Skip to content

Record your screen, ship a demo. Free and open-source, GPU-accelerated, no watermarks, no subscriptions. Windows, macOS, Linux. Actively maintained.

388 KB
Raw Blame

Manual end-to-end checklist

This checklist covers the real desktop capture-to-export path: the parts that unit, browser, and Playwright tests cannot exercise, including real screen capture, a physical webcam, the system tray, the native compositor, and export. Run it before promoting a release candidate and after any change to native capture, preview, or export.

"Manual" is about the input, not the operator. These checks need real OS mouse and keyboard events, not a human hand — so an agent with the computer-use MCP runs them, on demand, and a request for one section after a targeted change is as much a run as the whole file before a promote. Availability check and the rule on partial runs: AGENTS.md.

Sections marked v1.8.0 cover what that release changed: chat-driven editing through the agent tool set, clip-anchored modifiers, local transcription, the macOS Metal compositor, and the new effect controls.

Sections marked post-1.10.0 cover what has landed on main since the v1.10.0 tag: the AI camera background, the caption anchor model, pixel-resolution crop, editor window bounds, update settings, the Windows recording encoder and AAC changes, imported audio and voice-over recording, and transcript word editing with word insertion. Run the whole file for a release candidate; the marked sections are the ones with no prior release to fall back on.

Sections and checks marked v2.0.0 cover what v2 changed: the editor's Record mode and automatic zooms after a take, Apple's source picker and the permissions window on macOS, the export dialog's settings panel, and the webcam's capture resolution and frame rate. The v2 editor also reshaped controls older sections name (the top bar, the transport, the inspector); those checks were rewritten in place rather than marked.

Sections and checks marked post-2.0.0 cover what has landed on main since the v2.0.0 tag: the MCP server (in-app and stdio, recording and export included), the recordings folder and the low-disk refusal, microphone and Restart fixes, the zoom transition trails, the Glass Lens cursor and the click sound, names and terms for transcription, chat sessions kept per project, new agent tools, the Edit clip scrub, faster GIF export, Windows on ARM, the Windows clock and webcam-mode fixes, and Linux AppImage changes. On macOS, read Running the pass unattended on macOS before the first take.

How to run this

  1. Drive the real Electron app with computer-use — real OS mouse and keyboard events. For a release candidate, launch the CI-built artifact (see Running the pass on a CI-built artifact below); for a change under development, a dev build started with npm run dev.

    "Manual" here usually means an agent holding the mouse, so the tempting shortcut is not a browser shim: it is driving the real app through CDP instead. Do not. Playwright's .click(), javascript_tool-dispatched pointer events and anything else synthesised into the renderer arrive below the OS hit-test. On Windows and macOS the HUD is input-transparent until a real cursor move lifts it, so an injected click fires the DOM handler and comes back green while the path a user actually takes was never exercised at all.

    That trap is specific to the HUD and the countdown overlay — they are the only click-through windows; the editor is an ordinary one, and an injected click there does reach the handler a user would reach. The reason not to inject in the editor either is the first line of this file: this checklist covers what unit, browser and Playwright tests cannot. Drive it the way those tests already drive it and you have re-run the coverage you had, then written "passed" beside the parts nothing checked.

  2. The app is single-instance per userData path. If a leftover Electron/OpenScreen process still holds the lock, stop that process before relaunching; a second launch can exit successfully without opening a window. The lock is held by the OS and is released when the process dies, so there is nothing to delete on disk.

  3. For a dev build from a worktree, link or junction node_modules to the main checkout and provide the prebuilt native capture binaries for the platform before starting the dev build. Date those binaries against the change you came to test. Nothing rebuilds them, so an older helper runs the old code path in silence: the recording succeeds and the thing you were checking for is simply absent. See AGENTS.md for how to check one and how to rebuild it; when you cannot, test the CI-built artifact, because a dev build cannot answer a native question.

    A junctioned node_modules is only as current as the checkout it points at. Diff the two package.json dependency sets before trusting it: a dependency added after that checkout's HEAD is simply absent, and the failure does not name it — vite logs one Rollup failed to resolve import line among the build noise, the main process starts anyway, and the IPC handlers in the module that failed to bundle are never registered. Symptom seen: No handler registered for 'get-app-info' and no visible window, from a missing electron-updater. Installing the one missing package into the shared tree reconciles that tree against the older lockfile, so re-check the packages you need afterwards rather than assuming the install was additive.

  4. Grant computer-use access to the process name that actually owns the window: electron.exe or Electron.app for a dev build, and the installed name Openscreen for a packaged build or CI artifact. Do not grant access only to the installed app name when testing a dev build — it resolves to the installed executable and reports success while the dev window stays masked. This is step 4 and not step 1 for a reason: a dev build is not an installed app, so the resolver cannot find it until it is running and owns a window, and one unresolvable name voids the whole request.

    Ask for everything in ONE call — after the launches above, before the first check. request_access takes a list, and once a grant is in place the rest of the pass runs without a single further prompt: a full capture-to-export run is dozens of clicks and none of them ask again. So the only thing keeping a human at the keyboard is how many dialogs you raise and when. Raise one, before the first check, and the operator can walk away for the rest of the run; discover a fourth app you need an hour in and they cannot. That is also why this cannot move earlier — the resolver needs the app running, and one unresolvable name voids the batch. One exception, on Windows: wgc-capture.exe only exists while recording, so it is granted mid-pass, at the Hide desktop icons check. Tell the operator to stay for that second dialog. Beyond the app under test, ask for:

    • the desktop shell — the tray icon and the native save dialogs live there, and the tray is the only reliable way back to the HUD;

    • the OS settings app — needed to change display scaling, which is how DPI checks are run (see AGENTS.md; "the machine is at 100%" is not a reason to skip them), and on macOS to grant permissions;

    • on macOS the permission prompts, and on Linux the portal's source picker. Neither is an OpenScreen window, and which process owns each is to confirm on the first run: the screenshot note names a masked window's process, so ask for that name, then write it here.

      On macOS 26 the Screen Recording and Accessibility prompts are windows of universalAccessAuthWarn (measured 2026-10-14). It is not an installed app, so request_access cannot grant it: the prompt stays masked and the operator answers it. UserNotificationCenter is an installed app that request_access does resolve: granting it unmasks the system alerts it draws, an Automation consent dialog among them, so the agent can see what is blocking its clicks. That grant is computer-use access to those windows, not the Automation (Apple Events) permission itself, which stays the operator's answer to the prompt. The save panel shows nothing in screenshots either; typing the base name and pressing Return still saves.

    On Windows, an ungranted window does not just stay masked: it gets minimized. computer-use minimizes every window it was not granted whenever it brings an app forward, and leaves it so. Anything the pass checks on screen must therefore be granted, even a window the agent never clicks, such as the Hide desktop icons cover (see that check).

    Apple's source picker (macOS 15.2+) cannot be granted. It is drawn by Control Centre, which computer-use refuses, so an agent sees the picker and cannot click it: every take started through it waits for a human (rc.8, rc.12 and rc.13 runs). Do not try to grant it. Run the pass on the app's own picker instead, and keep Apple's picker for a short human slice: Running the pass unattended on macOS.

    Name them the way the Start menu does, in the system's own language. The resolver matches installed-app display names, not executables: on a French Windows the shell is Explorateur de fichiers and explorer.exe fails outright — notInstalled, with a nonsense suggestion attached — which then voids every other name in the same call. On an English install it is File Explorer. When unsure, ask rather than guess; the tool lists the installed names it knows.

    There is no way to pre-approve any of this in config: the request has to be answered live. That is upstream (claude-code#46907, closed stale), and bypassPermissions does not cover it either (#43172). Batching is the whole mitigation.

  5. Read AGENTS.md for the computer-use mechanics, screenshot permissions, tray interaction, and cleanup procedure. Read one check, perform it, observe the result, then continue; close each modal or popover with Esc before the next check.

  6. The recording HUD is protected from capture by default and is invisible in screenshots. For this session only, launch with OPENSCREEN_DISABLE_CONTENT_PROTECTION=1; this is the environment variable checked before setContentProtection(true). A packaged build gets it only when its executable is started from the shell where it is set. macOS 26 and later and Windows before 11 22H2 need no flag, and Linux has no protection to lift. Unset it before making any recording whose HUD must not appear in the video.

  7. A preview screenshot is downscaled. Settle every pixel-level question by exporting a frame and measuring the exported frame, not by judging fine edges, corners, shadows, or alignment from the preview screenshot.

  8. Keep the first real recording or imported project available for the editor sections. Log crashes, hangs, data loss, security issues, and reproducible visual failures as soon as they occur.

  9. Several v1.8.0 sections need a configured AI provider (chat editing, caption translation) or a built native compositor addon (preview, export). A dev build from a worktree needs the compositor addon installed for its platform, not only the capture binaries. When a prerequisite is missing, record the section as skipped with the reason; do not mark it passed.

    The AI camera background needs two more things that nothing else does: the ONNX Runtime shared library staged beside the addon (npm run fetch:onnxruntime, which a plain npm run dev does not run) and a recording that actually has a webcam track. Without the library the control is correctly absent — so an absent control is only a defect once you have confirmed the library is there. Check both directions before writing a verdict.

  10. Prefer a project with at least two clips from the same asset for the modifier sections. A single-clip project cannot exercise anchoring, reorder, or cross-boundary splitting at all, which is where the v1.8.0 timeline model changed.

Running the pass on a CI-built artifact

A release candidate is tested on what CI built, not on a dev build. The artifact carries one matched native payload built from the commit under test, so the stale-binary traps of a dev build do not apply to it. It comes from .github/workflows/build.yml dispatched by hand on the branch under test with release_tag left empty, which builds every platform and publishes nothing: gh workflow run build.yml -R getopenscreen/openscreen --ref <branch> -f arch=both (arch picks the macOS architectures).

Get the artifact. The run's page lists the artifacts at the bottom, under Artifacts, for anyone logged in to GitHub; each one downloads as a zip. From a terminal, find the run, then download an artifact by name. Artifacts expire after 30 days.

gh run list -R getopenscreen/openscreen --workflow build.yml --branch <branch>
gh run download <run-id> -R getopenscreen/openscreen -n openscreen-windows-x64
Artifact What it holds
openscreen-windows-x64, openscreen-windows-arm64 Openscreen.Setup.<version>-x64.exe or Openscreen.Setup.<version>-arm64.exe, the NSIS installer, with its .blockmap and the update-feed JSON
openscreen-windows-store Openscreen.Setup.<version>.appx, the Store package
openscreen-mac-arm64, openscreen-mac-x64 Openscreen-macOS-Apple-Silicon-<version>.dmg or Openscreen-macOS-Intel-<version>.dmg, with the update ZIP and its JSON
openscreen-linux Openscreen-Linux-<version> as .AppImage, .deb, .rpm and .pacman, with latest-linux.yml

<version> comes from package.json, not from the branch, so it does not tell two candidates apart. The results row carries the run id, and the version About OpenScreen reports.

Windows.

  • Run the Setup .exe. It is not code-signed, so SmartScreen's "Windows protected your PC" is expected; going past it is the machine owner's decision, not the agent's. The installer asks whether to install for the current user (%LOCALAPPDATA%\Programs\Openscreen\) or for all users (C:\Program Files\Openscreen\).

  • If the installer started the app when it finished, quit that copy (Quit OpenScreen on the HUD, or tray → Quit). It runs without the flag below, and its single-instance lock makes the next launch exit silently.

  • Launch from PowerShell, so the flag reaches the process (adjust the path for an all-users install):

    $env:OPENSCREEN_DISABLE_CONTENT_PROTECTION = "1"
    & "$env:LOCALAPPDATA\Programs\Openscreen\Openscreen.exe"
    
  • Grant Openscreen, not electron.exe.

  • openscreen-windows-store is unsigned as well, and Windows does not install an unsigned package. From a source checkout, powershell -File scripts/verify-appx-native.ps1 -Appx <file> -KeepRegistered registers it for a pass by hand, then grant the name its Start-menu entry shows. Loose registration needs Developer Mode, a machine-wide setting its owner turns on or not.

macOS.

  • Take the artifact that matches the Mac: arm64 for Apple Silicon, x64 for Intel.
  • The DMG is signed with the Developer ID, notarized and stapled only when the run has all six Apple secrets (MAC_CERTIFICATE_P12, MAC_CERTIFICATE_PASSWORD, MAC_CSC_NAME, APPLE_ID, APPLE_TEAM_ID, APPLE_APP_SPECIFIC_PASSWORD). Without them the .app is ad-hoc signed and the DMG is neither signed nor notarized; the job's Notarize DMG step then shows as skipped. On the Mac, spctl -a -vv -t install <dmg> answers accepted and source=Notarized Developer ID for a notarized DMG.
  • Gatekeeper refuses an unnotarized DMG that a browser downloaded. Clearing its quarantine attribute is a security override the Mac's owner decides on, not the agent. macOS ties permission grants to the code signature, and an ad-hoc signature changes with every build, so grants made to one candidate may not carry to the next.
  • Open the DMG and drag Openscreen into Applications.
  • On macOS 26 and later no flag is needed: launch it as a user would. On macOS 13 to 15 the flag takes a launch from Terminal, OPENSCREEN_DISABLE_CONTENT_PROTECTION=1 /Applications/Openscreen.app/Contents/MacOS/Openscreen. Started that way, macOS files the permission prompts under Terminal, not OpenScreen (2026-10-14, macOS 26.5): with OpenScreen allowed under Screen & System Audio Recording and Terminal not, the legacy launch still raised the Screen Recording prompt. Launch with open -a /Applications/Openscreen.app --env OPENSCREEN_DISABLE_CONTENT_PROTECTION=1 --env OPENSCREEN_DIAGNOSTIC=1 --stdout <log> --stderr <log> instead (add --env OPENSCREEN_MAC_SOURCE_PICKER=legacy for the unattended run): OpenScreen stays its own responsible process, the env reaches only it, and the main log lands in the file. While an Openscreen mcp stdio server runs, open says the app is already running and starts nothing; add -n.
  • Grant Openscreen, plus the picker and prompt processes from step 4.

Linux.

  • AppImage: chmod +x Openscreen-Linux-<version>.AppImage, then run it; add --no-sandbox if Chromium's sandbox cannot start. A package: install it with the distribution's tool (sudo apt install ./Openscreen-Linux-<version>.deb, sudo dnf install ./Openscreen-Linux-<version>.rpm, sudo pacman -U Openscreen-Linux-<version>.pacman), then run openscreen.
  • No flag is needed. Which name to grant is to confirm on the first run: a package installs a desktop entry, an AppImage has none until it is integrated.

Logs of a packaged build. The main process logs to its standard output, so a launch from a terminal shows it there, [content-protection] OFF for the HUD window included; whether the Windows executable prints into the PowerShell that started it is to confirm on the first run. However it was launched, the same lines are appended to main.log in the folder below, one --- launch <version> pid <pid> --- header per launch. Past 5 MiB it becomes main.old.log and starts over, so the folder never holds more than two files. Tray → Save Diagnostics writes a .zip of your choice: both log files whole, and diagnostic.json with the last 256 KiB of them (mainProcessLogs, with logFile and, if the file could not be written or read, logFileError) and the capture helper's recent output (helperOutput.windows, .mac or .linux). It leaves the open project out; a project file is in userData/projects. The renderer's console calls are stripped from packaged builds. Recordings and their .cursor.json and .session.json sidecars are in the recordings folder of %APPDATA%\Openscreen, ~/Library/Application Support/Openscreen or ~/.config/openscreen.

Where main.log is written (a diagnostic bundle's logFile names the exact path):

Platform Packaged build Dev build (npm run dev)
macOS ~/Library/Logs/Openscreen ~/Library/Logs/Electron: macOS names the folder after the app bundle, Electron.app in dev
Windows %APPDATA%\Openscreen\logs %APPDATA%\openscreen\logs
Linux ~/.config/openscreen/logs ~/.config/openscreen/logs

An instance started on another userData folder (--user-data-dir) writes to <userData>/logs instead, on every platform.

Launch and HUD

  • Start the app and confirm one launch window, the HUD, appears without a startup crash (on macOS the permissions window may open beside it: see the macOS section).
  • Confirm the launch window remains usable after the first device enumeration completes.
  • Confirm the HUD is visible when content protection is disabled for the test session.
  • Activate [data-testid="launch-tray-layout-button"] and confirm the bar switches between horizontal and vertical.
  • Confirm the chosen bar layout remains coherent when the HUD grows to show recording controls.
  • Activate [data-testid="hud-drag-handle"], drag the HUD across most of the primary display, and confirm it follows the pointer without drift.
  • Release the drag and confirm the HUD stays at the dropped position instead of jumping.
  • Activate the language button by its visible language code and confirm a menu of locale choices opens.
  • Press Esc with the language menu open and confirm it closes without changing the locale.
  • Activate the camera toggle [data-testid="launch-webcam-button"] and the microphone toggle [data-testid="launch-microphone-button"] (absent on macOS 13 and 14) and confirm each turns on with one click. On a Mac with no camera attached, confirm the webcam toggle stays off and reports that camera access is blocked or the camera is unavailable; recording remains available with the webcam off.
  • Open the gear [data-testid="launch-device-settings-button"] (Device settings) and confirm it lists the input devices with a level meter that moves when you speak, and the cameras with a live preview; pick another device, close with Close (the X), and confirm the toggle now records from it.
  • v2.0.0 — With a camera attached and no saved choice, open the gear and confirm a Camera quality row sits under the camera list and above the preview, with three radio choices, camera-quality-1080p, camera-quality-1440p and camera-quality-2160p, labelled 1080p, 1440p and 4K, and that 4K is the one checked.
  • v2.0.0 — Pick 1080p, quit and relaunch the app, reopen the gear, and confirm 1080p is still checked. recording-settings.json in userData carries "camQuality": "1080p".
  • v2.0.0 — Quit, hand-edit camQuality in that file to "720p", relaunch, and confirm the gear shows 4K checked rather than a row with nothing checked.
  • v2.0.0 — With no camera attached, or the camera disabled in the OS, confirm the gear reports "No camera found" and shows no Camera quality row.
  • Hover each HUD control with real mouse moves, on the horizontal bar and again on the vertical one, while idle and while recording. Confirm each shows exactly one tooltip that is readable and not clipped by the window (above a horizontal bar, beside a vertical one, the Record button included), that the system audio, microphone, camera, cursor and gear buttons still show theirs while recording, and that the cursor button is absent when capture falls back to the browser (macOS or Linux without their native helper). The HUD is click-through, so Playwright cannot do this: see tooltips.md.
  • Activate Hide recording bar and confirm the HUD hides without quitting the app.
  • Refocus the app from its system-tray icon and confirm the HUD returns to the foreground.
  • Activate Quit OpenScreen while idle and confirm the app exits cleanly, tray icon included.
  • Relaunch the app after closing it and confirm the single-instance behavior does not leave a duplicate HUD.

Source selection and recording

The in-app source selector below is the Windows one, and the macOS one before 15.2. macOS 15.2+ picks in Apple's picker (see the macOS section), and Linux in the portal's, after record is pressed (see the Linux section).

  • Activate [data-testid="launch-source-selector-button"] and confirm the source selector opens on its Screens and Windows tabs.
  • Select a screen or application card with data-testid="source-selector-card", activate [data-testid="source-selector-share-button"] (Select), and confirm the selector closes with the source name on the HUD.
  • With no source picked, confirm [data-testid="launch-record-button"] is enabled with the tooltip "Choose a screen or window to record", that activating it opens the source selector, and that picking a source starts the recording without a second click.
  • Confirm recording starts with a red stop state and an increasing elapsed timer, beside the pause, restart and cancel controls.
  • Confirm the configured system-audio, microphone, webcam, and cursor states remain visible, and locked, while recording.
  • Activate the recording control's pause action and confirm the timer stops advancing, then resume and confirm it advances again.
  • Activate the restart action while recording and confirm the current recording is discarded and a fresh recording begins.
  • Activate the cancel action while recording and confirm recording ends without opening an editor for the canceled take.
  • Confirm stopping opens the editor with the recorded screen asset loaded.
  • On Windows, stop once with system audio, microphone, webcam, and cursor all disabled and confirm the editor opens within a few seconds.
  • Record once with microphone only and confirm the resulting playback contains audible microphone audio.
  • Record once with system audio only and confirm the resulting playback contains audible system audio.
  • Record with microphone and system audio enabled and confirm both sources are audible and reasonably balanced.

Record mode and automatic zooms — v2.0.0

Record mode is the editor's pre-flight panel for the HUD: it edits the same settings, and Start recording hands them over. The HUD is closed while the editor is open, so every check here that ends in a take goes through that hand-over.

  • On a first run, with no saved project (otherwise the editor reopens the most recent one), open the editor and confirm the empty state leads with Record your screen, followed by Import a video and an Open project link; activate Record your screen and confirm it opens the Record tab.

  • Confirm Record mode lists Source, System audio, Microphone (absent on macOS 13 and 14), Camera, Editable cursor, Auto-zoom after recording (only while Editable cursor is On), and Hide desktop icons (macOS and Windows only). Editable cursor and Auto-zoom are both absent when capture falls back to the browser (macOS or Linux without their native helper), as the HUD's cursor button is.

  • Turn Camera on and confirm a live preview; turn Microphone on and confirm its level meter moves when you speak; switch each device from its menu and confirm the preview or the meter follows.

  • Pick a source from the Source row and confirm its name shows on the row and on the preview's badge. On Linux the row is plain text, "Your system will ask what to share", with nothing to pick.

  • Activate Cancel and confirm the editor returns to Edit mode.

  • Activate Start recording and confirm the editor window closes, the HUD comes back with the same source, audio, camera and cursor settings, and the take starts without another click (through the source picker first when no source was picked).

  • With Editable cursor and Auto-zoom after recording On, click a few places during a take, stop, and confirm the editor opens with zoom regions already on the timeline at those clicks.

  • Turn Auto-zoom after recording Off, record again, and confirm the new take opens with no zoom region; restart the app and confirm the row is still Off.

  • Turn Editable cursor Off and confirm the Auto-zoom after recording row disappears; confirm a take recorded that way opens without automatic zooms.

  • On Windows, turn Hide desktop icons On, record a screen, and confirm the recording shows the wallpaper where the icons were, and that the icons are back when the take stops. On macOS, confirm the icons are absent from the recording while they stay on the desktop.

    On Windows, grant wgc-capture.exe once the take is running, before the first click. The cover over the icons is a window of that helper, and computer-use minimizes every window it was not granted as soon as it brings an app forward. Ungranted, the icons come back at the agent's first click and stay for the rest of the take: that reads as a product defect and is not one (measured 2026-10-15 on v2.1.0-rc.1, Windows 11 26200). A user's own clicks never move the cover. The helper only exists while recording, so the step-4 batch cannot resolve it unless a take is already running.

Webcam capture quality — v2.0.0

The camera is captured at its own resolution and frame rate, and the resolution is the HUD's Camera quality setting (1080p, 1440p, 4K). Before that, nothing asked the camera for a size, so a UVC camera answered with the first format it lists, often 640x480, and the editor upscaled it. No unit test reaches a physical camera, so every check here is read off a real take. They need a physical camera, and ffprobe and ffmpeg from any install.

Where the take is. The webcam is its own file beside the screen recording: recording-<id>-webcam.mp4 from the Windows helper, recording-<id>-webcam.webm from the browser recorder (macOS, Linux, and Windows when the helper is missing). Both are in the recordings folder of userData, which the RECORDINGS_DIR: line the main process prints at startup names. Take the newest file, then:

ffprobe -v error -select_streams v:0 -show_entries stream=codec_name,width,height,avg_frame_rate,bit_rate:format=duration -of default=nw=1 <file>

A .webm is variable frame rate, so its avg_frame_rate says nothing about the camera. Count the frames instead, and divide nb_read_frames by duration:

ffprobe -v error -count_frames -select_streams v:0 -show_entries stream=width,height,nb_read_frames:format=duration -of default=nw=1 <file>

A Windows .mp4 is the opposite: written at a constant rate with the gaps padded by duplicate frames, so its frame count proves nothing either. On Media Foundation the camera's real rate is the delivered= count below.

What the Windows helper negotiated. Two places. The main process prints [native-wgc] capture started with a webcamFormat of width, height, fps and deviceName. The helper's own output is in tray → Save Diagnostics under helperOutput.windows, kept until the next take starts, so save it right after the take:

(Get-Content diag.json -Raw | ConvertFrom-Json).helperOutput.windows -split "`n" | Select-String "webcam"

The lines that matter are INFO: Native webcam format <W>x<H>@<F> (uncompressed|compressed) on Media Foundation, INFO: DirectShow webcam connected subtype <S> <W>x<H>@<F> on DirectShow, INFO: Webcam capture loop ended: delivered=<N>, and WARNING: Native webcam started but no visible frame was available before screen capture. To exercise the negotiation without the HUD, npm run test:wgc-webcam:win drives the helper with the camera named in OPENSCREEN_WGC_TEST_WEBCAM_DEVICE_NAME and a target from OPENSCREEN_WGC_TEST_WEBCAM_WIDTH, _HEIGHT and _FPS, and prints the same lines. That is a helper-level run, not a pass of the checks below.

Camera classes. A check tagged with a class needs a camera of that class. With none at hand, log it as skipped: no such camera, never as passed. An untagged check needs any one physical camera. The results row names the camera and the classes it covers.

  • [4K camera]: advertises a mode of 3840x2160 or more. List a camera's modes on Windows with ffmpeg -f dshow -list_options true -i video="<camera name>", on Linux with v4l2-ctl --list-formats-ext -d /dev/video<N>, and on macOS by asking for an impossible size, ffmpeg -f avfoundation -video_size 1x1 -i "<index>" (-list_devices true -i "" gives the index), whose error lists the modes the device supports.
  • [sub-4K camera]: every mode is below 3840x2160. Most laptop cameras and plain 720p or 1080p webcams.
  • [no-30-fps camera]: no mode at 30 fps or more at the size recorded, so 24 fps or less.
  • [30 and 60 fps camera]: offers 30 and 60 fps at the same size.
  • [DirectShow-only camera]: one Media Foundation does not enumerate, NVIDIA Broadcast for one. The helper output then carries WARNING: Requested webcam device was not found by Media Foundation; trying DirectShow.

Resolution and bit rate

  • [4K camera] Record about 20 s at each of 1080p, 1440p and 4K in turn, and probe each webcam file. The sizes are 1920x1080, 2560x1440 and 3840x2160, and bit_rate climbs with them. The encoder aims at 16, 24 and 40 Mbit/s; the BRIO measured in #875 gave about 14, 22 and 38. A bit_rate near 4 Mbit/s at every choice is the old fixed rate.
  • On Windows, confirm the size and rate in webcamFormat and in the INFO: Native webcam format line equal the file's width, height and avg_frame_rate. If the file is smaller than the choice on a camera that advertises the size, read the (uncompressed) or (compressed) tag on that line and the camera's mode list first: an uncompressed mode is preferred over a larger compressed one, whatever its size, so a smaller size can be the intended result. Write the modes and the outcome in the row instead of passing or failing it.
  • [sub-4K camera] Record at the default 4K. The take still has a camera and no "The camera could not be opened. Recording without it." toast appears. The file is the camera's best mode, for example 1920x1080 or 1280x720, and never 3840x2160 upscaled. bit_rate is that size's tier (8 Mbit/s at 720p, 16 at 1080p), not 40.
  • [sub-4K camera] Record at 1440p on a camera whose best mode is 1080p and confirm the file is 1920x1080.
  • Extract the last frame, ffmpeg -sseof -1 -i <file> -frames:v 1 last.png, and view it at 100%. It is the live picture: colours match the room with no green or magenta cast, no rows shifted or sheared, and not black. That covers the NV12 conversion, and a camera that stopped mid-take.
  • With a 4K take, the default, play the project in the preview and scrub across the camera, then export MP4 at 1080p. The preview keeps up without stalls, the export completes, and the last exported frame shows the camera.

