits of
+another package. The merged files are held in memory while the APK is
+repacked and signed (as for every patched APK), so a merge adds at most
+512 MB per file and 1 GB in all. Each file is read from its raw bytes and
+checked against both of its ZIP headers: the local header must name the same
+file with the same method, sizes and checksum as the central directory (or a
+data descriptor must carry them; ZIP64 sizes are understood), the data must
+end before the next entry, a stored file must be exactly its declared size, a
+compressed one must end exactly at the end of its data with nothing after it,
+and its checksum must match. So a forged header can't truncate a file or fill
+memory. A resource table counts as empty only
+if it defines nothing at all (styles and other complex entries included). A merged, re-signed APK can't update a
+copy signed by the publisher.
+
+## What won't run, and why
+
+- **Games that enforce their Meta store licence.** Many store games link
+  Meta's Platform SDK (`libovrplatformloader.so`), for the licence check and
+  for user ids, achievements, multiplayer, cloud saves or DLC. On the Frame the
+  SDK can't reach Meta's services, which only exist on Meta headsets. A game
+  that enforces the check then quits or hangs; one that doesn't may run
+  without its online features. Frame Control can't tell which from the APK, so
+  it installs these with a warning and **won't remove or work around the
+  check**, and doesn't bundle, call or link to any tool that does. If yours
+  quits, ask the developer for a build without the check, or get the game from
+  Steam or itch.io. Most SideQuest, itch.io and open-source games don't link
+  the SDK.
+- **32-bit games** (Quest 1 era): the Frame runs 64-bit ARM only. Refused.
+- **VrApi-only games**: the Frame has OpenXR, not Meta's older VrApi. They
+  install with a warning, but may not run.
+- **Anything else that needs Meta**: avatars, Quest-only device checks,
+  passthrough and other `XR_FB_*` features SteamVR lacks
+  ([vr-apks.md](vr-apks.md#what-steamvrs-android-runtime-supports-verified)).
+- **Private app data** (`/data/data/<package>`) can't be read over adb without
+  root, so only saves in `Android/data` come across.
+
+What usually works: OpenXR games without Meta's licence check, for example
+open-source and itch.io builds you sideloaded onto the Quest.
+
+Only import games you own, for your own use.
+
+## Evidence and limits
+
+**Verified locally** with a fake adb (`tests/fakeadb.py`, `tests/test_quest.py`):
+device detection (Quest, the Frame's tunnel, the Frame over USB, other
+Android devices, unauthorised, no permissions, offline, none, several),
+package listing and split detection, the refusals and the licence-check warning, the order of copies
+(APK checked before the OBB is copied), the two-step install then data copy
+(the game is never started by Frame Control, never replaced while open, and
+closed after its data is copied), split merging and signing, the split
+refusals and size limits (including a forged zip bomb), the space estimate,
+and cache clean-up after success and failure.
+
+**Not verified:** a real Quest (none was available), Quest system apps in
+`pm list packages -3`, real split APKs from the Meta Store, OBB and
+`Android/data` writes into a running Lepton instance on a real Frame
+(the OBB helper itself is also unverified there), whether games stay open
+without their OBB long enough for the second import, and the Linux udev rule.
+Large games are copied into memory while being merged or patched, as every
+patched APK already is.
@@ -0,0 +1,165 @@
+# Recording and Steam clips
+
+Frame Control can record what the headset shows, with its sound, and save it as
+an `.mp4` on your computer. It can also export the clips and background
+recordings from Steam's own game recording, at any length, as one `.mp4` each.
+Both are on **Tools → Capture & Live**. Everything is saved to
+`~/Videos/SteamFrame`, next to the screenshots in `~/Pictures/SteamFrame`.
+Nothing needs installing on the computer or on the Frame: the work is done by
+the `ffmpeg` that SteamOS ships.
+
+**Recording is tested on a headset; Steam clip export isn't yet.** On a Frame
+running SteamOS 0.5.6 (build 20261015.6141207), Record and Stop saved an
+8-second 1080p H.264 recording with AAC sound to the computer, Stop took about a
+second, and nothing was left running or stored on the Frame. That Frame had no
+Steam recordings, so the clip list came back empty and export has only been
+tested with files in Steam's format made by `ffmpeg`. The parts we're least sure
+of are listed under [What hasn't been checked](#what-hasnt-been-checked).