Frame rate

  • On a camera with a 30 fps mode, record about 20 s. On Windows, avg_frame_rate is 30/1 and, on Media Foundation, delivered= in the helper output is within 10% of 30 times the file's duration. Run it on an idle machine: a 4K take under heavy competing CPU load can drop frames (174 of 240 seen with a lint pass running), so repeat once idle before logging a failure. On a .webm, nb_read_frames divided by duration is within 10% of 30.
  • [no-30-fps camera] Record about 20 s. On Media Foundation, delivered= divided by duration is about 24 for a 24 fps camera, and fps in webcamFormat and the file's avg_frame_rate match it. If they say 30 while delivered= says 24, the file is padded with duplicates and nothing else shows it: log a failure, not a pass. On DirectShow, fps in webcamFormat, the rate in the INFO: DirectShow webcam connected line and avg_frame_rate all equal the rate the graph settled on. On a .webm, nb_read_frames divided by duration is about 24.
  • [30 and 60 fps camera] Confirm the helper output reads @30 for the size recorded and the file is 30/1, not 60.

During a take

  • While a take runs, activate the gear and confirm the panel does not open. After the take, open it and confirm the choice is the one from before the take.
  • Open the gear, then start the take by a route that does not dismiss the panel first. The panel closes on any pointer press outside it, on Esc and when the HUD loses focus, so the record button closes it. If every route you try does, log skipped: panel cannot be held open by real input; LaunchWindow.test.tsx covers the guard. With the panel open in the take, activate another Camera quality choice. The checkmark does not move, the take goes on, and the camera does not stop: the last frame of the webcam file is live. The webcam file keeps the resolution chosen before the take, and camQuality in recording-settings.json is unchanged.
  • Choose 1080p in the gear, open the editor's Record mode, which has no quality choice of its own, turn Camera on, activate Start recording, and confirm the take's webcam file is 1080p.

No visible frame, on Windows with Media Foundation

The helper waits up to 3 s for the camera's first visible frame before starting the screen capture. The warning is written to the helper output only, with no toast.

  • Cover the lens completely with a privacy shutter or opaque tape, since a fingertip lets light through. Record at least 5 s, stop, save diagnostics, and confirm helperOutput.windows carries WARNING: Native webcam started but no visible frame was available before screen capture. The take is still saved, with a black camera. If a camera's own noise with the lens covered reads as a picture, note that in the row.
  • Repeat with the lens open in a dim room where the scene is still readable, and confirm the warning is absent. A camera's NV12 is studio range, where black is 16 and not 0, and an all-black frame used to read as a picture.

Backends

  • [DirectShow-only camera] On Windows, record 4K and confirm the helper output shows the Media Foundation warning above, then INFO: DirectShow webcam format <W>x<H>@<F> and INFO: DirectShow webcam connected subtype <S> <W>x<H>@<F>. The take has a camera. The file's size is the connected size. fps in webcamFormat and the file's avg_frame_rate equal the rate the graph settled on, and when it differs from 30 the helper output says INFO: DirectShow webcam negotiated <N> fps (asked for 30).
  • On the browser recorder, on macOS or Linux, or on Windows in a dev build whose electron/native/bin/win32-x64/wgc-capture.exe is renamed away for the run, record at 1080p and confirm the .webm is 1920x1080 on a camera that offers it, the size closest to the choice otherwise. Put the helper back afterwards.
  • [4K camera] On the browser recorder, record at 4K with motion in front of the lens and confirm the .webm is 3840x2160 and its bit_rate climbs well past 18 Mbit/s, the ceiling the sidecar used to inherit from the screen recording's rate. A static scene can come out lower, so note whether it was one.

Sharpness

  • Take two recordings of one scene, holding a page of small printed text at arm's length: one at 640x480, from a camera that only offers 640x480 or from a release older than #875, and one at 1080p or 4K. Add a Full Camera segment over the same moment in each project, export MP4 at 1080p, extract a frame with ffmpeg -ss <t> -i <export> -frames:v 1 frame.png, and view both at 100%. The 1080p take resolves the text and the 640x480 take, upscaled about threefold, does not. With neither reference at hand, log skipped: no 640x480 reference.

Recordings folder, disk space, microphone — post-2.0.0

  • In Record mode, confirm the last row is Recordings folder, showing the folder's name and its full path on hover; activate it and confirm the native picker "Choose a recordings folder" opens.
  • Pick another folder, record a take, and confirm the file lands there. Restart the app and confirm the row still shows that folder.
  • After the change, open a project recorded before it and confirm its take still loads (d029d6952).
  • Confirm Reset to default appears only while a custom folder is set, and that it brings takes back under userData/recordings.
  • Delete the custom folder and confirm the next take recreates it. Point it at a removable drive, unplug it, start a take, and confirm the toast "Can't reach the recordings folder …" with no take started.
  • Confirm the row refuses a change while a take is running.
  • Point the recordings folder at a volume with less than 500 MB free (a small VHD or disk image) and confirm a start from the HUD, from Record mode and from a voice-over each refuses with "Not enough disk space to start recording (only … MB free)".
  • With no audio input listed by the OS, turn the microphone on from the HUD and from Record mode and confirm each refuses with "No microphone found" and stays off (#995).
  • Save the microphone as on, unplug the input, start a take, and confirm it records without the microphone, shows "No microphone found" and shows the toggle off. On macOS, Restart that take and confirm the next one starts.
  • Double-click the microphone toggle quickly and confirm the second click takes the first back: the microphone stays off.
  • Make a Restart's next take fail to start (the reliable way: fill the recordings volume below 500 MB during the take, so the disk check refuses it) and confirm "The new recording could not start. The previous one was deleted.", with no 0-byte recording-<id>.mp4 left in the recordings folder. Run it on each platform: the fix covers macOS, Windows, Linux and the browser recorder.

Editor opens and loads the project

  • Confirm the editor opens after a successful stop, in Edit mode, with the expected project title and asset and a "Recording added to a new project" toast.
  • Confirm [data-testid="preview"] is present and its current-time value starts at the beginning of the project.
  • Confirm the loaded video is visible in the preview rather than an empty state or broken-video state.
  • Confirm the timeline contains a clip for the recorded or imported asset.
  • Activate the project name ("Rename project"), enter a new non-empty title, and confirm the title changes.
  • Confirm the dot after the project name reports "Saved" on hover after the rename: edits are saved as they land, with no save step.
  • Switch among the Media, Edit, and Record tabs and confirm each selected tab visibly changes state.
  • Return to Edit after visiting Media and Record and confirm the preview, timeline, and inspector are back and usable. Media shows the media library over a short arrange-only timeline; Record shows its settings panel and no timeline.
  • In Edit mode, activate the Chat panel button in the top bar and confirm the chat panel opens or closes without changing the project.
  • Hover the timeline toolbar, the top bar (chat panel, undo, redo), the inspector rail and the Record mode rows with real mouse moves. Confirm each icon-only control names itself in one tooltip and shows no native one beside it, that the Add buttons, Play or Pause, Undo, Redo and Send show your own key as a chip, and that no tooltip covers the next control you would reach for (the rail opens its tooltips to the left).
  • Resize the chat panel by its visible divider and confirm the preview area resizes without moving the timeline content.
  • Resize the timeline by its visible top divider and confirm the timeline height changes without a layout crash.

Transport and preview

The transport sits in the timeline header: Play (which becomes Pause) and the time readout, nothing else. The ruler seeks, and Left / Right step 1/60 s.

  • Activate Play (also Space, or the key you set in the shortcuts dialog) and confirm [data-testid="preview"] changes data-is-playing from false to true.
  • Activate Pause and confirm playback stops and the preview reports data-is-playing="false".
  • Confirm the transport time readout advances while playback is running.
  • Confirm the playhead advances with the video instead of remaining at its starting position.
  • Seek while paused and confirm the preview frame changes to the selected time.
  • Seek while playing and confirm playback continues from the new time without a visible stuck frame.
  • While paused, press Left and Right and confirm the playhead and the preview step back and forward by 1/60 s.
  • post-2.0.0 — Play past the last clip and confirm the preview holds the last clip's own last frame instead of wrapping to the first clip, and that a later seek shows the right clip (#997).
  • Play through the end of the last clip and confirm playback stops at the total time and data-is-playing returns to false.
  • With a webcam recording, confirm the webcam picture-in-picture appears aligned with the screen content.
  • Add a full-camera segment, scrub into it, and confirm the webcam grows to fullscreen then returns at the segment end.
  • Confirm the preview's webcam, cursor, background, and region effects remain synchronized while scrubbing.

Timeline navigation (pan, zoom, scrub)

  • Confirm the timeline ruler displays time labels from the project start through its duration.
  • Click a position on the ruler and confirm the playhead and preview seek to that time.
  • Drag across the ruler or timeline track and confirm the playhead follows the pointer.
  • Hold Ctrl while scrolling over the timeline and confirm the timeline zooms around the pointer position.
  • Hold Shift while scrolling over the timeline and confirm the visible time range pans without changing the project.
  • Confirm the playhead remains aligned with the ruler and clip positions after zooming and panning.
  • Drag the navigator window and confirm the main timeline follows its visible range.
  • Drag a navigator handle and confirm the visible range narrows or widens without changing clip data.
  • Confirm an empty-area click clears any selected region and closes its selection inspector.
  • Confirm the reworked ruler keeps readable labels at the narrowest and widest zoom levels rather than colliding or disappearing.
  • Confirm the playhead stays exactly on the time it reports after zooming, panning, and resizing the timeline.
  • Rename the project, save it with Ctrl/Cmd+S, and toggle the export button's availability, and confirm the top bar keeps its layout instead of reflowing on each state change.

Clip operations

  • Switch to the Media tab and confirm the project asset appears as a card with its file name, duration, and size.
  • Drag a listed media asset onto the timeline strip below and confirm a new clip appears.
  • Click an asset card, activate Add to timeline in its detail pane, and confirm a clip is appended with an "Added to the timeline" toast.
  • Click a clip and confirm it receives a selected visual state.
  • Drag a selected clip before another clip and confirm the clip order changes.
  • Double-click a clip and confirm the Edit clip dialog opens; confirm its "Edit start and end" pencil on the clip card opens the same dialog.
  • Drag the "Adjust clip start" grip in Edit clip and confirm "Trim range" and "Final duration" change.
  • Drag the "Adjust clip end" grip and confirm "Trim range" and "Final duration" change.
  • Pick a ratio preset in Edit clip and confirm the crop frame snaps to it and keeps that shape while a corner is dragged; Free unlocks it.
  • Change the crop or the grips, press Apply, and confirm the preview and the clip reflect it; confirm Cancel leaves the clip unchanged.
  • Select a clip and activate "Delete clip"; confirm only that clip is removed.
  • Select a clip, use the configured copy and paste shortcuts, and confirm a duplicate clip appears.
  • post-2.0.0 — Duplicate a clip carrying a zoom, a speed region, an annotation and a Full Camera region, and confirm the copy carries all four; imported audio is not copied, by design (#1011).
  • post-2.0.0 — In Edit clip, confirm the trim bar has a playhead at the clip's in-point and the crop preview shows that frame on opening, not black. Click and drag along the bar and confirm the preview follows, the playhead stays inside the kept range, a trim grip dragged across it pushes it along, and scrubbing alone leaves Apply disabled.
  • With two or more clips on the timeline, activate the inspector rail's Edit clip button and confirm a "Choose a clip to edit" menu lists the clips with their time ranges and opens Edit clip for the one picked. Only one clip can be selected at a time.

Regions (trim/skip, zoom, speed, annotation)

The timeline toolbar adds each kind at the current time: Add a zoom (Z), Add a trim (T), Add a speed change (S), Add text, an arrow, an image or a blur (an annotation, A), Add Full Camera (C, offered only in a project with a camera). The key shown in each tooltip is your own binding. Selecting a pill opens its settings in the inspector.

  • Drag a trim region's left edge and confirm its start time changes.
  • Drag a trim region's right edge and confirm its end time changes.
  • Scrub across a trim region and confirm the preview skips the marked interval during playback.
  • Select a trim, activate Bring this part back in its inspector, and confirm the interval is restored.
  • Activate Add a zoom at the current time and confirm a zoom region appears.
  • Select the zoom region and pick each level in the Zoom level row, then type one in the field beside it; confirm the preview scale follows. Levels deeper than the recording can take without blurring are not offered, and a typed value out of range answers "Zoom goes from 1× to …×".
  • Drag the zoom focus point in the preview and confirm the zoom follows the new focus.
  • Change the zoom's 3D camera among Off, 3D Orbit, Screen turned left and Screen turned right, and confirm the preview orientation changes; with 3D Orbit under Auto focus, move the cursor or click from one side of the recording to the other and confirm the screen turns to that side; under Manual focus, drag the focus point to one side and confirm the camera settles on that side and stays there while the cursor moves.
  • With the cursor hidden (Cursor facet, Show cursor off) and the zoom under Auto focus, confirm 3D Orbit is not offered; switch the zoom to Manual focus and confirm it is.
  • Set a zoom region's Focus mode to Auto and confirm its focus follows cursor telemetry across the whole region.
  • Use Auto-enhance → Automatic zooms and confirm it adds suggested zoom regions when cursor telemetry supports suggestions, or says why not ("No room for automatic zooms" on a take whose zooms were already placed after recording).
  • Select a zoom region and activate Delete zoom in the inspector; confirm it disappears from the lane.
  • With a zoom, a speed, a trim, an annotation and a Full Camera region on the timeline, plus an imported audio track and captions, activate Clear timeline (the eraser button at the end of the toolbar, after Add Full Camera and a divider, or after the last add button when there is no camera; both are absent while no region exists) and confirm every region disappears in one step while the clips, the audio track and the captions stay, and one Ctrl+Z brings all the regions back.
  • Activate Add a speed change at the current time and confirm a speed region appears.
  • Pick each speed in the Playback speed row (0.5×, 1×, 1.5×, 2×, 4×) and confirm the lane label and preview timing change.
  • Enter a custom speed in the Custom speed field, commit it, and confirm the value is kept, with no preset pressed.
  • Play across a speed region and confirm the preview reflects the region's speed.
  • Select a speed region and activate Delete speed region; confirm normal speed returns.
  • Activate Add text, an arrow, an image or a blur at the current time and confirm an annotation region appears.
  • Select a text annotation, replace its text, and confirm the new text appears in the preview, its box fitted to the text.
  • Change the text's size (24, 32, 48, 72 or typed), its colour, and its Background plate (None, Dark, Light); confirm each change is visible in the preview and that the text stays readable on every plate.
  • Pick each Text animation (the row labelled "Select animation") and confirm the animation runs when the playhead enters the region.
  • Switch the annotation's type to Image, upload a supported image (JPG, PNG, GIF or WebP), and confirm the image appears in the preview.
  • Switch the type to Arrow, change its direction, colour and thickness, and confirm the arrow changes.
  • Switch the type to Blur, change its blur type (Smooth, Mosaic) and shape (Rectangle, Oval), and confirm the selected area is obscured.
  • Drag a text, image or arrow annotation in the preview, over the padding too, and confirm its position persists when the playhead leaves and returns; confirm a blur annotation cannot leave the footage it hides.
  • Select an annotation and activate Delete annotation; confirm it disappears from the preview and lane.
  • Use undo and redo after adding, editing, and deleting at least one region and confirm each operation restores the prior state.

Modifiers under a trim — post-1.10.0

A trim is skipped during playback but the playhead can still be parked on it. Both halves of this are preview-only — the render still cuts.

  • Park the playhead inside a trim and confirm the preview shows the frame that is actually at that time, not the first frame of the next kept segment.
  • Draw a zoom that lies entirely inside a trim, park the playhead on it, and confirm the zoom fires in the preview.
  • Repeat with an annotation and with a Full Camera segment entirely under a trim.
  • Confirm a region that merely overlaps a trim without being contained by it behaves as an ordinary region — this is not the same case, and testing it instead is the easy way to record a false failure. Check the saved spans rather than the pills, which look identical at normal zoom.
  • Confirm a zoom under a trim plays dry: full strength on its own span, with no ease-in reaching the kept frames beside the cut.
  • Confirm speed regions under a trim are not emitted — a still frame has no rate to show.
  • Export the range and confirm the trimmed span is absent from the output: the modifiers showing in the preview must not resurrect the cut frames.
  • post-2.0.0 — Put a trim inside a Full Camera segment and confirm the webcam stays full across the cut, in the preview and in an export, instead of shrinking back to its corner around it.

Zoom transitions and region editing — post-2.0.0

The zoom pill now shows only the hold; its move-in and move-out are drawn beside it as faint trails, on the speed-aware clock (#1028).

  • Add a zoom and confirm a trail before and after its pill, with the tooltips "Zoom-in: the view moves in until the zoom starts" and "Zoom-out: the view moves back out after the zoom ends".
  • Drag the zoom, then a trim and a speed pill near it, and confirm the trails follow live.
  • Put a trim across a trail and confirm that part is hatched, with "A trim removes … s of the zoom-in. The export jumps at the cut."; do the same with a clip junction and confirm "A clip junction removes …". Confirm the end of the timeline does not count as a junction.
  • Place two zooms at most 1.5 s apart and confirm one pan trail joins them ("Pan: the view moves straight to the next zoom") and the preview pans in about a second; put a trim between them and confirm the chain breaks into an out and an in.
  • Make two look-alike regions touch on one clip so they show as one pill, click one half, change it, and confirm only that region changes and the pill splits (#1017).
  • Drag a zoom, a speed, an annotation and a Full Camera pill into a neighbour and confirm each stops flush and keeps its length; resize one into a neighbour and confirm it stops there (#1008).

Modifiers are anchored to clips — v1.8.0

Zoom, speed, annotation, and full-camera regions are stored against a clip in that clip's own source time, not at an absolute ruler position. See timeline-model.md. These checks exist because the failure mode is silent: the pill stays where it was drawn while the effect fires somewhere else.

  • Draw a zoom wholly inside one clip, reorder that clip to another position, and confirm the zoom travels with the clip and keeps its length.
  • Confirm the moved zoom still fires over the same picture content, not at the ruler position it originally occupied.
  • Draw a region across a boundary between two clips, move one of those clips away, and confirm the region splits into one pill per clip instead of remaining one pill at the old position.
  • Put the two clips back side by side and confirm the fragments render as a single pill again.
  • Confirm two regions of the same kind with identical properties that touch display as one pill.
  • Change one of the two merged regions and confirm the pill separates into two.
  • Drag a zoom pill into a neighbouring zoom with a different level and confirm it clamps at the neighbour's edge and the neighbour does not move.
  • Confirm the same repel behaviour for two speed regions with different speeds.
  • Add a trim inside a clip that a zoom already covers and confirm the covered part is hidden without shifting any later region on the ruler.
  • Confirm the ruler still shows the trimmed span occupying its place while playback skips it.
  • Delete a clip and confirm modifiers anchored only to that clip disappear while modifiers on other clips are untouched.
  • Duplicate a clip and confirm its modifiers are duplicated with the copy.
  • Change a clip's in and out points in the Edit Clip dialog and confirm anchored modifiers clamp to the new range rather than drifting past it.
  • Select a zoom, copy it with the configured copy shortcut, move the playhead, and paste; confirm the "Region copied" and "Region pasted" toasts and a new zoom at the playhead with the copied length, level, 3D camera and focus, anchored to the clip it lands on.
  • Repeat the copy and paste for a speed region, a text annotation, and a trim (a trim pastes as a new trim of the copied length).
  • Copy with nothing selected, and paste before anything was copied, and confirm neither changes the project.
  • Save, reopen the project, and confirm every modifier is still on the same clip content after the reorder performed above.
  • Zoom and pan the timeline and confirm each pill's span still matches the time at which its effect fires in the preview.
  • Export a short range that covers a reordered clip and a trim, and confirm the exported frames agree with the preview about where each modifier fires.

Imported audio and voice-over — post-1.10.0

Audio is a clip-anchored region on its own lanes, not a free-floating pill: a track travels with the clip it was placed over through reorder, trim and delete. One user-visible track can be stored as several fragments (one per clip it covers) that share a trackId; the lane draws them as one pill, and each fragment's offsetMs is advanced so a bed spanning a cut does not restart at the boundary. Most defects in this area are that seam, so prefer a project with at least two clips and place a track across the junction.

Two things here cannot be judged from the preview alone and need an exported file: the mix and the fades. Export and probe.

  • Import an audio file and confirm it appears as a pill on an audio lane with a waveform, and that a clip lane above is unchanged.
  • Confirm the pill's label names the imported file rather than the project or the primary asset.
  • Play the project and confirm the imported audio is heard over the recording rather than replacing it.
  • Open the track's inspector and confirm output level (gain), fade-in, fade-out, loop and mute controls are present.
  • v2.0.0 — Confirm a newly imported music track starts at -18 dB with 1 s fades, and that its inspector's Reset audio returns to those values. Play it under speech and confirm the music dips while the voice speaks, in the preview and in the export (about 10 dB in the file).
  • Change the gain and confirm the change is audible in the preview; export and confirm the same level in the file.
  • Mute the track, confirm silence in the preview, and confirm the exported file has that track absent rather than merely quiet.
  • Set a fade-in and a fade-out, then export and inspect the waveform of the result — the ramps must be at the track's own edges. A fade timed to the end of the programme instead of the end of the track is the specific defect here.
  • Place a track that runs past the end of the programme and export: confirm it is cut off at the last frame and that its fade-out is not pulled forward to the truncation point.
  • Enable loop on a track shorter than its span and confirm it repeats to fill the span, in the preview and in the export.
  • Drag a track across a clip boundary and confirm it plays continuously across the junction, with no restart or click at the seam.
  • Reorder the clips underneath a track and confirm the track follows the clip it was placed over rather than staying at its ruler position.
  • Trim away a stretch the track sits over and confirm the track is silent through the cut while its own clock keeps running — the words after the cut must land on the picture they belong to.
  • Put a speed region under a track and confirm the track is not time-stretched with the picture, and that it still starts where it did.
  • Delete the clip a track is anchored to and confirm the track goes with it rather than being left pointing at nothing.
  • Drag a pill's edge and confirm the ghost of the rest of the file is drawn around it, and that the edge stops at the file's bounds instead of running past them.
  • While dragging an edge, confirm the readout shows in → out / length and that the numbers move with the drag.
  • Hold Alt and drag a pill's body: confirm the media slides inside a span that does not move, and that the rate follows the file rather than the timeline — a four-minute bed must be traversable without dragging four minutes of ruler.
  • Focus an audio pill and press Space: confirm it selects the pill without also toggling playback.
  • Record a voice-over against the timeline and confirm the timeline's own audio is silenced for the take, then audible again afterwards.
  • Confirm the recorded take lands on the voiceover lane at the playhead and plays back.
  • Restart the app, reopen the project, and confirm the imported audio still loads. Approval to read a file outside the recordings directory is granted by the picker for the session and by the project on load; if that second grant regressed, the waveform is empty and the track is silent only after a restart, which is why this check has to follow one.
  • Ask the assistant to add a music bed and confirm it places one on the correct lane; ask it to remove it and confirm the pill and its asset both go.

Transcript and captions

The transcript is the inspector's Transcript facet. Captions have no facet of their own: the facet's Captions button opens their settings in place of the transcript.

  • With no transcript, confirm the Transcript facet offers Transcribe now instead of showing an empty editor.
  • Start transcription for the loaded asset and confirm a visible in-progress state appears.
  • Confirm a completed transcription displays words in timeline clip order.
  • Click a transcript word and confirm the playhead seeks to that word's start.
  • Play the project and confirm the current word receives the cue highlight as playback advances.
  • Place the caret in the transcript and press Backspace or Delete; confirm the affected word becomes marked as skipped rather than disappearing from the transcript.
  • Hover a skipped word and activate its restore control (Restore "<word>"); confirm the word is kept again.
  • Activate Captions in the Transcript facet and confirm the caption settings open in place of the transcript.
  • Toggle Show captions and confirm captions appear or disappear in the preview; with no transcript, confirm the toggle is unavailable.
  • Pick each Style (Classic, Bold, Minimal, Light), then under Customize change the font, Bold, size, text colour, background plate and its opacity; confirm each committed change is visible.
  • Pick a language beside Translate, run it with a configured provider, and confirm translated captions appear, with a Display row to switch languages.
  • Switch Display back to "Original (transcript)" and confirm the source transcript returns.

Local transcription and captions — v1.8.0

  • Confirm the transcript pane states that transcription runs locally and that no upload occurs when it is started.
  • With the Whisper helper binary absent, activate the transcribe action and confirm the UI reports why nothing happened, and that the main-process log carries exactly one matching [stt] line. The failure now reaches a toast (transcriptionStore.ts) and the log (whisperServer.ts), but the sentence shown is one the app writes for itself — whisper-stt-server binary not found; build it via scripts/build-whisper-stt.sh, produced before any helper process starts — so it points a packaged-build user at a script they do not have. Helper stderr is a separate source, and only ever for a helper that did start. Verify against a build whose helper was deliberately not packaged, not only against a working one.
  • Run transcription in the packaged build and confirm the model is fetched or reused without an error about a missing cache directory.
  • Confirm a second transcription reuses the cached model instead of downloading it again.
  • Confirm the completed transcript reports the detected language on the asset's card in the Media tab.
  • Choose an explicit language under Regenerate as on that card, regenerate, and confirm the new transcript replaces the old one with its own word timings.
  • Confirm word timings are monotonic: click several words in order and confirm each seek lands later than the previous one.
  • Confirm silent stretches appear as a silence span with its duration rather than as missing text.
  • Activate a silence span's trim control and confirm a trim appears on the timeline covering that interval.
  • Restore that silence from the transcript and confirm the trim is removed.
  • Confirm transcription is unavailable with a clear message rather than a crash when the app runs outside Electron.
  • Translate captions, then delete the translation, and confirm the original transcript text and timings are unchanged.
  • Confirm a project carrying caption annotations from the old feature reports them and offers to remove them.
  • Play across a zoom region with captions on and confirm the captions stay in the frame instead of scaling and drifting with the zoom.
  • Export that range and confirm the exported frames show the same caption placement as the preview.

Caption placement — post-1.10.0

Captions are placed by an anchor and a margin, not by an invisible band: anchorV (top/bottom) with insetY, and anchorH (left/center/right) with insetX. The margin is reserved on the anchored side and applies to the plate, not the text, so it is measurable in an exported frame. Projects from before this change migrate their insetX.

  • Open the caption settings (Transcript facet → Captions), enable captions, and seek to a moment with speech; confirm a caption renders with its plate.
  • Confirm Position is a grid of six anchors, the camera position's minus its middle row: Top left, Top, Top right, Bottom left, Bottom, Bottom right.
  • Choose a top anchor and confirm the caption moves to the top of the frame and the slider relabels to "Distance from top".
  • Choose a left anchor and confirm the caption band moves to the left edge with its margin, and that a second slider reads "Distance from left"; choose a centre anchor and confirm that slider is absent.
  • Drag the distance slider to each extreme and confirm the caption reaches the true frame edge rather than stopping short at an invisible band boundary.
  • Pick a Style, change the size under Customize, and confirm no style shows as selected any more; pick the style again and confirm its look returns.
  • Export a frame and measure the plate's edge against the inset: with insetX: 10 on a 1920-wide output, the plate's left edge is at x=192. Measure the exported frame, not the preview screenshot.
  • Confirm the plate's margin is reserved on the anchored side — a right-anchored caption keeps its margin on the right as the text grows.
  • Open a project saved before this change and confirm its captions land where they did, with insetX migrated rather than reset.
  • Play across a zoom with captions on and confirm the captions do not scale or drift with the zoom, in the preview and in the export.

Transcript word editing and insertion — post-1.10.0

Three gestures on the word stream: double-click corrects a word, Backspace cuts it from the film, and typing between two words adds one. The third is DEV-ONLY and must be absent from a packaged build — see the gate check at the end of this section, and run it against a packaged build, not a dev one.

An added word is a clip, not a pause: the clip it lands in is cut in two and a generated clip goes between the halves, on its own asset with its own file. Until there is TTS that file is a test pattern over noise — deliberately unmistakable, because a held frame is indistinguishable from a decoder stuck at the end of a clip. So "the picture froze" is not a pass here: the mire must be visible, and at the insertion point rather than at the end of the clip.

  • Double-click a transcribed word, type a correction, commit it, and confirm the word changes in the pane and in the captions.
  • Confirm the correction survives re-transcribing the same asset, with the same timings.
  • Place the caret between two words, type, and confirm an amber clip appears on the timeline at that moment — not at the end of the clip.
  • Play across it and confirm the mire is on screen for the inserted stretch and the noise is audible, then the recording resumes exactly where it stopped.
  • Confirm the words after the insertion still line up with the picture: scrub to a later word and confirm the playhead lands on it, and that its caption appears at that moment and not earlier.
  • Confirm the inserted text gets a caption line of its own, over its own media, rather than being glued to the line before it.
  • Double-click the inserted word and make the text longer: confirm the amber clip grows and the film grows with it.
  • Make the text shorter and confirm the clip shrinks back — both directions, not only the one that is easy to see.
  • Delete the inserted word from the transcript and confirm the clip is one clip again, at its original length, with the recording continuous across the join.
  • Insert again, then drag the amber clip elsewhere on the timeline and confirm the two halves rejoin behind it, exactly as deleting it does.
  • Zoom out until the amber clip is a few pixels wide, select it, and confirm its edit and delete controls appear beside the clip rather than being clipped away inside it.
  • Confirm the amber clip behaves as an ordinary clip: it can be selected, moved, cropped and deleted from the timeline, and deleting it there removes the word from the transcript too.
  • Export a project containing an insertion and confirm the mire and its noise are in the exported file at the right moment and for the right duration.
  • On the voice-over lane, insert a word into a take: confirm the take splits, the generated audio plays between the halves, and the take's later content moves later by that much.
  • Confirm that insertion did not lengthen the film — the clips decide the length, and a take pushed past the last frame is clamped at export.
  • Insert a word into the film while a take is laid over it, and confirm the take's own span is unchanged: it keeps talking against a picture that has slid.
  • Ask the assistant to correct a misheard name and confirm it does; ask it to rewrite an inserted word and confirm it refuses with a reason rather than resizing generated media.
  • Against a packaged build: confirm typing between two words does nothing, that double-clicking an amber word does not open an editor, and that the transcript hint does not advertise adding words. Correcting an ordinary word must still work — that one ships.

Names and terms, transcript undo — post-2.0.0

  • In the Media tab, select a source and confirm its transcript card has a Names and terms field above Regenerate as, empty by default.
  • Type a rare name the take says, click Regenerate (↻) straight away without leaving the field, and confirm the new transcript spells it right, keeps its opening words and is not in Title Case.
  • Undo once and confirm the field gets its previous text back; switch projects and confirm the field is per project.
  • With focus in the transcript, cut a word with Backspace, press Ctrl/Cmd+Z, and confirm the cut is undone (#1012); inside a word being edited, confirm Ctrl/Cmd+Z still undoes typing.

AI chat and providers — requires a configured provider

  • In Edit mode, open the chat panel with the top bar's Chat panel button and confirm the chat surface appears.
  • Confirm the chat header shows controls for AI settings, history, and a new conversation.
  • Send a short request and confirm the user message appears in the conversation.
  • Confirm the provider returns an assistant response without an unhandled error.
  • Open the model picker and confirm the active model is visibly selected.
  • Change the reasoning effort when the configured provider supports it and confirm the chosen value remains selected.
  • Run an edit request that creates a supported timeline change and confirm the applied operation is visible in the conversation.
  • Use the conversation rewind control by its aria-label and confirm the rewind confirmation surface appears.
  • Confirm a rejected or canceled rewind leaves the timeline unchanged.
  • Open AI settings and confirm the provider list, connection status, and configuration form load.
  • For an API-key provider, enter a key and confirm the provider becomes connected without displaying the raw key afterward.
  • For a device-flow provider, confirm the challenge panel shows a user code and an Open login page action.
  • Open conversation history and confirm the current conversation is listed.
  • Start a new conversation, switch back to the prior one, and confirm each conversation retains its own messages.
  • Rename a conversation with its visible rename control and confirm the new title appears.
  • Delete a conversation with its visible delete control and confirmation prompt, then confirm it no longer appears.
  • post-2.0.0 — Hold two conversations in a project, quit the app right after a reply lands, relaunch, and confirm both conversations and that last reply are still in Conversation history.
  • post-2.0.0 — Switch to another project and confirm the history lists only its conversations; delete a project and confirm its <projectId>.chat.json beside the project file goes with it.

Chat-driven editing — v1.8.0, requires a configured provider

The agent may only call the fixed tool set in ai-agent.md; it never writes the document freehand. These checks are about the edit actually landing on the timeline, the turn being one undo unit, and a failed turn leaving the project intact.

  • With no provider connected, open the chat and confirm the "bring your own AI" welcome view appears with the composer disabled instead of an error.
  • Connect a provider and confirm the same panel becomes a usable conversation without restarting the app.
  • Ask the agent to cut the silences and confirm the result appears as trim regions on the timeline rather than as rewritten clips.
  • Confirm the seekable duration after that edit still reaches the full recording, so the trims are reversible.
  • Ask for a zoom on a described moment and confirm a zoom pill appears at approximately the requested time and the preview scales there.
  • Ask for a speed change over a described range and confirm a speed region appears with the requested factor.
  • Ask for a text annotation and confirm it appears in the preview with the requested text.
  • Ask for a full-camera segment and confirm the region appears and the camera fills the frame while it plays.
  • Confirm each applied operation is summarized in the conversation and that the number of summarized operations matches what the timeline gained.
  • Ask the agent to remove one of the modifiers it created and confirm that modifier alone disappears.
  • Ask the agent to restore the full timeline and confirm the trims it added are gone.
  • Confirm modifiers created by the agent are anchored like hand-drawn ones: reorder a clip and confirm they travel with it.
  • After a turn that applied several operations, undo once and confirm the whole turn reverts as a single unit rather than one tool call at a time.
  • Redo and confirm the whole turn returns.
  • Ask for something outside the tool set and confirm the agent explains rather than silently doing nothing or leaving an invalid document.
  • Send a request while the project has no asset and confirm a clear response instead of an unhandled error.
  • Use the rewind control on an earlier user message, confirm in the dialog, and confirm the timeline, the conversation tail, and the later checkpoints all roll back together.
  • Cancel a rewind at the confirmation dialog and confirm both the timeline and the conversation are untouched.
  • Confirm the context badge shows a percentage and that its tooltip reports used and budget tokens.
  • Activate the compact button (Summarize earlier messages to use less context) on a conversation with enough history and confirm an earlier-context summary message appears and the percentage drops.
  • Activate the compact button on a short conversation and confirm the "not enough history" message rather than a failure.
  • Confirm a compaction failure leaves the conversation history unchanged.
  • Use the copy control on an assistant message and confirm the message text reaches the clipboard.
  • Open the timeline toolbar's Auto-enhance menu, choose Smart cuts (it needs a transcript), and confirm the chat panel opens with the prompt prefilled and sent through the normal send path.
  • Confirm the edit produced by that auto-enhance request can be rewound like any other turn.
  • Choose Smart cuts with no provider connected and confirm the setup view appears instead of a failed send.
  • Choose Automatic zooms and confirm it adds zooms without involving the provider.
  • post-2.0.0 — Ask the agent to remove the filler words, then to cut a quoted phrase, and confirm the words are struck in the transcript and that consecutive words became one trim, with no sliver kept between them.
  • post-2.0.0 — Ask for a zoom "with the screen turned left", then a 3D orbit, and confirm the zoom's 3D camera row shows Screen turned left, then 3D Orbit.
  • post-2.0.0 — Ask for a vertical 9:16 format and confirm the aspect ratio changes with text annotations kept in place; ask for a title and confirm its box is fitted to the text, styled as the inspector would.
  • post-2.0.0 — Ask to stretch a region into its neighbour and confirm the agent reports the refusal, naming the neighbour, instead of landing a shorter region.
  • Restart the app, reopen the project, and confirm its conversations come back with their titles and messages while the provider configuration persists; rewind controls on messages sent before the restart are hidden, which is a known gap, not a defect to file.
  • Delete a project from the open-project dialog and confirm its <projectId>.chat.json is gone from userData/projects.
  • Confirm the provider API key is never displayed in the settings form after it is saved.

MCP server — post-2.0.0

Two servers. The in-app one is HTTP, in AI settings → MCP server, off by default. The stdio one is the installed executable started by the agent itself: claude mcp add openscreen -- "<exe>" mcp, with <exe> %LOCALAPPDATA%\Programs\Openscreen\Openscreen.exe on Windows or /Applications/Openscreen.app/Contents/MacOS/Openscreen on macOS. Both serve the same tools, recording, export and frames included (#1174). Drive these checks from a real MCP client against the installed build, never with a hand-written JSON request.

  • In AI settings, confirm Enable MCP server and Allow MCP clients to edit your projects, record and export are off by default; enable the server and confirm the status reads Running with a Port, an Access token (Show, Copy, Regenerate) and the Claude Code and Codex snippets. Change the port, Apply, and confirm a client reconnects on it.
  • With edits not allowed, confirm an edit tool, startRecording and exportVideo are refused, naming MCP edits; allow them and confirm the edit lands in the open editor and undoes with Ctrl/Cmd+Z.
  • From a client of the in-app server, startRecording on a window: the editor closes and the recording bar records, its countdown first, with no second recorder process. Click in the window, stopRecording: the editor opens on the take, laid out and zoomed as after a take started by hand, and the returned projectId is that project. An edit to it lands live and undoes with Ctrl/Cmd+Z; exportVideo to an absolute path and getFrames on it work. Confirm the recording bar's microphone, camera, system audio and cursor mode are back to what they were before the take, and that startRecording is refused while the editor has unsaved changes.
  • Confirm "<exe>" help lists mcp, register the stdio server in an MCP client, and confirm listSources lists displays, windows, microphones and cameras.
  • Run startRecording on a window, click inside it a few times, stopRecording, and confirm the returned project is in the editor's project list with one clip and the zooms its autoZooms count announces. Clicks close in time (3 s or less) and on screen share one zoom, and a click in the first seconds or the stopping click gets none, so there can be fewer zooms than clicks. On macOS 15.2+, confirm Apple's picker never appears; on Linux, confirm the portal asks.
  • Edit that project while it is closed, restoreCheckpoint with the returned checkpoint, and confirm the project is back to the raw take. Confirm an edit to a closed project with no checkpoint is refused.
  • Open the project in the editor, with the in-app MCP server turned off, then send an edit to it through the stdio server, a process apart from the app (#1174): listProjects reports the project open: true, the edit shows in the editor at once, Ctrl+Z takes it back in one step, and a later change in the editor (its autosave) keeps it. restoreCheckpoint with the checkpoint from stopRecording also lands in the editor as one undo step. Quit OpenScreen mid-session: the next stdio edit to that project, now closed, asks for a checkpoint (the app's left with it), and after createCheckpoint lands in its file. Relaunch OpenScreen and open the project: the next stdio edit goes through the editor again, with no restart of the MCP client.
  • exportVideo, then getFrames at a few ruler seconds after a trim, and confirm the frames show the moments asked for; edit the project again and confirm getFrames warns that the export is stale.
  • On each server, startRecording with camera: true (then a camera's name from listSources) and cursor: "system": the take opens with the webcam as a picture-in-picture, the system pointer is in the video and no zoom was planned. exportVideo at fps: 30 gives a 30 fps MP4; to a .gif with gifSize: "small" and gifLoop: false a GIF 480 pixels high that plays once; gifLoop on an .mp4 path is refused.
  • Drive a take from a script instead of the mouse, hand its action log to setPointerTrack, and confirm the export draws a cursor and zooms on the logged clicks.

Native compositor preview and export — v1.8.0

  • Confirm [data-testid="native-compositor-mount"] shows a live composited preview rather than an empty surface.
  • Confirm [data-testid="native-compositor-error"] is absent during a normal run.
  • Scrub back and forth across a clip boundary several times and confirm the preview keeps up without a stall on each crossing.
  • Seek to the very end of the project and confirm the last frame is shown instead of a blank or stuck frame.
  • Confirm no loading overlay remains on top of an already valid preview frame.
  • On a machine with no compatible GPU, confirm [data-testid="native-compositor-cpu-notice"] appears, the export dialog shows its CPU warning, and the export still completes.
  • Export the same project on macOS and on Windows and compare frames at identical timestamps for background, blur, shadow, roundness, padding, cursor, and text.
  • On macOS, export an MP4 from a project with audio and confirm the output has audio.
  • On macOS, export a frame containing a text annotation and confirm the text is upright, centred in its box, and that its background plate fits the text.
  • On macOS, export a frame containing a blur annotation and confirm the area is actually obscured.
  • On macOS, export a range with the cursor visible and confirm the cursor and its trail are rendered.
  • On macOS, export frames with each 3D camera preset and confirm the tilt matches the Windows render.
  • On macOS, export a frame with Blur background above 0% and confirm it matches the Windows render.
  • Export a range containing a zoom with an annotation and captions on screen, and confirm neither follows the zoom in the exported frames.
  • Confirm the packaged macOS app refuses to start or reports clearly when the compositor addon is missing, rather than failing at first render.

Export

The dialog is one settings panel: Format (MP4 / GIF), Quality (720p, 1080p or Source), Frame rate (24, 30 or 60), and for a GIF its size and loop controls. Every MP4 is H.264.

  • Confirm the top bar's Export button is disabled when the project has no asset.
  • v2.0.0 — With a loaded project, activate Export and confirm the dialog opens on one settings panel: Format (MP4 / GIF), Quality, Frame rate, and no idle hint plate where the progress block will appear.
  • Confirm the MP4 quality choices are 720p, 1080p, and Source, and that each one updates the displayed output dimensions.
  • Select 24, 30, and 60 FPS and confirm the selected frame rate remains visible.
  • Confirm there is no codec choice anywhere in the dialog: every MP4 export is H.264.
  • Select GIF and confirm GIF frame-rate (15, 20, 25, 30 FPS), size (Small, Medium, Large, Original), and Loop GIF controls appear.
  • Change GIF frame rate and size, toggle looping, and confirm the summary reflects the choices.
  • Start an MP4 export with Export MP4 and confirm the native rendering progress reports advancing frames or percentage.
  • During MP4 rendering, press Cancel and confirm it waits for native cleanup, returns to the same export options, leaves no partial MP4 (no .openscreen-mp4-*.partial.mp4 beside the destination), and preserves an existing destination; retry and confirm a complete MP4 is saved.
  • Confirm the export dialog reports Saved to with the output path after MP4 completes, and that Show in folder opens it.
  • v2.0.0 — Confirm the one-time star prompt under Saved to, when it appears, goes away on either answer and does not come back on the next export.
  • v2.0.0 — Export a take with speech at the Audio facet's default output level and measure it (ffmpeg -i <file> -af ebur128=peak=true -f null -, which prints both): integrated loudness about -16 LUFS (a very quiet voice gets at most +12 dB), true peak at most -1 dBTP. The limiter holds the true peak at -2 dBTP before the AAC encoder (256 kb/s), which puts a limited voice back up by up to about 0.7 dB, so a file reading -1.3 dBTP is in spec. A voice clipped at the source is the one exception on loudness: its crests come every pitch period, so the ceiling holds it near -17 LUFS whatever the gain (#1157), and only its true peak is checked.
  • Open the exported MP4 outside the app and confirm it plays through the expected duration with audio when the source has audio.
  • Start a GIF export and confirm frame rendering and file writing complete without an unhandled error.
  • During GIF rendering, press Cancel and confirm it waits for native cleanup, returns to the same export options, leaves no partial GIF, and preserves an existing destination; retry and confirm a complete GIF is saved.
  • Open the exported GIF outside the app and confirm it contains the expected motion and loop behavior.
  • post-2.0.0 — Time a GIF export of a short clip against v2.0.0 on the same machine and confirm it is several times faster on a multi-core CPU (about 9× on 8 cores, #952), and that Cancel still stops it mid-render.
  • Export a GIF from colour-rich footage long enough to exhaust the encoder's first code widths, and confirm no frame degrades into corrupted stripes or shifted colours partway through.
  • Open that GIF in a second viewer and confirm both decoders agree, since a code-width defect can decode differently per viewer.
  • Export a project containing audio, a trim, a speed region, a zoom, an annotation, captions, and webcam layout changes when available.
  • Compare that exported result with the preview for timing, skipped intervals, audio, webcam, captions, and effects.
  • For every pixel-level comparison, export a frame and measure it with an image tool rather than relying on a preview screenshot.

Export progress and speed-region audio — post-1.10.0

The percentage is computed in the renderer against a predicted frame total, and speed regions change that total. The audio stretch now runs through libavfilter atempo rather than WSOLA; the chain is capped at eight stages and its priming loss is compensated, so the stretched span must land on its target without a silence hole.

  • Export a project containing a speed region and confirm the progress bar reaches 100% rather than stalling at a fraction (a 1.25× region used to peg it at exactly 80%).
  • Confirm the bar keeps moving during the audio phase instead of freezing once frame rendering ends.
  • Confirm the reported frame total accounts for the speed region — a region at N× emits its span's frames divided by N.
  • Probe the finished file and confirm the audio and video durations agree to within one frame; a stretch that fell short would show as an audio track measurably shorter than the video. Record the export's frame rate alongside the two durations — "within one frame" is a claim about 1/fps, and without the rate written down it cannot be checked (at 30 fps the tolerance is 33 ms; at 60 fps it is half that, and the same numbers would fail).
  • Record something that is deliberately noisy across each speed boundary — count out loud, or play a continuous tone — before running silencedetect over the exported audio. A screen recording with only ambient mic is mostly silence, so the detector reports silence at the boundary whether or not atempo left a hole, and the check passes or fails for reasons that have nothing to do with the stretch.
  • With that audio, confirm no silence block coincides with a speed region's boundary. If one does, run the same silencedetect over the source and map the boundary back through the trims before recording a failure: silence that was already there is not a defect the stretch introduced.
  • Confirm the exported duration matches the source minus the trimmed spans, adjusted for each speed region. Compute it in frames, not seconds, and state the arithmetic: source frames − trimmed frames − (speed span × (1 − 1/N)) at the export's rate. A discrepancy of a handful of frames is not "close enough" — it is either explained or it is an open question, and the results log has to say which. The count is exact, not approximate: with P the programme seconds (each kept span divided by its speed, summed over every clip), the export emits ⌈(P − 0.001) × fps⌉ frames, whatever the number of trims, regions and clips. Three things make a correct export look off:
    • The clip end is the screen stream's real length, not the asset's durationSec, which can be the audio track's. Read it with ffprobe -show_entries stream=codec_type,duration on the source. A recording whose video ends 25 ms before its audio exports a frame fewer than the declared length predicts, and the progress total (computed from the declared length) is one frame high.
    • Count packets, not decoded frames: ffprobe -count_packets -select_streams v:0 -show_entries stream=nb_read_packets. -count_frames read one frame fewer than the file holds on a 60 fps export.
    • The video stream's duration is one frame short by the muxer's convention (it is the last frame's timestamp); the audio track lasts exactly frames ÷ fps. A 2451-frame 30 fps export reads 81.667 s video against 81.700 s audio, and that is in sync.
  • Cancel an export mid-render and confirm no audio decode threads outlive the attempt.

Settings, shortcuts, themes, i18n

The inspector's rail holds five facets: Composition (background, format, frame, motion), Camera layout, Audio, Cursor (only while a recording on the timeline has cursor data), and Transcript, whose Captions button opens the caption settings.

  • Open OpenScreen menu → Keyboard Shortcuts, change one shortcut, save it, use the new key in the editor, and confirm it triggers the configured action.
  • Confirm Ctrl/Cmd+S saves the current project with a "Project saved" toast.
  • Confirm Ctrl/Cmd+O opens the project dialog.
  • Open the Composition facet and switch its Background among the Image, Gradient, and Color tabs.
  • Select a built-in wallpaper and confirm the preview background changes.
  • Choose a color swatch, or a hex value in the Color tab's Custom row, and confirm the background changes.
  • Choose a gradient preset, then a one-colour gradient from the Gradient tab's Custom row, and confirm the preview background changes.
  • With an image or gradient wallpaper, pick each Animation (None, Drift, Aurora, Waves) and confirm the background moves in the preview; confirm the row is absent under a solid colour.
  • Drag Blur background, Padding, Roundness and Motion blur, and pick each Shadow (None, Light, Medium, Strong); confirm each changes the preview. Roundness is hidden at 0% padding unless a frame is on.
  • Pick each frame Style (Window, Laptop, Phone, Screen) and each Theme (Light, Dark), and confirm the preview draws it around the recording; export and confirm the frame is in the file.
  • With a zoom on a 3D camera in the project, confirm Depth of field appears under Motion (and is absent without one); turn it on and confirm the tilted zoom changes in the preview and in an export.
  • Open the Camera layout facet and choose each Preset (Picture in picture, Dual frame, Vertical stack, No webcam); confirm the preview arrangement changes.
  • Change Mirror webcam, Shrink on zoom, Camera shape (Square, Original) and its roundness, Webcam size, and each Position; confirm each change is visible.
  • In Webcam crop, drag the frame to move the camera's framing and a corner to zoom it; confirm the preview follows.
  • Open the Audio facet, change Output level, and confirm the preview and an export follow it; Reset audio brings it back.
  • Open the Cursor facet and toggle Show cursor and Auto-hide when inactive; confirm the preview changes.
  • Change Size, Smoothing and Motion blur, pick each Click bounce (None, Light, Strong), and toggle Click impact; confirm each committed value remains visible.
  • Pick each Cursor style and toggle 3D cursor; confirm the preview cursor changes. Under Cursor types, switch a type off and confirm that type is drawn as the arrow.
  • Record a take with Editable cursor Off (the system cursor), open it in the editor, and confirm the rail has no Cursor facet; confirm a take recorded with it On shows the facet, and an imported video does not.
  • Open OpenScreen menu → Switch to light theme (or dark) and confirm the editor switches between dark and light themes.
  • Open OpenScreen menu → Change language, choose a non-English locale, and confirm visible UI strings change.
  • Switch back to English and confirm the top bar, transport, inspector, and export labels return to English.
  • In the Composition facet, pick another Format (16:9, 9:16, 1:1, 4:3, 4:5, 16:10, 10:16, or a timeline shape under Original) and confirm the preview frame changes shape.
  • With a Format whose shape differs from the recording's, confirm the Recording row offers Whole and Follow cursor; confirm Follow cursor fills the frame with a window on the recording that follows the cursor, and Whole shows all of it.
  • Pick Auto in the Format row, drag padding from 0 to 100, and confirm the frame reshapes live with one border thickness on all four sides; switch the camera layout to Dual frame, then Vertical stack, and confirm the frame widens, then turns portrait, with a square camera. Export and confirm the file has the preview frame's shape and the same even border. Then add a clip of another shape, a differently cropped clip, or a clip without a camera, and confirm the frame does not move and Auto stays listed, disabled with the reason shown ("Clips differ"); pick 16:9 and confirm Auto is no longer offered.
  • Press Esc or click outside an open menu, popover, or dialog and confirm it closes.

App menu, About, and updates

  • On macOS, open the application menu and confirm it lists About OpenScreen, Permissions…, Save Diagnostics, Star on GitHub and Check for Updates.
  • On Windows and Linux, right-click the tray icon and confirm it lists Check for Updates and About OpenScreen. Outside the editor the tray is the only surface reachable by default there: the HUD is frameless and the editor and notes windows auto-hide their menu bar, so the Help menu appears only while Alt is held over one of those two windows.
  • On Windows and Linux, open the editor, hold Alt, and confirm the Help menu lists Check for Updates and About OpenScreen.
  • In the editor, click the OpenScreen wordmark in the top bar and confirm it opens a menu listing New project, Open project and Save project; Keyboard Shortcuts, AI settings, the theme switch and Change language; then Check for Updates, About OpenScreen and Star on GitHub. This is the discoverable path on Windows and Linux, where the two above are not.
  • Confirm the About row in that menu shows the running version, and that it matches what the About box then reports.
  • Open the wordmark menu and pick Keyboard Shortcuts; confirm the shortcuts configuration dialog opens and that only one dialog appears.
  • Open the wordmark menu and pick AI settings; confirm it opens the same provider dialog the AI panel's gear does, and that only one dialog appears.
  • Repeat that in Media mode, in Record mode, and in Edit mode with the chat panel collapsed — the three states in which the dialog had no owner before, and the reason the row must not be Edit-only.
  • Connect or disconnect a provider from the menu's dialog while the chat panel is open behind it, close the dialog, and confirm the composer and the model pill follow without reopening the panel.
  • Open the wordmark menu, then press Escape, click elsewhere in the top bar, and click the wordmark again — confirm each closes it and that the window does not start dragging instead of registering the click.
  • post-1.10.0 — Confirm a background check that finds nothing shows no dialog at all — the background path never reports "you are current".
  • post-1.10.0 — Confirm a failed or unavailable download stops short of the restart prompt rather than offering to restart into an installer that was never fetched.
  • post-1.10.0 — Reach Save Diagnostics from the tray context menu while idle, and from the Help menu on Windows and Linux (Alt) or the app menu on macOS; confirm each writes a bundle. It is deliberately not in the wordmark menu.
  • With the wordmark menu open, walk it with the Down and Up arrows and confirm focus wraps at both ends.
  • Switch the app language and confirm the wordmark menu's labels follow — Keyboard Shortcuts and AI settings matching the dialogs they open, Check for Updates and About OpenScreen the wording the macOS app menu and the tray use.
  • Open About and confirm it names the running version, the Electron/Chromium/Node versions, and the install channel.
  • Confirm the About box opens in front of the HUD rather than behind it.
  • On Windows and Linux, press Copy in the About box and confirm the clipboard holds that same block.
  • Open the HUD's device-settings panel and confirm its About row reports the same version.
  • Run Check for Updates from the menu, from the tray, and from the HUD panel, and confirm each reaches the same result dialog.
  • Start a second check while one is still running and confirm the HUD button stays disabled and reads "Checking…" until the first check's dialogs are done, rather than re-enabling into a click that does nothing.
  • Start a recording, then confirm Check for Updates is gone from the app menu, the Help menu and the tray for as long as the take runs, and returns when it stops.
  • Open the HUD's device-settings panel, start a recording with it still open, and confirm the update button disappears while the version stays. Then stop the take, reopen the panel, and confirm the button is back — a HUD that mounts mid-take must not lose it permanently.
  • On a Microsoft Store, Flathub, Snap, or Nix install, confirm no update affordance appears in the menu, the tray, or the HUD panel, while the version still shows.