+
+## Record
+
+Press **Record** next to **Live**. The button turns into **Stop**, and a line
+under the viewer counts the time. Press **Stop**: the recording is finished on
+the Frame, converted to `.mp4` there (without re-encoding) and copied to
+`~/Videos/SteamFrame/rec-<date>-<time>.mp4`. The line says when it's saved.
+
+- **What's in it**: the same picture as **Live**, SteamVR's copy of the headset
+  view (one undistorted 1920×1080 eye, from `/dev/video99`), at 30 fps and
+  8 Mbit/s, encoded with x264 (the Frame's hardware encoder crashes `ffmpeg`).
+  The sound is what the headset plays: the monitor of PipeWire's default
+  output, read through its PulseAudio server (`-f pulse -i @DEFAULT_MONITOR@`).
+  The microphone isn't recorded.
+- **Without sound**: before each recording Frame Control checks for a fraction
+  of a second that `ffmpeg` can read that audio. If it can't, the recording goes
+  ahead as video only, and the page and the saved message say so.
+- **Limits**: a recording stops by itself after 20 minutes and is saved as if
+  you'd pressed Stop. It needs about 2.9 GB free on the Frame to start (room for
+  a full-length recording twice, while it's converted), and checks the space on
+  your computer before copying. One recording runs at a time.
+- **Where it waits on the Frame**: `~/.cache/frame-control/video/`, not `/tmp`,
+  which is a RAM disk on SteamOS. It's deleted once copied.
+- **If the connection drops or the app quits** while recording, the Frame
+  stops `ffmpeg` cleanly when the SSH channel closes: it types `q` into
+  `ffmpeg`'s input, as at its keyboard, and kills it if it's still there 10
+  seconds later. Not a signal: SteamOS's `ffmpeg` 7.0 catches SIGTERM and
+  SIGINT and carries on recording. A recording that was never copied is
+  deleted when the next one starts.
+
+Recording and **Live** can run together; both read the same device.
+
+## Steam clips
+
+**Steam clips** lists what Steam's game recording has kept on the Frame: saved
+clips and background recordings, newest first, with the game, length and size.
+**Export** joins one into a single `.mp4` in `~/Videos/SteamFrame`, named after
+its folder (for example `clip_620_20261020_202134.mp4`), however long it is.
+Steam's own export is reported to stop at about 60 seconds. Exports run in the
+background; ones already on your computer are marked.
+
+How it works: Steam keeps each recording as MPEG-DASH segments, one folder per
+recording session.
+
+```
+~/.local/share/Steam/userdata/<account>/gamerecordings/
+  video/bg_<appid>_<YYYYMMDD>_<HHMMSS>/                 background recording
+  clips/clip_<appid>_<YYYYMMDD>_<HHMMSS>/video/bg_<...>/ a saved clip
+      session.mpd
+      init-stream0.m4s  chunk-stream0-00001.m4s  chunk-stream0-00002.m4s ...
+      init-stream1.m4s  chunk-stream1-00001.m4s ...
+```
+
+Each stream's `init-stream` segment followed by its chunks, in number order, is
+a playable fragmented MP4 of that stream. On the Frame, `ui/frame_clips.py`
+builds those files and has `ffmpeg` put the video and audio into one `.mp4`
+with `-c copy`, so nothing is re-encoded and it takes seconds. The result is
+copied to the computer and deleted from the Frame. The export needs free space
+on the Frame of about twice the recording's size while it runs.
+
+The listing doesn't rely on the folder names: any folder under `clips/` or
+`video/` (up to four levels down) holding `init-stream*.m4s` counts. A stream
+missing its init segment or its chunks is left out. Missing chunk numbers are
+shown as "pieces missing"; the rest still exports.
+
+## Copying new ones automatically
+
+**Settings → Copy to this computer** copies new screenshots and Steam clips
+here on their own. It's off until you turn it on, with a switch each for
+screenshots and Steam clips. While a headset is connected, Frame Control looks
+about once a minute and copies what's new with the same code as **Save** and
+**Export**: screenshots to `~/Pictures/SteamFrame`, clips to
+`~/Videos/SteamFrame`.
+
+- **Only what's new.** It copies what you take after turning it on, so the
+  shots and clips already on the headset don't all arrive at once. **Copy
+  everything now** copies everything that isn't on this computer yet, once.
+- **Deleted stays deleted.** What it has copied is remembered, per headset, in
+  `auto-sync.json` in Frame Control's data folder, so a file you delete here
+  isn't copied again (**Copy everything now** would bring it back).