Updating in place — post-2.1.2

Why this section is never skipped on a candidate. An update is the one path a release cannot fix after the fact: a build whose updater is broken can only be left through that broken updater, so everyone who installs it stays on it. Run this section for every release candidate, on every platform, and for any change to auto-updater.ts, background-update.ts, install-channel.ts, update-checker.ts, mcp/update-notice.ts, the update code in main.ts, the NSIS or publish config in electron-builder.json5, or the feed scripts (scripts/mac-update-feed.mjs).

A dev build cannot run it. The background check and the update button both require app.isPackaged, so npm run dev shows none of them. Test an installed build, on a channel that updates itself: NSIS, DMG, AppImage, deb, rpm or pacman. Never the Store package, which is kept current by the Store.

Getting a newer version to update to. Two ways, both against an installed build of the code under test.

  • The release candidate path, every platform. Install rc.N from its GitHub pre-release, then publish rc.N+1 (the normal RC flow, see release-and-secrets.md). An installed RC follows the newest rc pre-release on GitHub, not the stable release, and every RC publishes latest.yml, latest-mac.yml and latest-linux.yml, so rc.N finds rc.N+1 on its own. This exercises the code of rc.N, which is the point. The only way on macOS: editing anything inside a signed .app, app-update.yml included, breaks its seal, and Squirrel then refuses the update.

  • A local feed, Windows and Linux, for a change before any RC exists. Install the CI-built artifact of the branch (per user on Windows, so no UAC prompt gets in the way). Keep a copy of resources/app-update.yml in the install folder, then replace its content with:

    provider: generic
    url: http://127.0.0.1:8765/
    updaterCacheDirName: openscreen-updater
    

    In an empty folder, put any working installer of the same platform and architecture: the previous release, or the same artifact again. Its file name must contain x64 or arm64 (the architecture electron-updater matches). Write latest.yml beside it (latest-linux.yml for an AppImage), naming a version above the installed one, for example 9.9.9:

    version: 9.9.9
    files:
      - url: Openscreen.Setup.2.1.2-x64.exe
        sha512: <base64 SHA-512 of the file>
        size: <bytes>
    path: Openscreen.Setup.2.1.2-x64.exe
    sha512: <base64 SHA-512 of the file>
    

    Get the digest with node -e "console.log(require('crypto').createHash('sha512').update(require('fs').readFileSync(process.argv[1])).digest('base64'))" <file>. It must be base64, not hex: a hex digest fails every download with a checksum mismatch. Serve the folder on that port from the folder itself, for example with node -e "require('http').createServer((q,s)=>require('fs').createReadStream('.'+decodeURIComponent(q.url)).on('error',()=>{s.statusCode=404;s.end()}).pipe(s)).listen(8765,'127.0.0.1')".

    How to tell an install happened. The installed version may not move: a CI artifact carries the version in package.json, often the same as the last release. The tell is app-update.yml: the installer writes its own, so once an update went through, the file is back to provider: github.

Before each run, and to clean up. Empty the updater's cache, or a download from the previous run is reused without being fetched: %LOCALAPPDATA%\openscreen-updater, ~/Library/Caches/openscreen-updater or ~/.cache/openscreen-updater. Afterwards, restore the saved app-update.yml or reinstall, delete mcp-update-check.json from userData, and stop the local server.

What to watch. The first check runs a minute after launch, so wait at least that, plus the download. Nothing logs the updater itself; a failed check or download logs one [updates] line, which tray → Save Diagnostics keeps.

  • post-2.1.2 — Launch the installed build and wait. Confirm no dialog opens at any point. Once the download has finished, a blue update button appears in the HUD bar, just before the language button, and in the editor's top bar, just before Export, and the tray gains Update to .
  • post-2.1.2 — Start a recording and confirm the button is gone for the length of the take, then back.
  • post-2.1.2 — Quit with the button showing (Quit OpenScreen on the HUD, or tray → Quit), then relaunch. Confirm nothing installed on the way out: no installer, no UAC or password prompt, app-update.yml unchanged. The button is back a minute later. Only a click installs.
  • post-2.1.2 — Click the button. Confirm the app restarts into the update. On a per-machine Windows install, and on deb, rpm or pacman, the system asks for an administrator first (UAC, or the password dialog), and the update installs once granted.
  • post-2.1.2 — With an agent's session open (Openscreen mcp started in a terminal and left waiting on its input, or Claude Code or Codex connected to it), click the button. On Windows the update ends that session, as the installer replaces every running copy. That is expected: the click is the user's decision. The agent's client gets the new version the next time it starts the server.
  • post-2.1.2 — Stop the local server, or point url at a port nothing serves, and relaunch. Confirm the button still appears and opens the release page in the browser: when the self-updater fails, the release lookup on GitHub is the way out. This needs a GitHub release newer than the installed version, so run it on an RC whose stable successor is not out yet, or on any build older than the latest release.
  • post-2.1.2 — The agent-only user. On the installed build, write {"latestVersion":"9.9.9","checkedAt":<now in ms>} to mcp-update-check.json in userData, then start Openscreen mcp and send it an initialize request. Confirm the instructions in the answer open with "OpenScreen 9.9.9 is available" and tell the agent never to update OpenScreen itself. Delete the file, and confirm a Store install, or a dev build, says nothing even with it present. No computer-use is needed: a client that prints the initialize answer is enough, such as a short Node script that spawns the process, writes one JSON-RPC line and reads the first non-blank line back. On Windows the first line Electron writes is blank.

New effects and controls — v1.8.0

  • Type a zoom level outside the preset row (1.25, say) in the Custom zoom field, commit it, and confirm the preview scale and the retained value both follow, with no preset pressed.
  • Activate the timeline toolbar's Auto-Focus all zooms and confirm every zoom switches to automatic focus.
  • With the global toggle on, open a zoom's Focus mode row and confirm it shows Auto, disabled, with the note that the timeline's Auto-Focus button sets it, instead of silently ignoring a per-zoom change.
  • Turn the global toggle off and confirm per-zoom focus mode becomes settable again.
  • Set a speed of 16×, the maximum, and confirm the preview plays it; export that range and confirm the exported timing is correct.
  • Enter a speed above 16× and confirm the "Speed can't go higher than 16×" message rather than a silently clamped value.
  • Enable the webcam's Shrink on zoom and confirm the camera shrinks while a zoom plays and returns afterwards.
  • Choose each webcam layout preset, Vertical stack and Dual frame included, and confirm the preview arrangement changes.
  • Choose each webcam shape and confirm the mask changes in the preview.
  • Apply each text animation in turn and confirm the animation runs when the playhead enters the region.
  • Switch a text's Background plate from Dark to None and back, and confirm the dark plate returns rather than another colour.
  • Switch a blur annotation between Smooth and Mosaic and confirm the obscured area changes accordingly in the preview.
  • Set a blur shape to Oval and confirm the obscured area is elliptical in the preview.
  • Confirm a new blur offers only Rectangle and Oval: Freehand is listed only for a blur that already uses it, in a project from an earlier release, with the note that the export fills the rectangle around it.
  • Export a frame containing such a freehand blur and confirm the export covers its bounding box, which over-covers rather than under-covers, as that note states.
  • In Edit clip, change the crop Ratio between a preset and Free, apply, and confirm the preview reframes.
  • Confirm a cropped project exports with the cropped framing rather than the original.

AI camera background — post-1.10.0

The mask comes from the native compositor (ONNX Runtime + the vendored selfie-segmentation model), not from the renderer. Three things have to line up — the addon, the ONNX Runtime shared library beside it in electron/native/bin/<tag>/, and the model under public/mediapipe/ — and probeSegmentation answers with ready only when all three do. Check both directions: a control that silently does nothing is the defect this replaced.

  • With a webcam recording loaded, open the Camera layout facet and confirm a Camera background row offers Original, Cutout, Blur and Custom.
  • Choose Cutout and confirm the camera's background disappears in the preview, leaving the subject over the project background.
  • Choose Blur and confirm the background blurs while the subject stays sharp, and that a blur-intensity slider appears.
  • Choose Custom and confirm an image/color/gradient chooser appears and the selected wallpaper replaces the camera's background.
  • Confirm the first frames after switching modes may render unsegmented — the worker starts lazily. Scrub or let the preview advance before judging; a still, paused preview is not evidence the effect is inert.
  • Export a range with a camera background set and confirm the exported frames carry the same mask as the preview, not the untouched camera.
  • Remove the ONNX Runtime library from electron/native/bin/<tag>/ (in a packaged build, under its resources folder), restart, and confirm the whole Camera background row is absent rather than present and inert. The persisted mode stays in the document and the camera renders unsegmented; that is correct.
  • Put the library back, restart, and confirm the row returns without any other change.
  • On an Intel Mac, confirm the row is absent: upstream publishes no ONNX Runtime for osx-x64, so the probe can never answer ready there.

Glass Lens cursor and click sound — post-2.0.0

  • In the Cursor facet's Cursor style, pick Glass Lens and confirm the preview draws a glass lens that refracts the picture under it, for the arrow and the hand; turn 3D cursor on and confirm the 3D glass model with its shadow. Export and confirm both are in the file, on each platform: the shader is per backend.
  • On a take with clicks and Show cursor on, confirm Mouse clicks is off by default; turn it on and confirm each click sounds in the preview and a Click volume slider appears (-24 to 0 dB).
  • Export and confirm the clicks are in the audio, none from a trimmed-out span, at the timing a speed region gives them.
  • Turn Show cursor off and confirm the row disappears and the export has no clicks.
  • Pause during playback and confirm no pending click sounds after the pause.
  • Run the click checks on an installed build, not only a dev one: the click samples were missing from installers before b9ed58779.

Editor shell and dialogs — post-1.10.0

  • Double-click a clip and confirm the Edit Clip preview box carries the source's own aspect ratio, not a fixed 16:9 box — a portrait source must fill it rather than letterbox.
  • Focus the crop frame in Edit clip and press an arrow key; confirm the frame moves by 1% of the source, and that with Shift held it resizes from its right or bottom edge, keeping a picked ratio. The arrows must not seek the timeline behind the dialog.
  • Change the Ratio between a preset and Free, apply, and confirm the preview reframes and the clip's stale crop metadata does not survive.
  • Hold Ctrl and scroll over the ruler, over the hint labels, and over the navigator bar — not only over the lanes — and confirm the timeline zooms in each case.
  • Hold Shift and scroll over those same three places and confirm the visible range pans.
  • Open the clip picker, click outside it, and confirm it closes.
  • Select a clip and confirm its delete button is the same round raised button as the edit pencil, rather than sitting bare on the waveform; on a clip too narrow to hold them, confirm both sit beside it.
  • Switch to the light theme and confirm the gradient picker and the dialogs follow it instead of staying dark.
  • Confirm the top bar carries no second settings button beside the wordmark menu.
  • Start a transcription and confirm the status reads Downloading speech model (first run) or Starting speech model before Transcribing, then a percentage, on every surface that shows it.
  • Regenerate a transcript and confirm the busy label stays visible for the duration and is scoped to the timeline rather than leaking to unrelated surfaces.
  • Open the media asset card's Regenerate as picker and confirm it lists every whisper language (101 entries including Auto), sorted by localized name — not a hand-picked handful.
  • Choose a language, regenerate, and confirm the new transcript replaces the old one.
  • Hover the inspector's switches and choice rows, in Composition, Camera layout and Cursor, with real mouse moves and again with the keyboard. Confirm no tooltip only repeats a visible label, that the icon-only options (camera preset, camera position, frame) each show their name, and that Depth of field, Auto-hide when inactive, 3D cursor, Click impact and Shrink on zoom each show one line saying what they do, on the switch itself. See tooltips.md.

Deleting a project's recordings — post-2.0.0

  • Record a take, open Load Project, click the trash icon on its row, and confirm the row offers Also move its recordings to the Trash (Recycle Bin on Windows) with a size, switched off. Delete with it off and confirm the files are still in the recordings folder.
  • Record again, delete with the switch on, and confirm the screen recording (.mp4, or .webm from the browser recorder), its .cursor.json, its .session.json and any -webcam file are in the system Trash and can be restored from there.
  • Add one recording to two projects, delete one of them with the switch on, and confirm the row says the recording is kept for the other project by name, and that the other project still opens with its media.
  • Import a video from outside the recordings folder into a project, delete it, and confirm the dialog offers no switch and the file is untouched.
  • Delete the project that is open in the editor, and confirm its row offers no switch and says the recordings are kept: the preview holds the take open, which Windows will not move.

Editor window bounds — post-1.10.0

  • Move and resize the editor window, close it, reopen it, and confirm it returns at the same position and size.
  • Maximize the editor, close and reopen, and confirm it returns maximized.
  • Un-maximize, close, and confirm editor-window.json in userData records maximized: false with the restored bounds.
  • Hand-edit that file to a zero width and height and a non-boolean maximized, then relaunch; confirm the editor opens at its default size rather than restoring an unusable window.

Persistence (save, reopen, reload)

Edits are saved as they land. The dot after the project name reads "Unsaved" (hover title) only in the middle of a drag, or when a save failed.

  • Make a project change and confirm the dot after the project name reads "Saved" once it lands, with no save step.
  • Save with OpenScreen menu → Save project (or Ctrl/Cmd+S) and confirm the "Project saved" toast.
  • Close and reopen the project from the Open project dialog and confirm the asset and project title match before closing.
  • Confirm clip order and each clip's in/out and crop settings survive reopen.
  • Confirm trim, zoom, speed, annotation, and full-camera regions survive reopen with their positions and values.
  • Confirm background, effects, layout, webcam, cursor, aspect-ratio, and caption settings survive reopen.
  • Confirm the transcript and skipped-word ranges survive reopen.
  • Confirm the seekable duration after reopen reaches the recording duration, not merely the end of the last region.
  • Make a change, open another project, and confirm it opens with no unsaved-changes prompt, since the change was already saved; reopen the first project and confirm the change is there.
  • When the unsaved-changes prompt does appear (a change that has not reached the disk, such as after a failed save), confirm Cancel keeps the current project loaded, Save & continue opens the next project after saving, and Discard opens it without the change.
  • Open a project saved by a previous release and confirm it loads without a schema error.
  • Confirm every modifier in that migrated project sits on the clip content it covered before, not at a shifted ruler position.
  • Confirm a migrated project that had a region straddling two clips still renders it as one pill while the clips remain adjacent.
  • Save the migrated project, reopen it, and confirm nothing shifted on the second round-trip.

Platform-specific

Windows

  • Run the complete capture-to-export flow on real Windows with the packaged build.
  • Confirm a screen source and a single-window source both produce non-black video.
  • Confirm the system tray icon appears and changes to a recording state while recording.
  • v2.0.0 — With a physical camera, run Webcam capture quality on Windows: the Media Foundation checks, the no-visible-frame warning, and the DirectShow ones when a DirectShow-only camera is at hand.
  • Right-click the tray icon while recording, choose Stop Recording, and confirm the editor opens.
  • Relaunch with OPENSCREEN_DISABLE_CONTENT_PROTECTION unset, then confirm the HUD and notes window are excluded from captured video. Before Windows 11 22H2 the HUD is left unprotected and appears in the video; only the notes window must be absent there.
  • Disable hardware H.264 if the test machine supports that diagnostic path and confirm the software-encoder notice is clear and non-blocking.
  • Switch the recording HUD between displays and confirm it remains positioned on the intended display.
  • Switch the desktop to an odd-pixel window size and confirm the recorded frame dimensions remain valid.
  • On Windows 11 24H2 or later, record a single-window source, right-click inside that window and confirm the context menu is in the video; then open one that overhangs the window's edge and confirm it is cut off at the edge. Run it with nothing ordinary overlapping the recorded window: with another window (neither a popup nor a tool window) above it the menu was missing in the synthetic --window-popup fixture (46 runs on a second Windows 11 build 26200 machine: present in 25 of 25 runs with none above, absent in 21 of 21 with one above, see #910), so treat that as expected rather than a failure, without assuming the same for every real application's menu. Before 24H2 the menu is absent too, which is expected (secondary-windows applied:false in the log).
  • Run tray → Save Diagnostics and confirm a diagnostic bundle can be written.
  • post-1.10.0 — Record with no encoder override and confirm the helper's encoder-selection log line reports videoEncoderRuntime: "hardware". The plain sink-writer path never asked for hardware transforms before, so every ordinary recording ran the software encoder; on a slow machine that is what blew the stop-shutdown budget.
  • post-1.10.0 — Confirm forcing the software encoder still reports "software", so the default is a default and not a hard-wire.
  • post-1.10.0 — Record with microphone and system audio and confirm the resulting MP4 carries a valid AAC track at a legal rate (48 kHz).
  • Speak continuously for 30 s with the microphone on, then play the file in a neutral player (VLC, ffplay) and confirm there is no crackle. Include the moment you reach for the HUD to stop: that is where the holes of #911 clustered. In a waveform view, the defect looks like drops to digital silence of up to 10 ms in the middle of words.
  • Record with the webcam and confirm ffprobe reads profile=High, has_b_frames=0 and tv / bt709 / bt709 / bt709 colour tags on the screen and webcam tracks (#922, #923). Then export: solid colours must match the source within ±2 in RGB.
  • post-1.10.0 — On a device whose native rate AAC cannot take (96 kHz), confirm the recording still succeeds with the rate snapped to 48 kHz rather than failing at SetInputMediaType. The helper's own audio_sample_utils_test covers the accept/reject probes at build time; this check is the end-to-end half.
  • post-1.10.0 — Confirm a long recording's audio stays in sync, so the downsample remainder is carried across packets rather than drifting.
  • post-2.0.0 — Record 10 minutes or more with the webcam and system audio while the machine is under load (an export or a game running), and confirm at the end that clicks and lips still match the audio, that the webcam lines up with the screen, and that ffprobe gives the video and audio streams the same duration (#945).
  • post-2.0.0 — With a camera that cannot reach 30 fps at full size (a Logitech C920), confirm every Camera quality preset now records about 30 fps, smaller rather than a slideshow (#1112); a camera that does 30 fps at full size (BRIO, StreamCam) keeps its size.
  • post-2.0.0 — On Windows 10 22H2, confirm the HUD takes clicks on every launch (#1105). It is unprotected there and shows in recordings, by design.
  • post-2.0.0, Windows on ARM — Install the arm64 installer on an ARM device (Snapdragon) and confirm Task Manager shows OpenScreen as ARM64, that a take records, and that transcription works through the x64 whisper helper under emulation, slower but with no "binary not found".
  • post-1.10.0 — On a device that can be set to 96 kHz, record system audio while a 36 kHz tone plays and confirm the recording carries no 12 kHz component. That fold is what an inadequate anti-alias filter produces, and neither of the two checks above would catch it: the rate-snap check only asks that the recording succeeds, and the sync check only asks that frame counts stay aligned. Probe tones must sit well inside what AAC keeps — 12 kHz is fine; a 20 kHz probe was absent from the app's recording, and a 192 kbps AAC encode alone (tested with ffmpeg) removes it too, so it cannot be measured. Needs an endpoint whose shared-mode format is above 48 kHz and an integer multiple of it — check the Advanced tab's format list, and check every endpoint, not just the current format of the default one; a USB DAC is one way to get such an endpoint. Exclusive-mode support is not enough, because loopback reports the shared-mode format. If every endpoint really is 48 kHz, the check cannot run at all: forcing a lower encoder target instead does not work, since every AAC rate that would divide 48 kHz is rejected by the Media Foundation encoder on the host tested.

macOS

  • Run the complete capture-to-export flow on real macOS with the packaged build.
  • Grant the permissions through the permissions window (below) and confirm the HUD reflects the granted devices.
  • Record while switching Spaces with the HUD visible and confirm recording continues.
  • Stop a recording and confirm the editor opens without a crash during native recorder shutdown.
  • Confirm the tray or menu-bar item can refocus the HUD after it is hidden.
  • Confirm the HUD and notes window are absent from a full-screen recording. The capture leaves them out by window id on every macOS version, so this holds with the content-protection flag set too.
  • Confirm a physical webcam picture-in-picture records and plays back with the selected layout.
  • v2.0.0 — With a physical camera, run Webcam capture quality on macOS. The webcam is recorded by the browser recorder here, so the .webm checks apply and the Windows helper checks do not.
  • Export MP4 and GIF and confirm both files open in a native macOS media viewer.
  • Confirm closing and relaunching the packaged app does not leave an orphaned capture or editor window.
  • On the newest supported macOS, confirm the HUD and notes windows are visible on screen rather than blanked by content protection.
  • Confirm the HUD opens without waiting for the microphone permission prompt to be answered.
  • Confirm local transcription reports the device it actually ran on and completes on a Metal-capable machine.
  • Confirm the packaged .app contains the compositor addon and that the addon carries no build-machine path.
  • Confirm the packaged .app bundles its ffmpeg libraries and runs on a machine with no developer toolchain installed.
  • On a Retina/HiDPI display, record the screen and confirm the recorded frame is filled edge to edge — not the desktop drawn small in one corner of a black rectangle. Then do the same for a single window. Issue #418 shipped exactly this, invisible on every 1× display because a point size and a pixel size are the same number there; SCStreamConfiguration does not scale a frame up to fill an oversized buffer, so the surplus stays background black. Check the frame, not just the file's dimensions — the reporter's .mp4 was 3024×1898 as expected and still wrong inside.
  • With a second display attached at a different scale factor, record each display in turn and confirm both fill their frame. A machine whose displays all share one scale factor cannot catch a units mix-up.

Permissions window — v2.0.0

On macOS 15.2+ it opens at launch, until it has been closed once, while one of its rows was never asked. Before 15.2 it opens at launch while Screen Recording is missing or waits on a relaunch. Tray → Permissions… and the app menu open it any time. A Record-mode start with Editable cursor enabled may request Accessibility from macOS; a pending grant does not reopen this window or block recording. To see a first run on a Mac that has run OpenScreen before, its grants have to be reset first, which is the Mac owner's decision.

  • On a first launch, confirm the window "OpenScreen needs a few permissions" lists, on macOS 15.2+, System audio (Optional), Accessibility (Recommended), Microphone (Optional) and Camera (Optional); before 15.2, or with OPENSCREEN_MAC_SOURCE_PICKER=legacy, Screen & system audio (Required) comes first instead of System audio.
  • For each row whose permission has not been requested, activate its button (Allow, or Continue for the screen row) and confirm macOS raises its prompt; allow it and confirm the row turns to Allowed while the window stays open.
  • For a permission already requested but still denied, confirm the row offers Open Settings on the window's first appearance, and that it opens that permission's System Settings pane instead of raising a new prompt.
  • With Screen Recording ready, Accessibility ungranted, and Editable cursor enabled in Record mode, press Start recording. Confirm the macOS prompt may appear, the countdown still reaches an active take, and a warning says cursor effects may be limited while Accessibility is pending. Stop the take while leaving Accessibility pending, then start another take; confirm the OpenScreen permissions window does not reopen and recording still starts. Grant Accessibility in System Settings and confirm a later take has full cursor effects.
  • Refuse a permission, reopen the window from the tray, and confirm its row still offers Open Settings and opens the matching System Settings pane.
  • Before 15.2: confirm Get started stays disabled until Screen Recording is allowed; that the window offers Restart OpenScreen when macOS needs a relaunch to apply it; and that after System Settings' Quit & Reopen the window comes back.
  • Close the window with Get started and confirm it does not open again at the next launch.

Running the pass unattended on macOS

Apple's picker is owned by Control Centre, and computer-use cannot be granted Control Centre: no click and no key reaches the picker. Left as is, a human clicks Share Entire Screen for every take. Instead:

  1. Run the human slice first, in one sitting, before any Screen Recording grant: the Apple's source picker checks below, and the permissions window's 15.2+ rows, on a launch without the variable. One of them confirms the picker records with no Screen Recording grant, which only means something while the grant is absent: on a Mac that already holds it, that part needs the grant reset (tccutil reset ScreenCapture com.etiennelescot.openscreen), the Mac owner's decision, or it is skipped with that reason. A few minutes of clicks, after which the operator can leave.
  2. Then, once per Mac and app identity, the Mac owner allows OpenScreen under System Settings → Privacy & Security → Screen & System Audio Recording, and the app is relaunched: it reads that grant at launch. TCC keys the grant on the signature: a Developer ID build keeps it from one RC to the next, while an ad hoc signed local package asks again after each build.
  3. Relaunch with OPENSCREEN_MAC_SOURCE_PICKER=legacy from the shell, like the content-protection flag. The HUD's source button and Record mode's Source row then open the app's own Screens / Windows selector, which computer-use drives exactly as on Windows. Every other section runs from there without a human.
  4. macOS may ask, about once a month, whether OpenScreen may keep bypassing the private window picker. That alert is a system one too: the operator answers it. It is the only prompt left, and one more reason the grant has to be in place before the legacy run, not discovered during it.
  5. Say it in the results row. A pass run under legacy did not exercise Apple's picker, so the row names which takes used which picker. For a release candidate, the human slice is part of "the whole file": skipping it makes the row Partial.

Recording through the stdio MCP server or the CLI also bypasses the picker: startRecording and record pick a display or window by name, headless. They need the classic Screen Recording grant too, given to whatever launched them (the agent's terminal or IDE for the stdio server). That makes a take, not a HUD check — use it to get footage for the editor sections, never to tick a recording check.

Apple's source picker — macOS 15.2+, v2.0.0

These checks are the human slice: an agent cannot click Apple's picker (see above).

  • Activate the HUD's source button and confirm Apple's picker opens and the HUD hides until it closes.
  • Pick a display, then on another take a window, and confirm the HUD comes back with the pick's name on the source button and records what was picked, with no Screen Recording grant and no "bypass the system private window picker" alert.
  • Dismiss the picker without a choice and confirm the HUD comes back idle; press record with no source and confirm it opens the picker, then starts once a source is picked.
  • From Record mode, activate the Source row and confirm it opens the same picker rather than the app's own list.
  • After picking a display, change Hide desktop icons in Record mode and confirm the pick is dropped, so the next take asks for a source again.
  • With system audio never asked for (not from the permissions window either), turn system audio on and confirm macOS asks for it at once, not at the countdown; allow it and confirm the next take carries system audio. Refused, confirm the permissions window's System audio row offers Open Settings, and that turning OpenScreen on under "System Audio Recording Only" there brings system audio back.
  • Launch with OPENSCREEN_MAC_SOURCE_PICKER=legacy and confirm the app's own selector opens instead, with Screen Recording required.
  • With nothing playing on the Mac (no music, video or call app, so the output device is idle), pick a source, turn on the microphone and system audio, record while talking and play a sound halfway (afplay /System/Library/Sounds/Ping.aiff). Confirm the microphone is heard from the first second, not only from the sound on, and that the sound is in the take. Repeat with the microphone off: the sound is still in the take (#1173).

Microphone on macOS 13 and 14 — v2.0.0

  • On macOS 13 or 14, confirm the HUD, its Device settings and Record mode offer no microphone, and that a take records without trying to.

Linux

  • Run the complete editor-to-export flow on real Linux with the supported packaged or development build.
  • Confirm the HUD remains interactive on the supported Linux window manager.
  • Select a screen source in the compositor's portal picker and confirm the resulting recording is not black.
  • Select a single WINDOW in the portal picker and confirm the recording contains only that window, at the window's dimensions — not the whole screen. Check the pixel size, not just the look of it: ffprobe -v error -select_streams v:0 -show_entries stream=width,height -of csv=p=0 <file> should report the window's size, never the monitor's. This is the case that shipped broken in 1.8.0.
  • Record twice in a row and confirm the portal picker appears BOTH times, and that choosing a different source the second time actually changes what is recorded.
  • Confirm the HUD shows no in-app source button on Linux, and that the record button starts a recording directly instead of opening a picker.
  • Confirm the portal picker appears BEFORE the 3-2-1 countdown, not during or after it.
  • Start the same flow from the editor's Record mode (Start recording) and confirm it behaves identically to the HUD — no source to pick, picker first, then countdown.
  • post-2.0.0 — On a distro without libfuse2 (Ubuntu 24.04 as installed), confirm the AppImage launches directly.
  • post-2.0.0 — For a stable release, confirm the assets carry *.AppImage.zsync and latest-linux.yml, and that AppImageUpdate finds the update from the previous AppImage. A release candidate ships the .zsync but embeds no update information.
  • post-2.0.0 — Where DMA-BUF buffers fail (some NVIDIA or VM setups), confirm a take starts anyway, after the capture-retry-shared-memory warning.
  • post-2.0.0 — Run the CLI's record on Wayland and confirm the portal picker asks once, with the log line "Recording source: chosen in the system picker (--display and --window do not apply)".
  • Cancel the countdown after answering the picker and confirm the compositor's "screen is being shared" indicator goes away rather than lingering.
  • Confirm the system tray or supported desktop indicator can refocus the HUD when it is hidden.
  • Confirm microphone capture works with a physical device and the chosen device is audible in playback.
  • Confirm the HUD's Device settings lists the physical camera, or reports "No camera found" when there is none.
  • v2.0.0 — With a physical camera, run Webcam capture quality on Linux. The webcam is recorded by the browser recorder here, so the .webm checks apply and the Windows helper checks do not.
  • Confirm the native compositor preview loads without a blank surface or renderer crash.
  • Export MP4 and GIF and confirm the files open in a system media player.
  • Close and relaunch the app and confirm a saved project can be reopened without data loss.
  • post-2.0.0 — Run Deleting a project's recordings on GNOME and on KDE, and in the Flatpak build if it ships, and confirm the files land in the desktop's own Trash. Where the Trash refuses (a recordings folder on a drive with no Trash), confirm the row turns to Delete them permanently? and nothing is deleted until that is confirmed.

Results log

Date Build / tag Platform Pass/fail Notes
2026-10-25 fix/localize-export-validation-errors on d433e7ce; fresh npm run build-vite; release addon SHA256 71304292f7d4406fc53e550ee90363cd6f087d6f030f3787343f54e4d9c76be0 macOS 15.7.1, Apple M2; Electron 41.2.1; isolated synthetic-media profile Partial — localized errors and MP4 retry passed Computer-use switched the test profile to Simplified Chinese. An interior-gap export and a tail-only MP4 both rejected after confirming replacement of a seeded file. The error panel and notification used Chinese; the destination SHA256 remained 9db4e92b90f2b768062322a1c72408284f6f5caea473380d6e46e4090e2fcac5 after each failure, with no partial output. A normal four-second clip plus 10 ms tail then exported successfully in the same session at 1080p/30; independent ffprobe counted 120 frames / 4.000 s at 1752×1080, and full-file decode exited 0. Process file mappings confirmed the fresh addon and matching FFmpeg libraries were loaded. The standard addon script compiled Rust but failed during staging because the local SDK already contained @loader_path; test libraries were staged separately with that existing path preserved and ad-hoc signatures refreshed. This is not packaging validation. The profile also reported its missing optional whisper server; transcription was not tested. Skipped: physical capture/webcam, Presenter Overlay, unrelated editor features, native Windows/Linux execution, all-language visual review and release packaging.
2026-10-24 PR #1379 offset-origin follow-up; rebuilt addon SHA256 0dd01525a163d8889dd2aaa7088ea0c1e923956b14c430ff70537cba5c24e9d5 macOS 15.7.1, Apple Silicon Partial — offset source exported; interior gap still rejected Computer-use opened a synthetic video whose first decoded timestamp is 1.000 s, with a clip starting at 0. MP4 1080p/30 displayed Saved; independent probing counted 120 frames / 3.967 s at 1752×1080 and full-file decode exited 0. In the same session, opened the interior-gap project and confirmed replacement of that MP4: export reported the screen-start error, destination SHA256 stayed d45192c6c319006849fee014bc3882141bead0f8b0c43e7c895c51cefe1b7d7b, and no partial remained. Native regressions cover screen/camera offset sources and interior gaps with both zero and non-zero stream origins. Recording, Presenter Overlay compatibility, unrelated editor features, Windows/Linux runtime and release packaging were not exercised.
2026-10-24 dev post-merge follow-up to PR #1377 on 79c3aa24; rebuilt addon SHA256 53df86189b359d41b63a76e3eaafed5e9bf0e299fe0e89f8fd7950aa12bad6f3 macOS 15.7.1, Apple Silicon Partial — interior gap rejected; destination preserved; normal retry passed Computer-use opened a synthetic screen clip starting at 1.5 s inside a timestamp gap from 0.983 to 3.000 s, then confirmed replacement of an existing synthetic MP4. Export reported that clip 1 had no readable screen frame at its start. Destination SHA256 stayed 854b0911778da18c2a456d3eb227da749c0af2138ad52092180156ed138a6a48; no partial remained. In the same session, a normal four-second clip plus 10 ms tail exported successfully at 1080p/30: independent ffprobe counted 120 frames / 4.000 s at 1752×1080, and full-file decode exited 0. Native tests also cover camera gaps and failure after a valid preceding clip. Recording, Presenter Overlay compatibility, unrelated editor features, Windows/Linux runtime and release packaging were not exercised.
2026-10-24 dev CodeRabbit follow-up to PR #1377; rebuilt addon SHA256 8fe1baab14e941a5ba6d73b5596a27190a75edcd8b49630b451868f829bdaf71 macOS 15.7.1, Apple Silicon Partial — undecodable output rejected; destination preserved; retry passed Computer-use reopened the same synthetic tail-only project used in the previous run, selected an existing synthetic MP4, and confirmed the native replace prompt. Export now reports validation failure instead of Saved. Destination SHA256 remained 854b0911778da18c2a456d3eb227da749c0af2138ad52092180156ed138a6a48; no partial output remained. Without restarting the app, loaded the four-second valid clip plus tail and retried MP4 1080p/30: Saved, native walk 121 frames, independent ffprobe 120 decoded frames / 4.000 s at 1752×1080, full-file decode exit 0. This closes the publication-safety gap observed in the previous row; single-frame encoder timing itself remains unchanged. Native tests also cover absent/existing destination, malformed output, cancellation and successful H264/HEVC export. Recording, Presenter Overlay, unrelated editor features, Windows/Linux runtime and release packaging were not exercised.
2026-10-24 dev fix/reject-empty-exports, PR #1377 follow-up; rebuilt addon SHA256 2ad4ce50da3e2e8dd2a6dc67017048d5ae515f5b0d7143d509d727c9d09c29da macOS 15.7.1, Apple Silicon Partial — multiclip GUI export passed; single-frame muxing limit observed Computer-use opened an isolated profile and a synthetic four-second clip followed by a 10 ms tail starting at 3.99 s. MP4 1080p/30 displayed Saved; native walk reported 121 frames, independent ffprobe decoded 120 frames / 4.000 s at 1752×1080, and full-file decode completed without errors. Tail alone produced one muxed video sample with zero video duration and no decoded video frames; this additional single-frame muxing case is unresolved and no baseline attribution is claimed. Initial native-path entry failed; selecting the prepared project in the file list worked. Native export/inspection tests separately: 7 passed, 1 existing macOS frame-count-probe stub failure. Recording, Presenter Overlay compatibility, other editor features, Windows/Linux runtime and release validation were not exercised.
2026-09-23 dev fix/725-hidden-cursor-auto-zoom; rebuilt compositor addon SHA256 a0bcd61e17084c010141049a9978121e45be2985449ff1618ab6a3145f8381f6 Windows 11 Home 26200 Partial — issue #725 slider slice passed Human-driven real recording in the Electron dev build: moving a slider whose application cursor becomes hidden kept the reconstructed cursor hidden in the Studio recording and did not create a click-driven automatic zoom for that press. A visible control click still produced the normal automatic zoom. Sidecar inspection independently confirmed the hidden interval (visible: false) and that no zoom was centred on the press immediately preceding it. Recording, stop and editor open passed. Not covered: export, packaged build, audio/webcam, tray/HUD, other editor features, macOS or Linux.
2026-09-12 dev fix/643-gif-export-cancel, based on fbe461e9; rebuilt compositor addon SHA256 128c89fe65a459b3f17721d610ea9a78e0d3e980e9eb0b5fb2b0e5d018619e15 Windows Partial — export cancellation passed Computer-use OS input, real Electron and native save dialog, imported synthetic 8 s source. GIF 15 FPS / Medium (720p) / loop on: observed 55/120 frames, clicked Cancel while rendering (pending snapshot 104/120), returned to the same options; no final or partial output remained. Retried in that dialog to the same path and received Saved to; resulting GIF is 1280×720, 120 frames, 8,200,202 bytes, fully decodes with ffmpeg (exit 0). Automated real-addon tests separately pass existing-destination preservation, GIF retry, ordinary GIF and MP4 export. Desktop adapter required screenshot-based coordinates and Ctrl+O to reach the editor; HUD reachability is not claimed. An unrelated process error window partly obscured the retry success screenshot; file verification independently passed. Skipped: recording, audio/mic, webcam, tray/HUD, effect/caption comparisons, second viewer, macOS/Linux and release promotion. Task-local profile and evidence retained; test app stopped.
2026-07-31 dev build, claude/e2e-tests-v1-8-0-474894 (e9578f09) macOS 26.5, M1 Partial — 1 defect Ran launch/HUD, media, modifier anchoring, and export. Defect: a dangling asset blanks the preview. Modifier anchoring across a reorder verified in preview and in the exported frames. macOS export produced 1280×720 h264 + AAC at ~2× realtime. Chat sections skipped: no AI provider configured. HUD drag not runnable under computer-use (drop point is the desktop).
2026-08-13 installed v1.9.5-rc.1 Windows 11 26200, 1920×1080 @ 100% Partial — 1 defect Ran launch/HUD, source selection, recording, stop, editor open. Fragmented MP4 confirmed on the shipped artifact: 48 moof+mdat pairs over 47.6s, mvex present, mfra on clean stop. Defect: a recording that survives a helper kill is thrown away by the app — killing wgc-capture.exe mid-recording leaves a fully decodable 41s file (2460 packets, ffmpeg -f null - exit 0) with no .session.json and no .cursor.json, and stop answers "The recording could not be saved". Fixed in #363, re-verified end to end. Truncation ablation at 60%: plain MP4 unreadable, fragmented plays 29s. A dev build cannot test any of this — the prebuilt worktree helper predated the change and silently ran the old path. Editor/export/chat sections not run.
2026-08-14 release/v1.9.5 @ b1b81de5 (rc.2 candidate: dev TS + the CI-built rc.1 native payload, which is byte-identical since no native source changed) Windows 11 26200, 1920×1080 @ 100% Pass — no defect Regression net across the 65 commits since v1.9.2, not just the rc.2 delta. Four recordings. Every one a fragmented MP4 (mvex + ~1 moof/s, mfra only on a clean stop). GPU DXGI path still correctly opt-in (videoInput: cpu-rgb32) — the #336 regression has not crept back. No capture-pacing drift: HUD 00:59 → 60.067s at 60/1. Waveform correct in both directions: absent with no audio track, rendered with one. Audio muxes into the fragmented container (AAC 48k stereo) with 15 ms A/V drift, under one frame. Compositor renders and exports with no camera declared. #366: reopening returns to the saved project with its settings (Blur BG on, padding 9%) and mints no second project — 167→168 across a whole new recording. #363: helper killed mid-recording → editor opens on the recovered take (46 moof, no mfra), all three sidecars written, imported once. Export MP4 1080p60 from that recovered take: 46.0s / 2760 packets, decodes clean, duration matches the source exactly. Tray refocus works. NOT covered: DPI scaling — not re-run here, already validated when 60bb6d7c / 71cc88d6 landed; note that the display scale is a setting, so "this machine is at 100%" is never a reason a DPI bug cannot be tested (flip it to 150%, ~2 min). Also not covered: webcam PiP and the export webcam fixes, microphone, GIF, macOS/Linux, AI sections, packaging.
2026-09-02 worktree fix/wgc-aac-sample-rate (rebuilt wgc-capture.exe) Windows 11 Partial — helper-level only Not a capture-to-export HUD pass. Media Foundation SetInputMediaType for 96 kHz PCM into AAC returns 0xc00d36b4; 48 kHz succeeds. Live helper with snap on: forced 96 kHz encoder source reports 48 kHz and recording-started. Live helper with snap off: initialize fails at SetInputMediaType(audio) with 0xc00d36b4 (hardware, software, and plain MP4 retries). HUD, editor, export, webcam, and tray not run.
2026-08-14 installed v1.9.5-rc.1, macOS Apple Silicon DMG (CI-built, Developer ID signed). rc.2 is not published — only rc.1 exists on Releases; no native source changed between v1.9.5-rc.1 and origin/release/v1.9.5, so this artifact already carries the rc.2 native payload, but #366 (cross-platform TS) is absent from it macOS 26.5 (25F71), M1, 1920×1080 @ 2× Fail — 1 blocker The plan's assertion-1 criterion does not hold on macOS, in both directions. On a clean stop AVAssetWriter.finishWriting() collapses the fragments into a normal movie: ftyp mdat moov, mvex ABSENT, 0 moof, no mfra (45 s / 44.4 MB run). That is exactly the shape the plan calls the headline failure — and the pre-a6795d23 control recording (2026-08-10) has the same shape — so a clean-stop box walk cannot distinguish fragmented from plain on macOS; only the kill test can. Fragmenting is active: the takes whose writer died mid-fragment retain mvex + 1 moof per second of media (shipped-build writer-failure samples: 35 moof/36.0 s, 14/15.0 s, 3/4.0 s; plus 18 on a surviving-helper kill). The one kill on the shipped build is the exception that proves the scope — capture had already stalled ~12 s before the kill, so it carries mvex but 0 moof and only 1.0 s. No macOS file, clean or killed, ever carried mfra. Blocker: every app-driven recording truncates, then the app discards it. (Root cause and fix reported in #375 — the fragments carry a negative composition offset in a version 0 trun, where ISO/IEC 14496-12 8.8.8.2 defines the field as unsigned, because frame reordering was left on; AVVideoAllowFrameReorderingKey: false clears it and restores the crash-resilience the fragmenting was for. Verified at helper level there; this rc.1 run only reproduced the failure and validated nothing about the fix. Re-run this section against a CI build carrying #375 before rc.2 ships.) 3/3 takes stopped writing early while the HUD kept counting — media 4.0 s / 36.0 s / 15.0 s against HUD 02:02 / 01:30 / 01:04. Helper emits {"event":"error","code":"writer-failed"}; main log AVFoundationErrorDomain Code=-11800 … (-16341). Stop then hangs ~30 s on "Saving…" and drops the take: no .session.json, no .cursor.json, no editor. The app does surface the raw error in a toast (confirmed by hand on the same machine at 13:28–13:35 — my automated runs screenshotted after it auto-dismissed, so an earlier draft of this row wrongly said there was none). 44,561,966 / 328,337,979 / 139,631,607 / 17,187,009 bytes decodable and thrown away (147 GB free — not disk). Reproduced standalone with the shipped helper at 1080p30/8 Mbps, 2/2 (9 s, ~5 s), so it is not confined to the app's 4K60 path — but do not read that as load-independent: append rate demonstrably modulates how reliably it bites (#375 measures it reliable at ~57 fps and intermittent at 30 fps). Reproduced by hand, no automation involved, on six takes recording a YouTube page — and those six separate the trigger cleanly: system audio ON → 3/3 died at ~1.0 s and minted 0 projects; system audio OFF → 3/3 survived (3.3 s, 7.4 s, 25.0 s) and minted 1 project each. Audio is not the condition, only an accelerant — a controlled run with system audio off and not one screenshot taken during the capture (the screenshot layer hides non-allowlisted windows, so it was the last confound worth eliminating) died the same way: 8.008 s of video, 79,004,330 bytes then flat for 76 s with the helper still alive, 7 moof, 0 sidecars, 0 projects, same -11800/-16341. What audio changes is the window: with a track it is ~1 s, without one ~4–40 s. That reconciles the by-hand takes with mine — a take short enough to stop before the writer dies is clean, which is why 3.3 s and 7.4 s survived and 8.0 s did not, and why the 25.0 s one minted a project while still carrying mvex (never cleanly finalised). Turning audio off is therefore not a safe workaround. Untested here: microphone — this Mac has no input device, and whether a mic track triggers the same path is an inference, not a measurement. Helper A/B narrows the with-audio path to the fragmentation line: helper built twice from source identical to the rc.1 tag, differing only by writer.movieFragmentInterval (701 vs 700 lines) — with system audio at 1080p30, WITH the line writer-failed 2/2 (2.0 s, 1.0 s), WITHOUT it clean recording-stopped 3/3 (40.6 s, 37.9 s, 37.6 s). Read those counts as a sample, not a law: a later rebuild of the with-the-line arm survived 22.2 s at the same settings, so the failure is probabilistic and rate-dependent, and the byte-level evidence in #375 is what actually carries the case. The video-only local-vs-shipped gap (local survived 45 s, shipped failed 5/5) is explained by the same variable rather than by the released artifact — the shipped runs encoded at 56.6 fps against 29 fps locally. Kill test is confounded on the shipped build (capture already dead before the kill): 17.19 MB → only 1.0 s / 56 packets, 0 moof. On a helper that does not fail, a mid-write kill leaves 18 moof, decodes clean (ffmpeg -v error -f null - exit 0, 1373 packets) and no mfra — the shape the plan expects. #363 gap confirmed, and on macOS it fires with no kill at all: writer-failed alone loses the take; there is no app-side recovery. Audio: AAC 48 kHz stereo muxes into the fragmented container, video start 0.000000 vs audio 0.014479 → 14.5 ms drift, under one frame at 30 fps (measured on the 2.0 s written before the writer died). Compositor + export pass: preview renders with no camera declared; export MP4 1080p60 H.264+AAC via h264_videotoolbox (zero-copy VT), 5,726,865 bytes, 318 packets, decodes clean, duration matches to within 7 ms — source 26.713 s minus trims 19.910 + 1.513 = 5.290 s expected vs 5.283 s measured, under one frame at 60 fps. #366 not runnable as specified (absent from rc.1, rc.2 unpublished, and record→editor never completes); adjacent behaviour measured on an existing project — close+reopen kept 19→19 projects, exactly ONE project references the recording, and Blur BG / padding survived (showBlur=true, padding=16). NOT covered: Windows-only DPI and wgc-capture, GIF, AI sections, packaging (per plan); webcam PiP and microphone — this Mac has neither (Device settings reports "No microphone found" / "No camera found").
2026-08-22 installed v1.10.0-rc.3 — CI-built NSIS artifact from build run 32582966489 (openscreen-windows), App menu → About reports 1.10.0-rc.3, native payload complete and uniformly stamped (19 files in resources/electron/native/bin/win32-x64, all 17:59:10, so helper + compositor addon + av* DLLs are one matched CI set) Windows 11 26200, 1920×1080 @ 100% Pass — 2 minor defects Pause works, and the measurement that says so is the wall clock. createdAt 20:25:52.208 against a file finalised at 20:30:56.754 is 304.55 s elapsed for a 286.333 s file — 18.21 s shorter, exactly the paused interval, so capture was genuinely suspended. The HUD timer froze at 03:58 across two reads 7 s apart with the indicator amber, and resume was clean (04:01 → 04:08 over 7 s, no time lost). An earlier draft of this row called this a blocking defect, on the strength of comparing the file duration against a timer read before the stop click; with tool round-trips of ~20 s that comparison is worthless, and the packet count offered as corroboration proves nothing either — a file is continuous 60 fps whether or not capture was ever suspended. Written down because the wrong version of this measurement is easy to repeat: compare against wall-clock elapsed, never against the last timer you happened to screenshot. Capture is otherwise sound, on two takes. 15.8 s: fragmented (ftyp uuid pdin moov then 16 moof/mdat, mvex present), mfra on the clean stop, 1920×1080 @ 60/1, 948 packets = 15.8 × 60, ffmpeg -v error -f null - exit 0, both sidecars written. 286.3 s: 287 moof, mfra present, 17,180 packets, decodes clean, .cursor.json 1.3 MB. No pacing drift and no dropped frames over 4 min 46. Export passes and honours its settings: 720p/30 requested from a 1080p60 source gave 1280×720, avg_frame_rate 85900/2863 = 30.004, 8590 packets matching the frame count the progress UI itself reported, duration 286.333 s identical to source, decodes clean, 124.5 MB, written to the path chosen in the native save dialog and reported back as "Saved to …". Composition verified by extracting a frame and reading it at full resolution (not from a preview screenshot): gradient background, content inset as a rounded card with a drop shadow, content aspect ≈1.76 against the 16:9 target, synthetic cursor drawn. Note the exporter adds a silent AAC 48 kHz stereo track even though no audio source was enabled. Retracted: "the HUD language menu ignores Escape". It does not — the maintainer confirms the key works by hand. Claude Desktop swallows Escape before it reaches the app under test, so a synthesised press proves nothing about the app, and GetForegroundWindow() returning the HUD does not rescue the inference: the key never left the driver. The companion observation (an outside click on the HUD's own drag handle did not dismiss the menu) is withdrawn with it, since the HUD's own chrome is not "outside" the popover in any meaningful sense. What is established is that the blur path shipped in this RC works: 54e12706 fix(hud): dismiss the HUD popovers when the window loses focus dismissed the menu on a click to the desktop. Rule for anyone driving keyboard checks from computer-use: Escape is unusable as evidence, and any negative keyboard result needs a by-hand confirmation before it goes in this table. Behaviour vs doc: the record button is not disabled without a source — it opens the source selector. No recording starts, so the check's intent holds, but AGENTS.md still describes a disabled button with a "Please select a source to record" tooltip, and that is why no tooltip appears. Passed: single launch window, no startup crash; HUD visible under OPENSCREEN_DISABLE_CONTENT_PROTECTION=1; tray layout toggles horizontal↔vertical both ways; HUD drag follows the pointer without drift and stays at the drop point; language menu opens with its locale list; minimize hides the HUD without quitting (6 processes still alive); relaunching routes through the single-instance lock, restores the window and mints no duplicate; source selector opens, selecting a card enables Share, and the HUD label becomes the picked source (Tout l'écran); record → stop opens the editor with the asset, a timeline clip and a rendered preview; About reports the RC version. Local transcription works, on GPU — an earlier draft of this row reported it broken, which was wrong. Relaunching with stdout/stderr captured and importing a 15 s asset that carries an audio track settles it: [whisper-stt] boot: model=…\whisper-ggml\ggml-small-q8_0.bin host=127.0.0.1 port=64720 threads=16, ggml_vulkan: 0 = NVIDIA GeForce RTX 4070 Ti, model loaded; backend=whispercpp-vulkan, then [stt] done on whispercpp-vulkan: 1 chunk(s), 15.0s audio in 0.1s (0.01 rtf, 106.8x real-time). The pane switched to "1 caption lines, derived live from the transcript". The real (minor) defect is the error message: on an asset with no audio track the captions pane says "Failed to fetch", which reads as a network failure and sent this run hunting a broken STT server that was never involved — the pipeline simply has no audio to extract. It should say so. Second minor find, from the same stderr: listProjects cannot read three saved projects — one ZodError (transcript.segments[0].endSec must be greater than or equal to startSec, repeated across segments, words and transcripts[0]) and two SyntaxError: Unexpected non-whitespace character after JSON, i.e. truncated or double-written project files. They are skipped silently in the UI. Caption anchoring — the rc.2→rc.3 delta — is present but its rendering was not measured. The Position section carries exactly the model those commits describe: Bottom/Top, the note "Long captions grow upward — the bottom edge stays put", Distance from bottom defaulting to 1.5 %, and Left/Center/Right. What could not be checked is where a caption actually lands, because the only transcript obtainable here came from a 300 Hz sine and yielded one line that never surfaced at any scrubbed position. Closed out of band: the maintainer ran the caption sections by hand on a real spoken-audio recording and reports them correct, which is the coverage this automated run could not supply and the last gap standing between this RC and a promote. Also confirmed from stderr: [content-protection] OFF for the HUD window (OPENSCREEN_DISABLE_CONTENT_PROTECTION=1), so the flag does log its effect, and with the flag unset the HUD is correctly invisible to screenshots. The consequence matters more than the cause: the eight caption anchoring/margin/inset cherry-picks that are the entire delta from rc.2 to rc.3 are NOT covered by this run. Not run: restart and cancel actions; audio capture of any kind; webcam PiP; GIF; DPI scaling; HUD/notes exclusion from captured video with content protection ON (the whole session ran with it off, and the exported frame confirms the HUD is captured when it is off); regions, modifiers, timeline navigation, clip operations, persistence; macOS and Linux. Environment limits that shaped this run, worth knowing before the next one. parsecd.exe runs elevated and holds an invisible always-foreground window (ParsecMinFrameRate16); the moment OpenScreen loses focus every computer-use click is refused, and because the process is elevated UIPI makes granting Parsec useless — tray-icon refocus could therefore not be tested at all. Relaunching the app (single-instance raises it) is the way back. Dragging the HUD only works while every intermediate pointer position stays inside the HUD's own 904×698 mostly-transparent window; as soon as one lands on the desktop, the tier-"click" shell gate refuses the drag mid-gesture and leaves the button down — release it explicitly. Finally, the Microsoft Store package (EtienneLescot.OpenScreen, 1.9.6) shadows the NSIS install in request_access: every grant resolved to the Store bundle and the RC window stayed masked in screenshots while reporting success, until the Store package was removed. Screenshots do not interrupt a recording — that hypothesis was raised and disproved by running a 90 s capture with none taken and then taking one mid-capture with the helper surviving.