+- **Saved clips only.** Steam's background recording is a rolling buffer, not
+  something you chose to keep, so it isn't copied; export it by hand. A clip
+  Steam wrote to in the last 30 seconds is still being saved and waits for the
+  next look.
+- **It gives way.** It waits while an install, an export, a recording or
+  anything else is running on the headset, and stops between files if one
+  starts. While a copy is running, switching headsets waits for it, as it does
+  for an export. When the headset can't be reached it waits longer each time,
+  up to 15 minutes, and starts over once it answers.
+- **When this computer is full** (under 512 MB free, or a copy runs out of
+  space) it pauses and says so in Settings and above the lists. Free some space
+  and press **Resume**.
+- **With several headsets** it copies from the one you're connected to.
+
+Settings shows when it last checked, how many files it last copied and when,
+and the last problem it had. The screenshot and clip lists say when it's on.
+
+## For developers
+
+`ui/frame_clips.py` holds the logic, so other features can reuse it (automatic
+copying, in `ui/frame_autosync.py`, does):
+
+- `list_clips(run_frame)` and `export_clip(id, run_frame, run, fetch)` are the
+  computer's side; the server passes how to reach the Frame.
+- `list_recordings(userdata)` and `export(userdata, id)` are the Frame's side,
+  run as `python3 - list` / `python3 - export ID` with the file on stdin.
+- `Recorder` runs one recording from Start to the copy.
+- `parse_id` checks a clip id (`<account>/<clips|video>/<folder>...`, plain
+  names only, no `.` or `..`) before it reaches a path or a shell.
+
+The API, with the same guards as every other request:
+
+| Request | Does |
+|---|---|
+| `GET /api/record` | The recording's state: `idle`, `starting`, `recording` (with `elapsed`, `audio`, `limit`), `stopping`, `saving`, `saved` (with `file`) or `failed` (with `error`) |
+| `POST /api/record {"action": "start"\|"stop"}` | Start or stop |
+| `GET /api/clips` | Steam's recordings on the Frame, each with `saved` |
+| `POST /api/clips/export {"id"}` | Starts an export; poll `/api/job` |
+| `POST /api/open {"what": "videos"}` | Opens `~/Videos/SteamFrame` |
+| `GET /api/autosync` | Automatic copying: `screenshots`, `clips`, `since` (the watermark), `checked`, `copied`, `copied_time`, `error`, `paused`, `waiting` (clips still being written) |
+| `POST /api/autosync {"action": "settings", "screenshots", "clips"}` | Turn either on or off; turning it on sets `since` to now |
+| `POST /api/autosync {"action": "now"\|"resume"}` | Copy everything not here yet, once; or resume after running out of space |
+
+They aren't available when the server runs on the Frame itself (the iPhone
+app), since they save to a computer.
+
+Tests: `tests/test_clips.py`, and `tests/test_autosync.py` for automatic
+copying (with a fake headset). They run the recording's shell under bash with a
+stand-in `ffmpeg`, so Stop, the time limit and the clean-up are tested as they
+run on the Frame. Where `ffmpeg` is installed, they also make a 65-second DASH
+recording with Steam's file names and export it with the real `ffmpeg`.
+
+## What hasn't been checked
+
+The 2026-10-24 read-only clip list on build `20261015.6141207` again returned
+zero clips: its `gamerecordings` directory held only `gamerecording.pb`. No
+Steam recording was created during that check, so a real Steam clip export is
+still unverified. The recording and Stop path were verified earlier on that
+build, as described above. Remaining checks:
+
+- Audio and video sync. Both inputs are stamped with the wall clock, because
+  the camera device and the audio server otherwise count from different clocks.
+- The exact layout of `gamerecordings` on the Frame's SteamOS build, and
+  whether a clip's segments start mid-recording. The parser is deliberately
+  loose about the layout.
+- How much x264 at 1080p30 costs the Frame while a game is running.
+- Automatic copying: that 30 seconds without a write is a safe sign Steam has
+  finished saving a clip, and how a once-a-minute listing feels on the Frame.
@@ -4,6 +4,9 @@ Where to get the Steam Frame's operating system, what's inside it, and how to
 run it for testing without the headset. For recovering a Frame that won't boot,
 see the boot menu and boot-loop entries in
 [how-the-frame-works.md](how-the-frame-works.md#facts-worth-knowing).
+To keep your game saves, settings and Android app data through a factory reset
+(`Erase User Data`) or a re-image, and to get out of Desktop Mode, see
+[backup-and-recovery.md](backup-and-recovery.md).

 ## Downloads