2026-08-22 installed v1.10.0-rc.3, macOS Apple Silicon DMG from the GitHub release (Openscreen-macOS-Apple-Silicon-1.10.0-rc.3.dmg, via Safari), unmodified: CFBundleShortVersionString 1.10.0-rc.3, spctl accepted, source=Notarized Developer ID (M4LK7C6S84), the only copy of com.etiennelescot.openscreen on disk macOS 26.5, M1, one display (captures at 1920×1080; scale factor not re-measured, the 2026-08-14 row records this display at 2×) Partial: no defect Computer-use pass, mostly on the caption anchoring that is the whole rc.2 to rc.3 delta. Restored here: the row was written in #486, then lost when a merge of main into that branch dropped it, and #486 was closed as already logged, which was true only of the 2026-08-23 window-capture slice below. Captions, on an imported clip of real speech (say, about 52 s, muxed to MP4, imported through New Project): auto-transcription fired on import (English detected, 26 lines, near verbatim), and every placement was measured on exported frames (crop=1:1, rawvideo), never the preview. Bottom at the 1.5 % default: plate bottom at y=1063/1064 of 1080 (16 px, 1.48 %), pixel-identical for two single-line captions 741 px and 999 px wide, both centred at x≈960. Top at the 1.5 % default, 9:16 (needed to force wrapping, since 16:9 phrases never exceed 7 words): plate top at y=29 of 1920 (1.51 %), identical for a 2-line and a 3-line caption, which grow downward (y 29 to 251 vs 29 to 353). Left: plate edge at x≈108 of 1080 for both. Recording: a clean stop (second click on the record/timer control) saved recording-1787474633337.mp4 (227.4 MB, 27.1 s, ftyp mdat moov), .cursor.json and .session.json, made a project and opened it in the editor; a SIGKILL of the helper mid-recording still left a decodable fragmented file (13 moof, 14.0 s), so the #375 fragmenting holds. MP4 export at source 1080p and at 9:16 both completed. Driver note, not a defect: a respawning UserNotificationCenter took OS focus and blocked synthetic clicks until killed before each click; the Cancel recording (X) icon discards by design, and the stop-and-save control is the record button itself. Not covered: Right alignment and non-default top/bottom distances (not measured), #418 HiDPI frame fill, window capture (see the 2026-08-23 row), webcam and microphone (none attached), GIF, AI features, persistence across reopen, Windows and Linux.
2026-08-23 installed v1.10.0-rc.3 (Developer ID, unmodified) run with OPENSCREEN_SCK_CAPTURE_EXE pointed at a helper built from this branch macOS 26.6.2 (25G83), M1, 1728×1117 @ 2× Pass — fixes a blocker Window capture section only. Before: selecting any window in the source picker kills the helper the instant start() builds its filter — Assertion failed: (did_initialize), function CGS_REQUIRE_INIT, file CGInitialization.c, line 44, SIGABRT, -[SCContentFilter initWithDesktopIndependentWindow:] → SLSGetDisplaysWithRect. 6/6 attempts on the shipped rc.3, no file, no error surfaced in the UI (the HUD returns to idle as if nothing happened). Display capture is unaffected and always worked, which is why this went unnoticed: the two paths diverge at makeCaptureTarget, and only the window branch resolves a rect through SkyLight. After: record → 25.2s → stop → editor opened on the take, recording-1787475175449.mp4 12,559,123 bytes / 25.18s / 2674×1684, the MP4 and both sidecars written (.cursor.json, .session.json), one project minted, zero crash reports. Helper-level A/B on an identical request JSON isolates the change: shipped signed helper → assertion, no file; this branch's helper → recording-started/recording-stopped, 4.49s / 1336×840 decodable MP4. NOT covered: webcam PiP, microphone, system audio (all off for these runs), export, GIF, AI/transcript sections, Windows, Linux. Not covered by unit tests either — Package.swift scopes the Swift test target to what runs without a screen, a display server or a TCC grant, and this crash needs all three.
2026-09-03 dev build, worktree github-issue-385-38d731 @ 437e4bd2 (main). Natives rebuilt from this tree: wgc-capture.exe and compositor_view.node both dated 2026-09-03, verified by string probe ([segmentation] HIT / control OPENSCREEN_EXPORT_ENCODER HIT; helper control encoder-selection HIT). ONNX Runtime 1.27.1 staged. Windows 11 26200, 1920×1080 @ 100% Partial — no defect, one validation unresolved Post-1.10.0 slice. Passed: hardware H.264 is the Windows default (videoEncoderRuntime: "hardware" on the plain path); recording AAC valid at 48 kHz; camera background Original/Cutout/Blur/Custom all render in preview and in the exported frames; the control is correctly hidden when onnxruntime.dll is removed and returns when it is restored; caption anchor presets Top/Left move the band and relabel their sliders, and the exported plate's left edge measured x=192/1920 = 10.00% against insetX: 10; crop field down-arrow steps one source pixel (100 → 99.9479) and the preview box carries the source aspect; a zoom lying entirely inside a trim fires in the preview with the playhead parked on it (zoom 19.956–22.927 inside trim 19.931–23.596, checked in the saved project, not by eye); export progress reached 100% with a 1.5× speed region present and reported a real frame total; exported audio and video durations agree within one frame at the export's 30 fps — 41.200 s audio vs 41.167 s video, a 33 ms gap that is exactly 1/30; editor-window.json persists bounds and a hand-planted zero-size/non-boolean state is rejected on relaunch; Regenerate-as lists all 101 whisper language entries; regions and settings survive a restart. Skipped: Ctrl/Shift+scroll zoom and pan — the computer-use scroll action does not carry a modifier, so the fix that moved the wheel listener to the whole pane could not be exercised either over the ruler or over the lanes; tray context menu (Save Diagnostics, Update Settings) — the desktop shell is granted at tier click, which blocks right-click; Update Settings in general — dev build, app.isPackaged is false so the submenu is correctly absent; the illegal-AAC-rate snap end-to-end — no 96 kHz device on this machine, covered instead by the helper's own audio_sample_utils_test MF probes, which passed at build time. Note, not a defect: Escape does not reach the app through this driver (it failed to close the Edit Clip dialog too), so any Escape-based check here is untestable rather than failing — same conclusion as the rc.3 retraction. Unresolved — do not read this row as clearing it: the exported duration ran 0.48 s (≈13 frames at 30 fps) under the trim/speed arithmetic — 1236 frames against a predicted 1249, from source 46.300 s minus a 3.665 s trim minus 0.990 s saved by a 2.971 s span at 1.5×. Ruled out: the source being shorter than its container claims (-count_frames gives exactly 2778 = 46.300 s × 60 fps). Not reproduced under controlled conditions, and not attributable to any post-1.10.0 change — the trim/speed length arithmetic predates them — so it is logged as an open question rather than a defect against this slice. The silence check in the same run is also weaker than it looks: that recording was ambient-mic screen capture, i.e. mostly silence, so it could not have distinguished an atempo hole from the source's own quiet. Both are why the two checks above now demand controlled audio and frame-level arithmetic.
2026-09-04 signed follow-up candidate 256ec082 on feat/fix-av1-webm-clip-switch, packaged from a raw-Git-blob build tree (Openscreen.exe 32469490…3E6E, compositor_view.node 090B606C…AA3A) Windows 11 Pro Insider Preview 26340, 1707×1067 logical Partial — no defect AV1 clip-switch section only, driven by real computer-use against the packaged app. Three-clip synthetic project: H.264 blue 0–1.50 s, malformed AV1 1.50–1.54 s (AV1CodecConfigurationRecord version 127, avg_frame_rate=0/0, unknown-sized EBML, and a nonzero source start of 0.400 s), H.264 green from 1.54 s. Real pointer control throughout — hover then click on the click-through HUD's Open Studio, ruler clicks and the previous-clip control to seek, the play control to run. Seeking into the malformed AV1 clip renders that clip's own frame rather than a neighbour's: the preview turned red at 0:01.5, against blue at 0:01.2 and green at 0:02.3. Read those colours as the three fixtures' own identities — blue (0,0,255), AV1 (254,0,0), green (0,255,1) at source 0.400 s, measured from the files, not sampled off the screen: no screenshot from this session was retained, so the on-screen colours are a qualitative reading and the identification rests on the clip-switch log below. Playback from 1.2 s crossed both boundaries and reached the third clip (green at 0:03.4) before stopping at the end of the timeline. The app's own stderr binds that traversal to the fixtures in order (h264-before-blue.mp4 → legacy-av1-v127-zero-rate.webm → h264-after-green.mp4) and shows the malformed traits being exercised live — 38 × Unknown version 127 of AV1CodecConfigurationRecord found! plus Unknown-sized element at 0x3cd inside parent with finite size — with no panic, fatal or abort line, the process still logging and still accepting UI input at the end of the pass, and closing Studio from the UI ending the app. (Liveness was polled during the run, but those polls were not persisted; what is checkable afterwards is the log and the fact that every later click landed.) The AV1 clip is 40 ms, narrower than the 0.1 s playhead readout, so the AV1 frame is identified by clip-boundary navigation and by the logged clip switch, not by the readout alone. The fixtures are synthetic, not a reporter-supplied file. No Playwright ran in this pass — the earlier Playwright run against the same package is supplemental and is not manual evidence. Exit code not captured: the waiter script threw before recording it, so "normal exit" rests on the UI close, the process disappearing and the absence of any Windows error event for the app. Not run: capture/HUD recording flow, webcam, microphone, system audio, export, GIF, transcription, editor operations beyond seek and play, macOS and Linux. Environment note for the next run: request_access resolves only Start-Menu-registered applications and refreshes that index on a delay, so an unpacked build needs a temporary shortcut and a retry before it can be driven at all; and re-fronting through that shortcut launches a duplicate instance on the default profile instead of raising the isolated one.
2026-09-04 three packaged arms differing in exactly one file: hashing all 181 packaged files shows resources/electron/native/bin/win32-x64/compositor_view.node is the only path that differs. PR base 9f64ae6d built for this run as CDA2618F…164B; an older pre-fix addon from an unrelated branch build (18B44DFB…0CDA, 2026-09-02); this branch's 090B606C…AA3A. Shared Openscreen.exe 32469490…3E6E and project B915113B…3E93 Windows 11 Pro Insider Preview 26340.9233 Pass — reproduces and clears the synthetic repro of issue 554 Scripted, not a manual OS-input pass, and each arm ran once. All three arms were driven by the same Playwright script against the real packaged Electron process, so this row is a controlled A/B and does not replace the computer-use row above. The base build crashes where this branch's build does not. Playing H.264 → malformed AV1 → H.264, the PR-base arm terminated with exit code 3221225477 (0xC0000005, access violation) without reaching the third clip. The older pre-fix addon advanced its playhead to 1.907 s but still showed the first clip's blue [0,15,255,255] at 1.915 s, then terminated with the same 0xC0000005 before a screenshot could be taken. This branch's arm crossed at 1.908 s, rendered the third clip's green [0,216,0,255] (retained as a screenshot), and closed with exit code 0. All three arms logged Unknown version 127 of AV1CodecConfigurationRecord and reached the AV1 clip's cursor sidecar, each recording loaded=FAIL; those [live] lines are native-compositor output, so the addon was live in every arm. Limits: the media is the synthetic fixture matching the reported AV1 configuration, not the reporter's unavailable recording, so this clears the reproduction rather than the reporter's file; interaction was script-injected rather than real OS input; the host is build 26340, not the reporter's 26200; n=1 per arm against a memory-corruption crash; and the post-playback canvas probe reads a different, smaller canvas element than the pre-playback one, so only the green frame has a screenshot behind it.
2026-09-05 Windows single-frame decoder ownership follow-up; source SHA-256 FAF87561...87117, bundled addon 94bc16d7...43a2c, app.asar 8063a377...96f04 Windows 11 build 26340; packaged Electron 41.2.1 Scoped pass: nonzero AV1 seek and cross-clip playback Fresh package, unique profile, no addon override; all 15 packaged native files hash-match staging (the existing packaging filter excludes standalone ffmpeg.exe). Native keyboard opened Studio after HUD pointer targeting failed; Studio seek/play/close used native input. AV1 at timeline 1.500 s/source 0.400 s rendered red [255,24,0,255] with a retained screenshot. During native playback from the first H.264 clip, read-only CDP samples observed blue followed by third-clip green [0,216,0,255] at 1.571 s; native logs confirm the malformed AV1 and following H.264 sources. PID 35472 exited 0 through window Close. An earlier sandboxed launch failed during Chromium GPU startup and is excluded from this pass. Limits: synthetic media, two native playback attempts in one successful session, no repeated stability or reporter-file claim; capture/webcam/tray/export not run. Exact-endpoint 4.540 s and one subsequent seek showed first-clip blue after native free-run wrapped: unchanged clip-identity synchronization paths, outside this resource/preflight follow-up; endpoint correctness is not claimed. Historical A/B rows remain separate.
2026-09-05 installed v1.11.0-rc.1, CI-built DMG from build run 33958219380 (openscreen-mac-arm64, release/1.11.0-rc.1 @ 9f30f1cb; the tag was later re-cut to 5d9e8703 with #618), Developer ID signed and notarized (spctl: accepted, Notarized Developer ID), ticket stapled; About reports 1.11.0-rc.1 (dmg), Electron 41.2.1; native payload one uniformly stamped set (33 darwin-arm64 files, Sep 5 11:36) macOS 26.5 (25F71), M1, 1920×1080 @ 1× plus a 2304×1296 @ 2× pass; no camera, no microphone Partial: 1 blocker, 4 minor Real OS mouse and keyboard through computer-use. Written at the time on claude/e2e-results-v1-11-0-rc1, which was never opened as a PR. Blocker (#616, fixed by #618): packaged STT is dead: the .app ships the ffmpeg dylibs but no ffmpeg executable, every stt:transcribe fails with FfmpegUnavailableError: stt:native-extraction-unavailable, and the UI says only "Failed to fetch" (a cached transcript can still show "Transcript ready" beside it). The Whisper helper itself loads on whispercpp-metal; relaunching the unmodified build with only OPENSCREEN_FFMPEG_PATH set made it work (65.5 s in 19.3 s, 0.29 RTF), so transcript, captions and word editing below ran with that variable. Regressions not reproduced: #375 (65 s take with system audio finalised: video 65.491 s / audio 65.530 s, -12.8 dB mean, ftyp mdat moov, all sidecars); the rc.3 window-capture CGS_REQUIRE_INIT crash (window 1468×854, 1161 frames, no crash reports); #418 at a real 2× (screen 2304×1296, four edges 100 % non-black; window 2304×1058). Passed: HUD layout toggle, drag (tracks the pointer, clamps at the menu bar, stays dropped), locale menu, device panel ("No microphone/camera found"); source picker, record, pause (timer froze at 00:31 for 16 s), stop to editor in under 10 s; play/pause/seek, ruler seek, navigator zoom/pan; z/t/s/a regions, text annotation with Typewriter, blur Mosaic/Gaussian/Oval obscure; export frame math (run A 1883 vs 1882.53 expected, run B at 2× 1872 vs 1872.11; audio and video within 0.98 and 0.99 frames at 30 fps), no silencedetect hole at either speed boundary with the level flat at -9.7 dB, caption plate left edge x = 192 for insetX 10 and top edge y = 16 for 1.5 % (measured on raw frames); GIF 938 frames 1280×720, loop count 0, no LZW corruption, decodes in ffmpeg and Preview; MP4 in QuickTime; packaged word-insertion gate (typing between words inert, a correction propagates to captions); About, Save Diagnostics, Check for Updates ("up to date (1.11.0-rc.1)"), tray Update Settings, 3 modes persisted across restart ({"mode":"download"}); shortcut rebind Z to X; every modifier and composition setting survived quit and relaunch. Minor defects: (1) the diagnostic bundle carries no logs (recentLogs, mainProcessLogs, helperOutput empty); (2) the blur inspector has no intensity/block-size control (the model stores 12/12) and no freehand shape; (3) a display-resolution change strands the HUD off-screen until restart (the editor window re-clamps, the HUD has no persisted position); (4) GIF at 15 fps writes a 6 cs delay, 16.67 fps, 56.28 s for a 62.40 s programme. Not defects, checked: export silence at 30.1999 s is the source's own silence through the trim; timeline.speedRanges: [] is schema layout; 58 non-monotonic-DTS lines are an ffmpeg null-muxer artifact (0 in ffprobe); an "unrendered" annotation was white on white. Skipped: webcam, microphone, voice-over, a second display at another scale (no hardware); Escape, modifier+scroll zoom/pan, hover controls on transcript words and silence chips (driver limits); AI chat, clip operations, imported audio, undo/redo, Background/Effects/Layout/Cursor facets, theme/locale switch, restart/cancel while recording, HUD minimize/close, Windows, Linux.
2026-09-07 locally built and installed v1.11.0-rc.1 from release/1.11.0-rc.1 @ 027f9e3c (rebased on main, includes #618). Apple Silicon DMG, Developer ID signed; local development artifact, not published or notarized. Verified payload includes executable ffmpeg 8.1.2, compositor, SCK helpers, Whisper/ONNX and FFmpeg dylibs; codesign --verify --deep --strict passes. macOS 26.5, M1 Partial: STT blocker fixed; 1 system-audio defect No-cache STT proof: opened a project whose asset had zero transcript segments, then used the installed build without OPENSCREEN_FFMPEG_PATH. UI progressed Transcribing 0% → Transcript ready; logs show Whisper Metal loading, native audio extraction and [stt] done on whispercpp-metal for the 0.2 s AAC asset (1.9 s inference). The project subsequently contains a new transcript asset id, language, one segment and one word. This rules out the cached transcript masking the former silent ffmpeg failure. A 65.5 s cached-project regenerate also completes normally (19.3 s / 0.29 RTF), but is supporting rather than decisive evidence. An asset with no decodable audio now exposes NoAudioTrackError in the UI instead of the previous FfmpegUnavailableError. System audio defect: with System audio ON and microphone OFF, recorded the entire screen for 20.38 s while a local 440 Hz / 48 kHz tone was actively playing through afplay (exit 0). The produced MP4 is healthy (3840×2160 H.264 plus AAC-LC 48 kHz stereo) but its audio is digital silence: AAC averages 2 kb/s and volumedetect reports mean/max -91.0 dB over 1,957,760 samples. Two shorter controls (0.55 s and 3.23 s) show the same silent track. Capture itself, source selection and clean stop work; this is an audio-content failure, not missing track creation. Continued editor slice: imported a local synthetic WAV; waveform, audio inspector, mute and loop all worked. Native Edit-menu undo/redo correctly changed then restored a Zoom modifier (keyboard shortcuts remain inconclusive under this driver). Theme toggled both directions; Cursor inspector toggles and controls were reachable; Edit Clip dialog opened with source preview and crop fields. Chat panel and AI settings opened, but no request was sent because the loaded project contains user recording/transcript. Webcam/voice-over remain unavailable: this Mac reports no camera or microphone. Remaining prior gaps: restart/cancel, clip drag/reorder/copy-paste, modifier anchoring under trims, second mixed-scale display, and Windows/Linux.
2026-09-07 Correction to the local v1.11.0-rc.1 retest above. macOS 26.5, M1 Pass: system audio Retracts the synthetic-tone defect. The afplay control captures digital silence on this machine, but is not representative of the ScreenCaptureKit route. A subsequent user-driven browser/YouTube capture with System audio ON produces a 99.46 s H.264 + AAC-LC 48 kHz stereo MP4 whose audio measures mean -29.3 dB, max -2.6 dB; the editor waveform visibly matches the signal. System-audio capture, source selection and clean stop pass.
2026-09-11 dev build (npm run dev) of fix/582-antialias-decimation, base fbe461e9, final commit 8d97f799. The 96 and 192 kHz A/B ran the helper from a clean npm run build:native:win of the tree 8d97f799 carries; the 384 kHz A/B and the 48 kHz app run used helpers built with the same command from earlier revisions of this branch, whose helper sources differ from 8d97f799 only in comments and whitespace. The A/B's pre-fix helper was built from fbe461e9 in a separate worktree. No binary hash is given; build from the commit SHA. Windows 11 Pro 26340, 2560×1600 @ 150% (as read after the runs) Scoped pass — no defect; the 96, 192 and 384 kHz (factors 2, 4 and 8) A/B ran and the alias is gone Anti-alias decimation (#582) slice. 48 kHz path, in the app, with real OS input against the click-through HUD: source picker → Entire screen → Share; system audio on (helper launched with system: { enabled: true }); record with countdown; pause at 00:32 (timer yellow, control switches to resume) and resume (red at 00:36); stop → editor opens with the take; playback of the saved recording runs, playhead advancing through 0:04.6 and 0:11.6. Recorded stream aac, 48000 Hz, 2 ch, 192000 bit/s, 38.38 s. The endpoint ran at 48 kHz, so nothing decimated on this pass: the 1 kHz reference measured -26.97 dBFS, and the 4 kHz and 12 kHz alias bins were empty (-128 and -125 dBFS). The 20 kHz probe component is absent (-136 dBFS), and an AAC encode alone removes it too: the same tone encoded to AAC 192 kbps by ffmpeg, with no capture involved, reads -131 dBFS at 20 kHz against -13.99 dBFS at 1 kHz — so a 20 kHz probe cannot test the capture path. Decimation path, A/B'd against the pre-fix build at three rates: a USB DAC was made the default render endpoint and its shared-mode format set to 96000, 192000 and 384000 Hz in turn, so loopback reports that rate, the encoder target stays 48000, and the factor is 2, 4 and 8; the helper's own audio-format event confirmed the rate in every run. The helper was driven the way the app drives it (config JSON argument, stop on stdin), not through the app UI. Fold-bin readings, before → after, for 96 / 192 / 384 kHz: a 36 kHz tone (folds onto 12 kHz) -22.25 → at most -115.32, -23.90 → at most -138.02, -24.28 → at most -111.54 dBFS; a 42 kHz tone (folds onto 6 kHz) -28.20 → at most -137.14, -30.40 → at most -135.09, -30.96 → at most -117.25 dBFS. The after readings bound what the recordings hold and sit near each recording's floor; they are not the filter's leakage, which the design puts at -123 to -155 dBFS at these points. The 1 kHz reference reads -13.98 or -13.99 dBFS in every recording; each pre-fix recording shows its alias only in its own fold bin; every pre-fix reading matches the box average's frequency response to within 0.08 dB; the 384 kHz runs were repeated and agree to 0.01 dB. AAC track length is not used as evidence about the filter's delay or about frame accounting (tracks are whole 1024-sample AAC frames, and repeat runs of the same helper differ by up to six); that output frame counts do not drift is asserted by the helper's suite. Beyond these recordings the behaviour is covered by that suite (95 assertions, including a real-AudioMixer case) plus eleven mutation counter-proofs and a control that must stay green. Check the Advanced tab's format list, and every endpoint, before concluding a machine cannot run this — a device's current format is not its capability. Not run: microphone, webcam, tray menu, export, and every non-Windows platform. A/V sync was not eyeballed: against the output timeline this change puts a decimated stream 39 + 1/factor output frames late (0.81–0.82 ms at a 48 kHz output; the filter's own group delay is 40·factor source frames), below what watching can resolve, so it is asserted numerically in the suite instead.

| 2026-09-13 | 6862a380 + microphone USB-label resolver patch; rebuilt release helpers | macOS 26.6.2, Apple Silicon, Shure MV7 | Partial | Real desktop HUD: selected explicit Shure MV7 (14ed:1012), microphone on, system audio/camera off; started/stopped capture and reached saved editor (16.1 s). Logs confirm the rebuilt helper path and decorated label. Raw AAC is non-silent (mean -37.4 dBFS, peak -13.3 dBFS). Separately, the resolver matched the decorated label to the real AVFoundation MV7 ID. Default input was also MV7, so audio alone does not prove non-default routing. Listening, mixed/system audio, export, non-default-input recording and other platforms were not run; this is not full acceptance. | | 2026-09-15 | dev build (npm run dev), worktree e2e-macos-pr-652-b93321 on pr/509-hud-session-prefs: main pass on 8fa0740a (PR #652 head before the scope narrowing), rescope smoke on ac36aab4 | macOS 26.5, M1 | Pass — 1 defect found, fixed on the branch before merge | Settings-persistence slice for PR #652, driven with real OS mouse/keyboard. Capture prefs persist across relaunches from both the HUD and the Rec stage — display source, system audio, mic toggle, cursor mode, auto-zoom — and recording-settings.json matched the UI after every restart (final smoke state: mic on, system audio on, editable-overlay cursor, autoZoom on, screen:1:0). A window source is restored after restart; when that window no longer exists, refocusing the HUD falls back cleanly to "Screen". Re-enumeration on HUD focus: no flicker, no permission prompt, no errors. New-project appearance defaults applied without touching existing projects (on 8fa0740a, before the narrowing dropped them). Defect: the HUD stop button did nothing — select source → record → stop pill left the take recording with no log line. Root cause: in startWhenDevicesReady, when every device is ready on the first check the async body never reaches an await, so its finally clears startWhenDevicesReadyInFlight before startWhenDevicesReadyInFlight.current = pending runs; the ref keeps a settled promise forever and every later stop (or countdown cancel) returns it. The existing tests all make the gate wait once, so none caught it. Reported on PR #652; fixed on the branch in e159ced0 (merged to main as 1df32fed). The fix was not re-verified in a dedicated run — the post-narrowing smoke on ac36aab4 ("drop the Settings dialog") confirmed prefs still persist and an existing project still opens, nothing more. Not covered: explicit mic/camera device restore and preview stop on reset (host has no camera, mic is default-only), the full capture-to-export flow, export, transcript/AI sections, Windows/Linux. | | 2026-09-15 | dev build (npm run dev), worktree editor-presets on feat/editor-presets (PR #675) | macOS 26.5, M1 | Pass — no defect | Style-presets header slice, driven with real OS mouse input. Confirmed the Presets trigger is in the editor header between the Media/Edit/Rec switcher and theme control, and is absent from the Composition pane header. Opened the menu from the new location and confirmed the built-in OpenScreen preset plus the empty user-preset state. Created E2E Header Smoke; the success toast appeared and the resulting .openscreenpreset file contained format/version metadata and appearance-only state. Applied the built-in preset and observed Applied “OpenScreen”; renamed the user preset to E2E Header Renamed and verified the file rename on disk; updated it from the current project and observed Preset updated; deleted it through the confirmation UI and verified the menu returned to No saved presets yet and the preset directory contained no files. Not covered: capture, media import, preview/export, transcript/AI sections, Windows/Linux, or preset portability between machines. | | 2026-09-16 | worktree trim-modifiers-visibility-771405 on fix/privacy-blur-follows-zoom @ cd927180; compositor_view.node rebuilt from this tree by npm run build:native:compositor (2026-09-16 10:26, contains the new privacy_mask shader comment), ffmpeg DLLs hash-identical to the vendored SDK; renderer rebuilt with npm run build-vite | Windows 11 26200, NVENC export | Partial — no defect | Privacy blur under zoom and 3D tilt, through the app's real export (electron . export <project> -o <mp4>). Source: synthetic 1920×1080 clip with a red box at 55–70 % on both axes. Eight legacy projects (rest, 2.2× zoom centred, 2.2× zoom at 0.75/0.75, 2.2× zoom + iso), each exported without and with a white mosaic blur annotation over the box; frame at 3 s. Red pixels still visible: rest 0 (clear 42 777), centred 0 (207 933), off-centre 0 (207 932), iso 0 (140 398); the masked iso frame shows the mask following the tilted box. Before the fix the same geometry leaked 89–100 % of the box (D3D11 pixel test privacy_blur_under_zoom.rs). Not covered: the GUI editor preview (the HUD rendered as a black rectangle in computer-use screenshots, so the pass went through the CLI export, which shares compose_frame with the preview), oval and Gaussian styles, motion-blurred zoom ramps (unit-tested only), macOS and Linux rendering (Linux lib tests pass under WSL/lavapipe; Metal only compiles on macOS CI). | | 2026-09-16 | worktree wf_62afabe4-4fb-7 on feat/3d-depth-of-field @ 7587de8e + the macOS golden and bench commits; compositor_view.node rebuilt from this tree by npm run build:native:compositor (string probe DOF_MAX_LOD HIT, control OPENSCREEN_EXPORT_ENCODER HIT), ffmpeg DLLs hash-identical to the vendored SDK; renderer rebuilt with npm run build-vite | Windows 11 26200, NVENC export | Partial — no defect | Depth of field on a tilted zoom, through the app's real export (electron . export <project> -o <mp4>). Source: synthetic 1920×1080 text-and-grid clip. Three legacy projects: iso zoom with focus at 0.92/0.08 and no depthOfField key (the default must apply), the same with depthOfField: false, and a flat zoom. Frame at 3 s: with the default the far (bottom-left) half of the tilted screen is visibly defocused and the near half stays sharp; with false the whole plane is sharp; the flat project renders a flat, sharp screen. Export wall time was the same with and without the effect (2.22–2.36 s over two warm runs each). Not covered: the GUI editor. Computer-use access to electron.exe was declined for this run, so the toggle on, the toggle off and the disabled state on a project without a tilt were checked only by the jsdom test depthOfFieldToggle.test.tsx. The preview cost was measured by the A/B harness instead (tilted_depth_of_field.rs, preview shape at 1280×720 + readback_direct: +0.03 ms/frame on hardware, +2.5 ms/frame on WARP). macOS and Linux rendering were not run here (the macOS golden runs on macOS CI; the Linux golden is opt-in under WSL). | | 2026-09-24 | installed v1.13.0-rc.4 — CI-built DMG Openscreen-macOS-Apple-Silicon-1.13.0-rc.4.dmg from the GitHub Release (tag 187e0d8d), Developer ID signed + notarized (spctl: accepted, Notarized Developer ID), About reports 1.13.0-rc.4 (dmg); native payload one uniformly stamped set (all darwin-arm64 files 2026-09-22 10:05) | macOS 26.5 (25F71), M1 Mac mini, one 3840×2160 display; no camera and no microphone attached | Partial — 8 defects, none blocking in what ran | Full-file attempt with real OS mouse/keyboard; everything not run is listed. Passed: launch/HUD (layout toggle, drag follows pointer and stays dropped, language menu, minimize → tray refocus, close/relaunch, no orphan processes after Quit); source picker (Screens/Windows tabs); record with countdown, pause freezes timer, resume, restart resets, cancel opens no editor; stop from HUD and from tray Stop Recording (tray icon turns red while recording); recording 3840×2160 H.264 + AAC 48 kHz filled edge to edge (measured: 0 dark px on all four borders); system audio captured (−10.9 dB mean); relaunched without the protection flag → [content-protection] OFF … macOS 26.5.0 logged, HUD visible on screen and absent from the recorded frame. Editor: rename, Media/Edit/Rec modes, chat panel, play/pause, seek bar + ruler seek, trim edges + playback skips the trim, zoom levels/3D camera (Off, 3 fixed angles, orbit) in preview, auto zooms from clicks, speed presets + custom 2.5× + 150 → "can't go higher than 100×" toast, text annotation (text/colour/background/typewriter), arrow (direction/stroke), blur (mosaic/gaussian/oval), delete + ⌘Z/⌘⇧Z, second clip by drag from Media, reorder (modifiers stay anchored to their clip, checked in the document), Edit Clip crop (↓ on W = 99.97395 = one px of 3840, partial "8" stays editable, 80 % crop reframes), clip delete leaves other clip's modifiers. Transcript: local Whisper on whispercpp-metal (4.5× RT), "no data leaves the device" notice, word click seeks, cue highlight, Backspace cuts a word (trim created), double-click correction propagates to captions and to the translation, typing between words is inert in the packaged build; captions on/off, Top/Left anchor with relabelled sliders, translation FR via provider then delete → original back. Chat (DeepSeek): 2-op turn applied and undone/redone as one unit, "cut the silences" explained instead of guessing, rewind cancel leaves everything, rewind confirm rolls back. Imported audio: pill named after the file, gain/fades/mute/loop controls, exported gain exactly −4.0 dB, fades at the track's own edges (1.45 s ramps), loop fills to the end with no seam, mute → track absent (noise floor), recording audio unchanged under the bed, track not time-stretched by a 2× region and cut at the last frame without pulling the fade forward, still audible in an export after an app restart. Export: MP4/GIF options + dimensions, progress reaches 100 %, frame math exact (take 2: 1086 − 55.5 trim − 18.5 speed = 1012.1 expected, 1013 written; take 1: 1032.6 expected, 1032 written; audio = video duration in both at 30 fps), no silence hole at speed boundaries (silencedetect, the one gap maps to a pause already in the source), caption plate left edge x = 192 for insetX 10 and top edge y = 16 for 1.5 % (measured on PPM), annotation/captions do not follow a zoom, 3D tilt + text + oval blur + cursor all in exported frames; GIF cancel returns to options with no partial file, full GIF 1280×720, 508 frames, infinite loop, opens in Preview; MP4 opens in QuickTime. Settings: background image/colour/gradient, padding/roundness/shadow, cursor hide + style, light theme (gradient picker + dialogs follow), French UI incl. app menu and wordmark menu, custom shortcut (T→Y works, restored), ⌘S/⌘O dialog. App menu About / Save Diagnostics (valid JSON bundle) / Check for Updates → "up to date (1.13.0-rc.4)"; same from the HUD panel (which shows version, "No microphone found", "No camera found"); tray Update Settings has the 3 modes and the pick survives a restart; editor bounds (resize, maximize, un-maximize persisted; zero-size + maximized:"yes" file → sane default window). Defects: (1) Preview only — a text annotation with an animation renders inconsistently during playback in the native preview (blank, partial "R", or complete across repeated plays of the same span) while paused seeks and the export are correct; related: the first paused frame after creating an annotation or reopening a project is sometimes stale until the next seek. (2) Clicking the editor behind the Export modal while the native Save Exported Video panel is up sends that panel behind the editor: invisible, modal stuck on "Starting…", its Cancel inert — recovered only by pressing the panel's Cancel through Accessibility; reproduced twice. (3) HUD dialogs (Check for Updates from the HUD panel) grey a 900×700 area of the desktop — already fixed on main by #733, not in rc.4: cherry-pick candidate. (4) GIF delay is truncated, not rounded (crates/compositor/src/gif_export.rs:231, 100 / fps): 15 fps → 6 cs → plays at 16.7 fps, 30.48 s for a 33.87 s programme (+11 %, not the ~4 % the comment allows); 30 fps has the same error. (5) Hovering a Backspace-cut transcript word shows no Restore button (undo and the trim inspector still restore). (6) Whisper appends a hallucinated repeat of the whole transcript over the trailing ~8 s of silence, stacked at 34.4 s. (7) The "frame-stepped and muted" hint (speed.previewFrameSteppingHint) is rendered nowhere at 20×. (8) Cosmetic: the shortcuts dialog lists Undo/Redo as Ctrl+Z / Ctrl+Shift+Z / Ctrl+Y on macOS. Checklist drift, not defects: no loop or fullscreen transport control exists (transport.loop key unused); blur has no intensity/block-size control (fixed 12/12) and freehand only for legacy docs; projects autosave, so the unsaved indicator and Save/Discard/Cancel prompt never appear; the app menu puts Save Diagnostics between About and Check for Updates. Skipped — no hardware: microphone (mic-only, mic+system balance, HUD mic prompt), webcam PiP/layouts/shapes/full-camera/shrink-on-zoom, AI camera background (no webcam track), voice-over, second display at another scale factor; single Space: record while switching Spaces; driver limits: Ctrl/Shift/middle-mouse timeline zoom/pan (modifier+scroll never lands), Esc on menus; not run: chat history/rename/delete/new-conversation switch, compact context, copy message, API-key entry and device-flow panels, auto-enhance AI path, update download/install/restart flows (no newer release), old-project migration and legacy caption annotations, region merge/repel/cross-boundary split, clip copy/paste, transcript silence-span trim, re-transcription, word insertion (dev-only), notes window exclusion, Intel/Store/Windows/Linux blocks. Main-process logs clean (no error lines) across both launches; defect screenshots kept locally. | | 2026-09-24 | installed v1.13.0-rc.6 — CI-built DMG Openscreen-macOS-Apple-Silicon-1.13.0-rc.6.dmg from the GitHub Release (tag 6eb86b17, build run 35990025043), Developer ID signed + notarized (spctl: Notarized Developer ID); native payload one uniformly stamped set (34 files, 2026-09-24 13:10). rc.6 = rc.5 (#733 backport) + cherry-picks of #748, #749, #750 (2 commits), #719 (2 commits), #666 | macOS 26.5 (25F71), M1 Mac mini, one 3840×2160 display; no camera, no microphone | Pass — risk-scoped slice, no defect | Scoped to what rc.5→rc.6 changed, not a full-file run (the full pass is the rc.4 row). Real OS mouse/keyboard; every level, duration and frame claim measured on exported files. #733 HUD dialogs: HUD device-settings → Check for Updates shows "up to date (1.13.0-rc.6)" with no grey rectangle over the desktop (rc.4: ~900×700 greyed). #743 file panels (widest blast radius — every save/open dialog): Export MP4 → Save Exported Video is now a sheet on the editor; three clicks on the editor behind the modal leave it attached and visible (rc.4: panel fell behind, modal stuck on "Starting…"); Cancel returns the modal to its options. Open Project → Browse files… and the audio-import picker (m) also open as sheets on the editor and complete. #744 GIF timing: GUI GIF export 15 fps / Medium: 517 frames, delays 7/6 cs alternating (345×7 + 172×6), playback 34.47 s = frames/15 exactly, against 34.40 s for the MP4 of the same project (rc.4: 30.48 s for a 33.87 s programme); CLI GIF 15 fps: 543 frames, 36.20 s = 543/15. #747: Keyboard Shortcuts → Fixed shows Undo "⌘ + Z" and Redo "⌘ + ⇧ + Z / ⌘ + Y". #719 CLI export of v2 projects: three hand-written v2 projects over a real take — with cropRegion {0.25,0.25,0.5,0.5} the CLI export on rc.4 ignored the crop (full frame), on rc.6 it renders the centre crop; cursorSize: 6 renders a cursor twice the default (3.0) size, compared frame-for-frame against the same project without the key; an out-of-range crop {-0.2,0.5,0.6,0.8} exports without error (clamped); the v2 crop project also opens in the GUI with the crop applied. #666 imported-audio gain: the Output level slider now goes well below −12 dB (set −37.5 and −39.0 dB); exported level of the 1 kHz bed by single-bin lock-in over 11–13 s: −31.1 dBFS at −4 dB (rc.4) → −65.9 dBFS at −39 dB (rc.6), Δ 34.8 dB for 35 dB requested, muted floor −96.2 dBFS. Smoke: record from the editor's Rec stage (tray icon red, stop from tray Stop Recording) → editor opens on the 13.23 s 3840×2160 H.264 + AAC take with an auto zoom → add a trim (0.68 s) → Export MP4 1080p30: 12.60 s (13.23 − 0.68 = 12.55 expected, within 2 frames). Main-process log: no error lines. Not covered: everything outside this slice (see the rc.4 row); #743 on Windows/Linux (the helper keeps owning the HUD there — unit-tested only); the native-bridge project save path was exercised only through Open; webcam/mic still untestable on this machine. Open, not in rc.6: #742 (animated annotation during playback, preview only), #745 (not reproduced), #729 (Whisper hallucination on trailing silence). | | 2026-09-24 | claude/zoom-speed-adjustment-fb8abd, the zoom-pace PR for #683, on d951e532; compositor_view.node rebuilt from its first commit cc88c423 (the later motion-blur fix at speed boundaries does not touch these measurements, taken with motion blur off) by npm run build:native:compositor (2026-09-24 22:17), ffmpeg DLLs hash-identical to the vendored SDK; renderer rebuilt with npm run build-vite | Windows 11 26200 | Partial — no defect | Zoom pace and speed regions, through the app's real export (electron . export <project> -o <mp4>). Source: synthetic 1920×1080 clip, a 160 px white square centred. Two legacy projects: plain at 1× with a 1.25× zoom on [2 s, 3 s] and a 5× zoom on [6 s, 8 s]; sped with the whole clip at 4× and the same zooms at 4× the source times. Zoom scale measured on every frame of the 60 fps exports (area-weighted width of the square). New addon: plain and sped match frame for frame within a maximum scale difference of 0.0001; 1.25× moves 2→98 % in 0.53 s, peak 0.9 e-folds/s; 5× in 1.12 s, peak 2.8 e-folds/s; each zoom reaches 98 % before its region starts and leaves when it ends. Same app build pointed at the 2026-09-13 addon (OPENSCREEN_COMPOSITOR_VIEW_NODE): 5× peaks at 13 e-folds/s at 1× and 31 at 4×, where the whole zoom-in takes 0.18 s. Not covered: the GUI editor preview (same plan_frame as the export, not driven), Full Camera under a speed region and chained zooms (unit-tested only), macOS and Linux rendering. | | 2026-09-25 | claude/ratio-auto-screen-studio-df6b2e (Auto format), tree of ace7af33 + 972f8cf3 before the default flip; the test projects state auto explicitly, so the flip does not change them. compositor_view.node not rebuilt (no Rust change in the PR): the 2026-09-24 22:17 build from zoom-speed-adjustment-fb8abd, its whole bin/win32-x64 directory copied aside and loaded through OPENSCREEN_COMPOSITOR_VIEW_NODE; renderer rebuilt with npm run build-vite | Windows 11 26200, NVENC export | Partial — no defect | Auto format, preview geometry and real export. Preview: the renderer in browser mode (vite-dev, ?windowType=editor), 4:3 fixture tests/fixtures/sample.webm with website/static/video/webcam.mp4 as camera, padding 50, geometry read from the DOM. Fixed 16:9 + top / bottom left 265 px empty per side against 42 px on top; Auto gave a portrait 259×418 frame, square 207×207 camera, 26 px on all four sides. Side by side, padding 0 → 100: frame 2.36 → 1.82, border 0 → 84 px, camera square throughout. Fixed 16:9 side by side gave a 4:5 camera and the block at 80 % of the width, which is the old formula's answer (computed, not re-measured on the old build). Export: electron . export <project> -o <mp4> --quality source, synthetic 1280×720 blue screen and 640×480 green camera, wallpaper #303030, padding 50, roundness and shadow 0, frame at 1.5 s parsed as PPM. Top / bottom 858×1258, camera 687×686, margins 86/86/85/86; side by side 1240×506, camera 405×406, margins 50 on all sides; PiP 1194×736, screen margins 74/74/74/76. Not run: the real Electron editor window (preview checked in browser mode only), the native live preview, the default flip on a fresh recording (unit tests only), a non-English Format menu, macOS and Linux. | | 2026-09-25 | worktree zoom-overflow-layouts-698eb0, claude/zoom-overflow-layouts-698eb0 on ff3c3f51 + the slot-mask change; compositor_view.node rebuilt from this tree (cargo build -p compositor-view-napi --release), baseline addon built the same way from ff3c3f51; ffmpeg DLLs hash-identical to the vendored SDK; renderer rebuilt with npm run build-vite | Windows 11 26200, NVENC export | Partial — no defect | The block layouts' screen slot masks the zoom and the 3D (overflow hidden), through the app's real export (electron . export <project> -o <mp4>). Source: synthetic 1920×1080 testsrc2 screen and 1280×720 bars webcam, 4 s. Legacy 16:9 projects, padding 50 %, a depth-4 (2.2×) zoom at focus (0.3, 0.6) on [0.5 s, 3.5 s], frame at 2 s. dual-frame: the baseline addon spills the zoomed screen under the camera and off the frame; the new one crops it to its rounded slot, camera and gap untouched. dual-frame + iso: the footage tilts as in any layout and the slot, which stays upright, crops it; where the tilted footage leaves the slot, the wallpaper shows and the shadow follows the visible part. picture-in-picture: the whole export is bit-identical between the two addons (PSNR inf on all 240 frames). D3D11 frames through live::Player, the preview engine, also checked vertical-stack, the right preset and the follow-cursor camera. Not covered: the GUI editor preview (same plan_frame, not driven), a window or device frame in a block layout (left unmasked), privacy blur and click impacts in a slot (unit-tested only), macOS and Linux rendering (WGSL validated with naga, Metal not compiled). | | 2026-09-26 | worktree motion-blur-padding-issue-11befb, claude/motion-blur-padding-issue-11befb on 1a669c22 (#794); compositor_view.node built from this tree (cargo build -p compositor-view-napi --release), baseline addon built the same way from 218bdbb0 (main); ffmpeg DLLs hash-identical to the vendored SDK; renderer rebuilt with npm run build-vite, plus a local, uncommitted pass-through of editor.frame in normalizeProjectEditor (the v2 CLI path drops it) | Windows 11 26200, NVENC export | Partial — one defect found and fixed | Motion blur of the framed screen, through the app's real export (electron . export <project> -o <mp4>). Legacy 16:9 project: phone frame, padding 50, motion blur 1, shadow 0.6, a depth-3 zoom on [3 s, 8 s]; sources: synthetic 1920×1080 testsrc2, then solid red and solid green to isolate the footage (red/green difference). Static frames: only encoder noise between the two addons. Ramps (60 fps frames 132 and 501): the baseline keeps the shell's rim and bezel razor-sharp around a smeared picture; the new addon smears them with it (87–109k px of shell and shadow change that are never footage). Footage outside the union of the sharp apertures at n−1 and n, +2 px: 1,185 px on one frame at first (a ~20 % tint over the bezel in the four output corners, where the phone overflows the output), fixed in 1a669c22, then 0 px over all 55 moving frames. Not covered: the GUI editor preview (same compositor path, not driven), window chrome and laptop/monitor frames on this machine (window chrome covered by the Linux lavapipe test), macOS (Metal not compiled here). | | 2026-09-26 | worktree motion-blur-padding-issue-11befb, branch claude/cli-export-documents on 218bdbb0 + this change; renderer rebuilt with npm run build-vite; compositor_view.node the one in the worktree's electron/native/bin/win32-x64 (loading, not rendering, was under test) | Windows 11 26200, NVENC export | Partial — no defect | openscreen export on a project saved by the editor, through the app's real CLI (electron . export <project> -o <mp4> --json). Before: a copy of a real schema-7 editor project (%APPDATA%/openscreen/projects/proj_741d2d06…, monitor frame, 2 zooms with a 3D camera, 2 trims, speed regions ×5 and ×2 in legacyEditor) failed at once with Project file is not a valid .openscreen project. After: exported in 34 s, 1920×1080 60 fps, 21.62 s against 21.58 s expected (38.25 s − 4.60 s of trims − 9.44 s and 2.63 s taken by the two speed regions); the frame at 2 s shows the zoomed, tilted screen inside the monitor bezel. info on the same copy: version 8 after the load-time upgrade, the recording found, 2 zooms, 2 trims (its speed count, read from legacyEditor.speedRegions, was fixed after this run and is covered by projectCommands.test.ts). pack refuses it with pack does not support projects saved by the editor yet. A legacy v2 project (synthetic 4 s testsrc2, one zoom) still exports: 4.00 s. The original project file was only copied, hash-identical afterwards. Not covered: pack for documents (refused, not implemented), captions (its runner still reads v2 only), a document whose media moved (relink is the editor's own, unit-tested there), macOS and Linux. | | 2026-09-26 | worktree fix-3d-cameras-left-right-02603d, claude/fix-3d-cameras-left-right-02603d on cbf85dc7 + this change; compositor_view.node rebuilt from this tree by npm run build:native:compositor, the rest of bin/win32-x64 copied from the installed v1.13.0-rc.6 (ffmpeg DLLs hash-identical to the vendored SDK); renderer rebuilt with npm run build-vite. Reference: installed v1.13.0-rc.6 (= v1.13.0) through its own CLI | Windows 11 26200, NVENC export | Partial — no defect | 3D cameras against v1.13.0, openscreen export on both builds, same synthetic 14 s 1080p grid and cursor sidecar (three held positions), zoom depth 2, hideCursor, motion blur 0. During the hold (2.5, 3.5, 7.5, 11.5 s), left here and iso in v1.13.0 differ only in a ~50×65 px box around the cursor sprite (0.05–0.09 % of pixels > 24; the cursor size changed since); same for follow-cursor. iso here renders byte-identical to left. right is the mirror of left: screen edge 1–2 px apart, row by row. With motion blur 0.35 the orbit differs while the camera moves (mean abs diff 9, 9 % of pixels): the mode-8 trail added on 25/09, kept on purpose. Not covered: editor preview, GUI, macOS and Linux, zoom ramps (their pacing changed with #683). | | 2026-09-26 | worktree openscreen-screen-mask-zoom-e21ce9, claude/openscreen-screen-mask-zoom-e21ce9 on 1d67754c (integration/demo-never-ugly) + the frame-container change; compositor_view.node rebuilt from this tree by npm run build:native:compositor, baseline addon and renderer built from 1d67754c in a scratch worktree; ffmpeg DLLs hash-identical to the vendored SDK; renderer rebuilt with vite build | Windows 11 26200, h264 export | Partial — no defect | Block layouts under a window or device frame: the frame is the fixed container, the footage zooms and tilts inside it, and the block lays the frame out beside the camera. Through the app's real export (electron . export <project> -o <mp4>). Source: synthetic 1920×1080 testsrc2 screen and 1280×720 mandelbrot webcam, 8 s. Legacy projects, default padding, a depth-4 zoom at focus (0.3, 0.6) on [1.5 s, 3.5 s] and a depth-3 iso zoom on [5.5 s, 7.8 s], frames at 0.2, 2.8 and 7.0 s. dual-frame (16:9) and vertical-stack (9:16) under window, laptop, phone and monitor. Baseline: the whole frame zooms and tilts over the camera, and at rest the laptop's deck and the monitor's stand run under it. New: the frame stays at rest, lined up with the camera on its body, deck and stand clear of it; the flat zoom and the iso tilt stay inside the aperture, over the device's black glass or the window chrome's fill, with square top corners under the title bar. picture-in-picture with and without a laptop, no-webcam with a laptop, and both block layouts without a frame: every decoded frame identical to the baseline (framemd5, 480 video frames plus audio). Not covered: the GUI editor preview (same plan_frame and computeCompositeLayout, not driven), follow-cursor inside a frame (unit-tested only), the Auto format with a frame in a block layout (its resting shape ignores the frame, so the camera is not exactly square there), macOS and Linux rendering (WGSL validated with naga, Metal not compiled). | | 2026-09-27 | worktree camera-size-zoom-balance-7e14d7, claude/camera-size-zoom-balance-7e14d7 on 7aa00935 (integration/demo-never-ugly) + the camera size change; compositor_view.node built from this tree (cargo build -p compositor-view-napi --release), baseline addon built the same way with the old 1/zoom curve; renderer built with vite build for each side, the baseline with the old 10–35 bounds; rest of bin/win32-x64 copied from a 26/09 build, ffmpeg DLLs hash-identical to the vendored SDK | Windows 11 26200, NVENC export | Partial — no defect | Camera size range and shrink on zoom, through the app's real export (electron . export <project> -o <mp4>). Source: a real 1920×1080 recording with its 640×480 webcam; legacy 16:9 projects, picture-in-picture bottom right, shrink on zoom on, a depth-3 (1.8×) zoom on [4 s, 10 s]; frames at 2.5 s and 7 s, camera boxes measured by difference with a camera-less export. During the zoom: default 25 %, 150×113 → 210×158 px; minimum 10 → 15 %, 60×45 → 126×95; maximum 35 → 50 %, 210×158 → 420×315, each equal to the computed rect. At rest: 378 and 540 px wide at 35 and 50 %. Not covered: the GUI editor preview and the slider (unit-tested: min 15, max 50), the square camera, macOS and Linux rendering (same plan_frame, not compiled here). | | 2026-09-30 | installed v2.0.0-rc.1 (CI artifact, sha512 matched latest.yml), installed over 1.10.0 | Windows 11 26200, 1920×1080 @ 125 %, AMD Radeon iGPU | Partial — 2 defects | Launch/HUD, source selector (window and full screen), record with pause/resume, stop→editor ~0.65 s, transport, seek, zoom/speed/trim/full-camera regions with undo/redo, autosave, imported audio track (mixed at the right place and gain), voiceover take, Rec mode, MP4 export (duration = source video stream − trim − speed gain, exact), GIF export (ffmpeg and Pillow decode byte-identical). Defects: HUD walks up on a vertical→horizontal flip (#951, fixed in rc.5 by #953); transcript word times VAD-compressed, ~−14 s (fixed for rc.3, #948). GIF export ~3 frames/s vs MP4 ~84 fps (#952). Not defects, re-triaged: the Whisper Vulkan boot crash came from a helper staged from an unrelated branch (409cf4d7); "HUD not topmost" is computer-use's own doing (a screenshot clears WS_EX_TOPMOST on the HUD); the 98 % progress stop was the Windows audio tail (#942). Not covered: installer UI (ran /S), updater, tray, cancel/restart, notes, reopen from library, AI chat, macOS/Linux. | | 2026-10-01 | installed v2.0.0-rc.6 (CI artifact, sha512 matched latest.yml), installed over rc.1 | Windows 11 26200, 1920×1080 @ 125 %, AMD Radeon iGPU | Partial — 1 defect (transcript) | Targeted at the Windows changes since rc.1. #951: two vertical↔horizontal round trips with real clicks, state read after each, HUD back at the same bounds (302,149). Recording (full screen + mic + system audio + webcam, instant pause/resume mid-take): screen and webcam H.264 High, BT.709 primaries/transfer/matrix, TV range; audio runs 28 ms past video (rc.1: 0.87 s); stop→editor ~1 s. Screen at 57.2 fps average — 181 missed 60 Hz ticks over 52 s, timestamps stay on the clock (video 52.17 s vs audio 52.20 s) — known, #945. STT: helper boots on Vulkan, VAD on CPU, no crash; on a ground-truth clip (12.12 s silence then speech), first word at 12.10 s and every sentence boundary within 0.12 s; the "en" CTC aligner downloads in the background and only applies from the next transcription. Defect: the last word before one pause ("happened.") lands after it as a 20 ms word, its audio given to the previous word; without the aligner the display order also inverts ("nothing The happened."). Reported to #948. Export: 3-clip MP4 720p30, progress reaches 100 % (4200/4203), 4200 frames, video 139.97 s / audio 140.00 s, each clip's speech at the expected offset, decodes clean. Mixer (#911, instant resume): a second take with system audio only (mic off) while a 26 s TTS WAV played at 76 % volume (output peak meter 0.985 during the take). No punched hole anywhere in the voice: an alignment-free scan for 2–12 ms dips below 5 % between loud neighbours found none, and the same scan finds 10 ms holes injected into the take. Instant pause/resume (0.32 s button to button) skips 0.330 s of the source, so no voice before the pause is lost (the bug was 80–90 ms). Take is +3 dB over the source, peak −0.1 dBFS, 0 full-scale samples. The first take of this pass had the speakers muted at 0 %, which silences WASAPI loopback, so its audio was not usable. Not covered: preview limiter; MCP server; keep-awake; tray; GIF; macOS/Linux. | | 2026-10-01 | installed v2.0.0-rc.8 (build run 36883700385, tag 33a16fe1, helper from build-whisper-stt run 36830983393; NSIS sha512 matched latest.yml), installed over rc.5 for all users, About reports 2.0.0-rc.8 (win32 x64 · nsis, Electron 41.2.1) | Windows 11 Home 26200, 1920×1080 @ 125 %, Logitech StreamCam (sub-4K, MF) | Partial — 2 defects | Webcam: torn and mis-exposed frames in the webcam file (some frames split horizontally into two pictures, others blue-dark or washed out). The maintainer attributes them to the camera itself, not OpenScreen; the same frames are in the rc.1 take of 2026-09-30. Defect 1 (#960): captions in the preview are stale after switching projects. Two clips of one asset, captions Bold top-left: after opening another project and reopening this one, clip 1 shows no caption or the wrong line (0:06 showed "we are recording the screen…", 0:12 and 1:05 nothing) until Show captions is toggled; saved data is intact. Defect 2 (#961): after a working session, closing the editor leaves OpenScreen running with no window (6 processes, tray only), seen twice: once after takes (HUD, Record mode, restart, cancel), once after only the project dialog, Edit clip and a chat turn. An instance that only opened the editor and closed it quit as window-all-closed intends, so some hidden window outlives those surfaces. Loudness in spec: MP4 export at the Audio facet's default output level (never changed in this project), −15.8 LUFS integrated, −1.4 dBFS sample and −1.4 dBTP true peak; the check asked for a sample peak under −1.5 dBFS, which the AAC overshoot past the −1.5 dBFS limiter cannot meet, so its wording is corrected in this change. Passed: HUD layout toggle both ways, tooltips idle and recording (with continuous real moves; a teleported cursor shows the previous control's tooltip, a driver artefact), language menu, Device settings (devices, live preview, Camera quality 1080p/1440p/4K, 4K default, 1080p persisted across quit/relaunch as camQuality, hand-edited 720p reads back as 4K, device switch persisted), Hide recording bar → tray click brings the HUD back, Quit OpenScreen (0 processes), second launch exits 0 with one HUD. Source selector (Screens/Windows, Cancel), record with no source opens it and starts after the pick, countdown, pause (span absent from the file) / resume, restart (new helper process, timer reset), cancel (no editor, no file), gear locked during a take. Stop → editor, 53.0 s take with 3 automatic zooms at the clicks, desktop icons hidden in the second take for its whole length. Files: screen and webcam profile=High, has_b_frames=0, tv/bt709×3, AAC 48 kHz, encoder-selection hardware. Auto-transcription after the take; Regenerate as English. Transcript: clip order, silence spans with durations, monotonic word seeks (4.0 → 11.2 → 16.4 → 21.3 → 21.8 → 27.2 → 56.9 s), live word highlight, Backspace marks skipped + restore, silence-span trim and restore, #959 checked: sentence-final "time." cut 11.91–12.51 against a speech offset of 12.44 (silencedetect −40 dB) and next speech at 16.46; typing between words does nothing (packaged gate). Transport, Left/Right 1/60 s, stop at end, navigator narrow/pan. Regions by T/S/A/C, trim edges, Bring this part back, undo/redo; zoom levels, custom 1.25 kept, 9 → "Zoom goes from 1× to 2.86×", 3D Orbit/left/right; speed presets, custom 3×, 20 → "Speed can't go higher than 16×"; annotation text/size/plate; Full Camera. Clip reorder carries every modifier; Edit clip grip (0:10.7–0:53.0) and 1:1 crop, Cancel leaves it. Export dialog v2 layout and sizes (1224×720 / 1838×1080 / 1910×1122); MP4 1838×1080 30 fps, video 92.033 s vs audio 92.066 s (1 frame), 2762 frames vs 2760.7 computed (105.941 − 10.312 trim − 5.41 × 2/3 for 3×), frames match the preview at 16.5/28/40 s; GIF cancel (no partial file) then full GIF 816×480, 1383 frames at 15 fps, loop, ffmpeg and GDI+ agree. Imported audio: −18 dB, 1 s fades, ramps at the track's own edges in the export (3.0→4.0 s, 22.0→23.0 s). Captions toggle, Bold, top/left anchors and Distance from left, captions unscaled under a zoom. Camera layout presets with Auto reshaping, shape, mirror, Cutout. Cursor facet style and Show cursor. Wordmark menu, About (version, Copy to clipboard), light/dark theme, French and back, shortcut rebind (Q adds a zoom) and reset, Ctrl+S toast, Ctrl+O, an older project opens without prompt, reopen keeps regions/settings/audio track. Editor bounds and maximized state restored. Chat (deepseek-v4-flash): "Add a 2x zoom from 0:07 to 0:09" applied as 2.20× with a reason, one undo/redo unit. Not run: HUD drag (the HUD window now ends ~45 px above the bar, any longer move lands on the click-tier desktop; short in-window moves did not move it, inconclusive); Esc (Claude Desktop swallows it); Ctrl/Shift+wheel (the scroll tool holds no modifier); tray right-click menu (Stop Recording, Save Diagnostics, Update Settings, About from tray) and the Alt-held Help menu; level meter and mic-only/system-only takes, music ducking (nobody spoke); "No camera found"; 4K-camera, no-30-fps, DirectShow, covered-lens and sharpness checks (no such camera / setup); window source and 24H2 context menu; DPI and multi-display (single display, scaling not changed); voice-over recording; region copy/paste and merge/repel, clip duplicate/delete; translation; most AI-settings and chat-history checks; camera background export of Cutout/Custom; caption plate measurement; legacy-project migration checks beyond opening one; macOS and Linux. Also seen: an English text read by the French SAPI voice auto-transcribed as French (detection, not app). | | 2026-10-01 | installed v2.0.0-rc.8, CI-built DMG Openscreen-macOS-Apple-Silicon-2.0.0-rc.8.dmg from the GitHub Release (tag 33a16fe1, build run 36883700385, sha256 7a926621...8e99), installed over rc.7, Developer ID signed, notarized and stapled (spctl: accepted, Notarized Developer ID), About reports 2.0.0-rc.8 (dmg); native payload one uniformly stamped set (34 files) | macOS 26.5 (25F71), Mac mini M1, 3840×2160 @ 2×, no camera, no microphone | Partial: 1 should-fix, 4 minor | Real OS mouse and keyboard through computer-use against the installed release (an rc.7 pass was not run: computer-use access was denied while nobody was at the Mac). Defect (#965, fixed by #971): the HUD is recorded into every take after an editor round trip in a picker session: excludedWindowIds [7408] then [7458], take C frame at 10 s shows the HUD with its 00:09 timer; makeCaptureTarget(picked:) reuses picked.filter. Minor: #966 the lane says "Press Z" after zoom was rebound to X; #967 with no camera the HUD toggle saves camEnabled: true, so Record mode shows Camera On; #968 raw Error invoking remote method 'stt:transcribe' after a 60 s ffmpeg timeout behind a pending Downloads TCC prompt; #969 the French wordmark menu wraps and splits the version as 2.0.0-/rc.8. One unexplained app exit, not recurring. rc.1 blocker gone: 2 clips + trim + 1.5× speed + zoom + annotation exported 5,123 frames against 5,121 computed (85.351 s), audio and video end together at 85.38 s, silencedetect tone continuous across the cut and both speed boundaries. Silent system audio not reproduced (-17 to -20 dB mean); a pending Accessibility grant no longer blocks recording. Passed: HUD layout, drag within 2 px, tooltips, language menu, Device settings, Hide recording bar and menu-bar restore, Quit, second instance; menu-bar menu and About; picker-less record, pause (00:17 held 6 s), restart, cancel, stop to editor, Record mode handover; capture H.264 High 3840×2160 bt709, keyframe 0.983 s, about 30 Mbit/s; 3 automatic zooms at clicks; transcription on whispercpp-metal (46.1 s in 23.7 s), French punctuation (the rc.8 change), word seek, skip, trim silence; captions; editor regions, undo/redo, theme, French; rc.7 camera change (Square 70 %, Original 40 %), layouts, Full Camera, Cutout and Blur; MP4 1080p60 and GIF 778×480, 1,282 frames, loop 0; a schema 6 project opens and saves as 8, settings persist; AI chat zoom with undo/redo; Save Diagnostics with no key. Not run: Apple picker interactions (owned by Control Centre, not grantable; the operator clicked Share Entire Screen), Escape checks, microphone, webcam capture, voice-over (no devices), window source, multi-display, Spaces, notes exclusion, Permissions window, Ctrl/Shift+scroll, clip reorder, imported audio, translation, update flows. | | 2026-10-02 | installed v2.0.0-rc.12, CI-built DMG Openscreen-macOS-Apple-Silicon-2.0.0-rc.12.dmg from the GitHub Release (tag 5853f923, build run 37010880013, sha256 d76ee67c...ae15), app byte-identical to the DMG, Notarized Developer ID, About reports 2.0.0-rc.12 | macOS 26.5 (25F71), Mac mini M1, 3840×2160 @ 2×, no camera, no microphone | Partial (rc.8 to rc.12 changes only): 1 minor regression | Real OS input through computer-use; the operator clicked Share Entire Screen in Apple's picker twice. #965 to #969 verified fixed (take B after the editor logs the new HUD id [341] and the HUD is absent at 12 s, the picker is asked again by design; "Press X to add zoom" in English and French; camEnabled cleared to false; translated unreadable-media hint after chmod 000; French menu rows on one line). Regression, not filed: inside a 4× speed region the preview lags 1.8 to 3.2 s of source and stays 2.2 to 2.7 s behind after it, against 0.6 to 1.8 s on rc.8 run from its DMG on the same project; export unaffected; suspected e358d771 / b519f236. Preview at 1× on 1080p and 4K within 0.1 s, cut crossing, entry into a clip past its end shows its last frame, webcam PiP and Full Camera verified; read over CDP, the hidden screen <video> decodes 0 frames and the <audio> runs at 4×; closing the editor quits (#961), but the countdown-overlay path was not exercised. Also seen: transcription took 23 to 25 s while 4K clips were playing, against 3.3 s right after a take. Not run: everything outside the rc.8 to rc.12 changes, the Windows shared-texture transport, audible preview audio, camera and microphone. | | 2026-10-03 | v2.0.0-rc.13 CI artifact (build run 37072118778, tag 018d22ab, same crates/ electron/ src/ scripts/ as main at 0624c148) against v2.0.0-rc.12 and v1.11.0-rc.1, each run from its installer's extracted app tree; editor checks on a dev build of main with rc.13's whole bin/win32-x64 | Windows 11 Home 26200, Ryzen 5 7520U, AMD Radeon iGPU (D3D11, h264_amf) | Partial — no defect | Export-path slice for #982/#983/#984. cargo test -p openscreen-compositor --lib --tests on the real GPU: 456 passed, including the D3D11 render tests for the device frame, cursor model, click impact and cursor tap. Benchmark A/B in one session: rc.12 4.78× its floor, rc.13 1.71×, 1.11.0-rc.1 1.72× (numbers in rendering-performance.md, "The same fix on Windows"). rc.13's scoring exports are byte-identical to rc.12's. Twelve-second variants exported by both builds: the 3D cursor with click impact, the laptop frame with its shadow, the window frame, and a colour with Aurora are byte-identical; the two blurred-background variants differ invisibly (PSNR ≥ 67 dB). Frames extracted with ffmpeg show the device frame, 3D cursor, window chrome and rounded corners. Preview while the background is cached: image/colour/gradient switches, blur 100→0→50 and a window resize are all followed, Aurora keeps moving, None stays still, and playback frames show no leftover trail rectangle. Not run: real capture and HUD (computer-use screenshots came back uniformly grey, so the editor was driven over CDP, with injected input rather than OS input), Radeon GPU Analyzer register counts, Intel iGPU. | | 2026-10-03 | installed v2.0.0-rc.13 (NSIS from the release, sha512 matched latest.yml), installed over rc.12 for all users, About reports 2.0.0-rc.13 | Windows 11 Home 26200, 1920×1080 @ 125 %, AMD Radeon iGPU, built-in USB2.0 HD UVC WebCam | Partial — 1 defect | Scoped to the app changes between rc.8, the last computer-use pass, and rc.13; the export path itself is covered by the rc.13 row above. Real OS mouse and keyboard input throughout; computer-use screenshots came back uniformly grey again, so the app was observed through PrintWindow on its own windows and DOM reads over CDP. Defect: the webcam shows its first frame when a clip is entered past the webcam's last frame from another clip. Two clips of one take (screen video 50.98 s, webcam 51.24 s, audio and clip 51.31 s): scrubbing from clip 2 to 51.2 s in clip 1 shows the screen's last frame, as 227b9b84 intends, but the webcam's frame 0 (2 of 2; ground truth from ffmpeg: an empty chair from 18.7 s to the end). Scrubbing to the same point from inside clip 1, or to 51.0 s from clip 2, shows the right webcam frame. Cause: open_and_seek_clip and seek_pair in crates/compositor/src/live.rs seek the webcam with seek_to and fall back to seek_to(0.0), where present_frame uses seek_to_or_last; unchanged on main. Paused scrub only, in a ~70 ms window at the end of each clip. Passed: #961: one take with the countdown overlay, Record mode, editor work and an export, then closing the editor left 0 processes. #966: after rebinding Add zoom to Q and Add audio to J, the empty-lane hints read "Appuyez sur Q pour ajouter un zoom" and "Appuyez sur J pour ajouter un audio, V pour enregistrer une voix off", and Q adds a zoom. Wordmark menu in French: every row 31 px (one line), menu 261 px wide, "2.0.0-rc.13" in one piece. Preview: playback from 0:12 to 0:45.8 across a trim (0:15.0 → 0:18.5) and a 1.5× region with no canvas stall over 150 ms, 32–50 picture changes a second, wall time matching the edit; stops at 0:51.3 on the screen's and the webcam's last frames. #960 (#964): A → B → A with the playhead at 0:06, the preview is pixel-identical to before (mean difference 0, against 77.9 for project B). Background image → gradient → colour followed in the preview. #968: a media file whose read the OS refused (icacls deny on the test file) — Regenerate shows the translated "Impossible de lire l'audio de ce média…" hint inline and in the "Échec de la transcription" toast, without the IPC wrapper; the main process logs MediaUnreadableError … Permission denied. MP4 export from the GUI, 1752×1080 30 fps: 2930 frames, video 97.633 s against audio 97.666 s, zoom, background, webcam and cursor present in extracted frames, −18.3 LUFS. Also seen: the green "Aucune parole détectée" chip from the earlier successful run stays next to the unreadable hint; each clip of the take still carries 0.32 s of audio past its video (what is left of #942), and the export ends each clip at its video (97.63 s, against 98.22 s from the clip lengths the timeline shows). Not run: the HUD with no camera (#967; a camera is attached), the macOS HUD fix (c2a8559e), tray, microphone and system-audio content, GIF, macOS and Linux. | | 2026-10-03 | installed v2.0.0-rc.13, CI-built DMG Openscreen-macOS-Apple-Silicon-2.0.0-rc.13.dmg from the GitHub Release (tag 018d22ab), installed over rc.12, spctl: accepted, Notarized Developer ID (M4LK7C6S84), About reports 2.0.0-rc.13 | macOS 26.5 (25F71), Mac mini M1, 3840×2160, no camera, no microphone | Partial, no defect | Scoped to the app changes between rc.8 and rc.13, on macOS: the Windows row above covers the same diff on Windows and leaves out #967, c2a8559e and macOS, which this row takes. Real OS mouse and keyboard input through computer-use (screenshots worked on this Mac); Apple's picker is owned by Control Centre and cannot be granted, so the operator clicked Share Entire Screen for each take. c2a8559e (HUD kept out of takes after the editor round trip): take 1 picked from the HUD's record button (excludedWindowIds: [582]), stop, editor, Record mode showed the source as Choose what to record, Start recording reopened Apple's picker instead of reusing the pick, take 2 recorded with excludedWindowIds: [639]; frames extracted from both files (28:26 and 13:04) show no HUD where it sat. #967 (no camera on this Mac): with camEnabled: true and no camera identity in recording-settings.json, launch rewrote it to false and the HUD showed the camera off; same with a stale identity (camDeviceId + camDeviceName of an unplugged camera); clicking the toggle keeps it off with the "Camera access is blocked" toast; Record mode reads Camera Off. #961: closing the editor after take 2 (countdown overlay, Record mode handover) left 0 processes; so did closing it after a long editor session (eight project opens through the native panel, transcriptions, Media mode, an export, the export dialog's save panel cancelled). #966: with Add zoom bound to X in shortcuts.json, the empty zoom lane reads "Press X to add zoom", and "Appuyez sur X pour ajouter un zoom" / "Drücke X, …" in French and German. Wordmark menu (3bee2d25): French and German, every row on one line, "2.0.0-rc.13" in one piece. #960 (b834416c): project A with captions → project B → A again: captions show at once at 0:06.4, 0:10.2 and 1:05.2 without touching Show captions (0:05.3 shows none, a 4.77 to 5.57 s silence in the saved words). Preview: playback from 0:12 to the end across four transcript silence cuts, a 2× and a 16× region: picture and captions follow, the readout advances with the edit, stops at 1:06.2; no stall seen, 8 [pipeline] décodage opens for one seek and four cuts (consistent with a preload per cut; not compared with rc.12). 227b9b84 / 83fbcecc: a two-clip project whose first clip is a file with 8 s of video and 10 s of audio (no camera): seeking from clip 2 to 0:09.5 shows clip 1's last frame, the same as 0:07.7, no stderr. A video-only file plays with its picture (e73097c8). #968: project whose first asset is chmod 000: MediaUnreadableError … Permission denied in the main log, clip 2 still transcribed, the asset card shows "Transcription failed" and "Couldn't read this media's audio. Make sure OpenScreen can access the folder it's in, then regenerate." without the IPC wrapper; after chmod 644, Regenerate runs again (not persisted as no-audio). Metal compositor (8caaa20d, dd5e914b, 0624c148, 2fb994eb): laptop frame, 3D cursor, click impact, image background blurred 53 %, Aurora, captions; MP4 1080p60 export of 3374 frames, video and audio 56.23 s against 56.21 s computed (66.154 s minus 5.09 s of cuts minus the 2× and 16× gains), full decode clean; extracted frames show the laptop with its shadow, the blurred background, the 3D cursor with its shadow, rounded screen corners and captions. During playback the preview followed image → image, blur 52 → 1 %, image → colour and back. Also seen: the playhead keeps its time when another project is opened (0:41.0 from A shown in B and C). With a hand-written camDeviceName and a null camDeviceId, the HUD keeps the camera on and the file at true (useCameraHudSync compares the empty selection with the empty id and returns early); the app always writes both together, so this state is artificial. Out of scope: a 10 s extract with speech under loud video sound transcribes as no speech in Auto (detected en, p 0.35) and as "Musique" in French. Not run: webcam and microphone (none on this Mac), so the Windows row's webcam-past-the-end defect cannot be checked here; click impact not confirmed (the fixture's cursor sidecar has no clicks, and none was identified around the nine clicks of take 2 at preview scale); Aurora motion on a plain colour; the 4× preview lag of the rc.12 row (not re-measured: this pass played 2× and 16× regions without timing the lag); Esc; tray; GIF; A/B against rc.12; Gatekeeper (the DMG came through gh, no quarantine flag); Intel. | | 2026-10-03 | installed v2.0.0-rc.15 (build run 37124242835, tag ccd8b449, NSIS sha512 matched latest.yml), installed over rc.12 for all users, About reports 2.0.0-rc.15 (win32 x64 · nsis, Electron 41.2.1, Chromium 146.0.7680.188, Node 24.14.1). v2.0.0 was promoted from this candidate during the pass and differs from it only by the version bump (eff86dcb), so the results hold for the stable code | Windows 11 Home 26200, 1920×1080 @ 125 %, AMD Ryzen 5 7520U / Radeon iGPU, built-in USB2.0 HD UVC WebCam (best mode 1280×720@30), Microphone Array (AMD) | Partial — 6 defects | Whole-file pass on Windows. Real OS mouse and keyboard input throughout. Computer-use screenshots came back grey again, so the app was observed through PrintWindow on its own windows and DOM reads over CDP, and every picture and sound claim below was measured on the file with ffprobe/ffmpeg. Defects: (1) #1004, HUD drag at 125 %: the HUD window grows about 1.5 px per drag step (1180×882 → 1232×935 over ~43 steps) and the bar jumps by half the growth on release; after a long drag down the bar sat partly under the taskbar (centre y 1042, work area ends at 1032). (2) #1005, Edit clip: Reset, Cancel and Apply sit inside the scrolling body. In a 1240×1000 window (794 css px tall) the card is clipped above Apply, and a click where Apply should be lands on the backdrop, closing the dialog and discarding the edit. It fits when the window is maximized. (3) #1006, the rail's "Choose a clip to edit" menu renders under the timeline toolbar: Clip 2 and Clip 3 are covered (elementFromPoint returns _tlToolbar) and clicks on them do nothing. (4) #1007, a clip duplicated with Ctrl+C / Ctrl+V does not carry its anchored zoom or Full Camera region: duplicateClip in src/lib/ai-edition/document/timeline.ts copies only trimRanges. (5) #1008, zoom repel across a clip boundary: a zoom dragged into a neighbour that ends at its clip's end jumps past it into the next clip, shrinking from 12.97 to 8.23 s and leaving a zero-length fragment on the first clip. Ctrl+Z restores it. (6) #1009, the wordmark menu does not close on a click on the bare top bar. The header is a -webkit-app-region: drag area, which never delivers the document mousedown that AppMenu listens for. It does close on the preview, on a top-bar button and on the wordmark. Passed: HUD and capture. One HUD, [content-protection] OFF logged; horizontal ↔ vertical layout; every tooltip, idle and recording, beside or above the bar and unclipped; language menu (15 locales); camera and microphone toggles in one click; device settings: three inputs, level meter, live camera preview, camera quality row (1080p persisted across a relaunch; hand-edited 720p and an absent key both read 4K). Hide bar, then the tray overflow icon brings the HUD back; Quit leaves 0 processes; a second launch exits 0. With no source the record tooltip reads "Choose a screen or window to record" and one click opens the picker, then starts. While recording: source button disabled, toggles aria-disabled, gear inert, tray tooltip "Recording: Tout l'écran". Pause freezes the timer and resume advances it; restart starts a new helper and deletes the first file; cancel deletes the file and opens no editor. Stop opens the editor with "Recording added to a new project". Screen H.264 High bt709 1920×1080; AAC 48 kHz; webcam 1280×720 ~30 fps at 8 Mbit/s, the camera's best mode, not upscaled, matching webcamFormat. Microphone-only, system-audio-only and all-off takes behave; window source 1270×668, non-black; an odd 1001×601 window gives an even 992×596 file; a context menu inside the window is recorded and cut at its edge. Without the flag the HUD and an open Notes window are absent from the take. Record mode: rows, live camera preview, microphone meter, device switch, source modal with badge, Start recording hands over to the HUD; auto-zoom on gives one merged zoom over three clicks, off gives none, and the setting persists; editable cursor off hides the auto-zoom row and the Cursor facet; Hide desktop icons shows a bare wallpaper and restores the icons after. Passed: editor. Rename, Saved dot, tabs, chat panel toggle, tooltips with key chips, dividers; play, pause, seek playing and paused, ±1 frame arrows, stop at end; ruler click and drag, navigator narrow and pan, labels without collision, playhead within 0.5 px; Ctrl+S toast with a stable top bar. Clips: media card drag and Add to timeline, reorder, Edit clip grips, 1:1 crop held through a corner drag, Free, Cancel, pencil, delete. Trim, zoom (levels, custom, out-of-range message, focus drag, 3D modes), speed (presets, custom, 16× cap message, playback at 16×), annotation (text, sizes, plates, palette, Typewriter, image, arrow, blur mosaic and smooth, oval, clamped, dragged into padding); Ctrl+D, Ctrl+Z, Ctrl+Shift+Z. Modifiers inside a trim fire on the parked frame. Copy and paste of zoom, annotation and trim with their toasts and properties; an empty copy is a no-op; deleting a clip removes its modifiers; Edit clip clamps them; a Full Camera region across a junction splits and merges as clips move. Imported music: fragments across a junction with offsetMs, −18 dB and 1 s fades, gain, Reset audio; voiceover lands at the playhead. Transcript: clip order, silences, monotonic word seeks, live cue, Backspace skip and restore, silence trim; captions in four styles and six anchors; French translation with a Display row, and deleting it leaves the words untouched; typing in the packaged gate does nothing; a double-click correction survives Regenerate as English (model reused); 101 languages, detected language on the card. Composition: colour, gradient and one-colour gradient, wallpaper, animations move during playback only (21–23 distinct frames in 3 s, None 1), blur, 9:16 with Whole and Follow cursor, Auto padding with even borders, shadows, four frame styles in both themes. Camera layout: four presets, mirror, square, size (default 40, max 60), roundness, positions; background Cutout, Blur with an intensity slider, Custom image and colour, and the custom-colour mask reaches a 720p export; webcam crop corner zoom (124 %) and frame move. Audio facet: +6.5 dB raises the export from −17.2 to −12.7 LUFS with the peak held at −1.3 dBFS; reset returns 0 dB. Cursor facet: hide, style, size (to 5.8 of 6) and 3D cursor all show in the preview. Depth of field appears once a zoom has a 3D preset and changes the preview. Wordmark menu: rows as listed, About row version, arrows wrap both ways; Keyboard Shortcuts as one dialog: Z → Q changes the lane hint, Q adds a zoom, Z does nothing, reset restores Z; Ctrl+O; AI settings as one dialog; light and dark themes, dialogs included; French UI with one-line menu rows (#969), back to English; About lists the versions and channel, and Copy puts the same block on the clipboard; the HUD panel's version row and Check for Updates ("Checking…", disabled, the same dialog as the menu, re-enabled after). AI chat (deepseek): "add one zoom at 2x from 5 s to 8 s" lands as 5–8 s at 2.20× with the reason given, is undone by one Ctrl+Z and redone. Closing the editor quits the app and writes editor-window.json (maximized: false); after a relaunch the last project's settings and the main project's two clips, crop, regions, music, voiceover and captions are as left, and a paste before any copy changes nothing. A schema-4 project from July opens, migrates to schema 8, and keeps its clip, seven zooms and annotation. Export: MP4 1080p30 of the two-clip project, 10288 frames, audio and video within one frame, −16.0 LUFS and −1.4 dBTP, annotation upright, caption plate at x = 108 for a 10 % inset, Full Camera where mapped, music ducked about 10 dB under speech with fades at the track's own edges, muted track absent, voiceover not stretched by a 2× region. GIF 25 fps: cancel removes the partial file and leaves the existing destination untouched; the full render (127 s, #952) gives 434 frames, infinite loop, consistent in ffmpeg and GDI+. Also seen: a take with a 4 s pause had audio 1.09 s past its video, against 0.02–0.17 s without a pause; three dedicated paused takes gave 0.40, 0.15 and 0.03 s. The export ends each clip at its last video frame (10288 frames against the 10320 the progress counted), so the timeline and the export disagree by that tail. The editor reopens 4 px taller than it closed at 125 %. Not run: the A/V offset across a pause measured directly (the flash-and-beep rig never painted on the captured desktop), the HUD with no camera (needs the camera disabled in Windows), tray right-click menus (Update Settings, Stop Recording, Save Diagnostics: Explorer is granted click-only), the Alt Help menu (an Alt tap showed nothing, inconclusive), Ctrl/Shift+wheel on the timeline, cursor auto-hide, smoothing, motion blur and click effects, annotation animations other than Typewriter, Depth of field in an export, chat rewind, history and Smart cuts, a software-encoder take, the crackle and hole scan, multiple displays, 4K and 1440p cameras, 150 % DPI, macOS and Linux. | | 2026-10-03 | main @ 0624c148 (sources identical to v2.0.0-rc.13 in crates/, src/, electron/), packaged locally (electron-builder --mac dir --arm64, ad hoc signed), compositor_view.node SHA256 7c1d38d664d29fa78615c386aa86e53456f1e662289ad53a2a128d588b99d560; baseline the installed v2.0.0-rc.12 DMG (Developer ID) | macOS 26.5 (25F71), Mac mini M1, 1920×1080 @ 2× | Partial, no defect | Shader split, corner fast path, background cache and trail scissor (#982, #983, #984) on Metal. Not a computer-use pass: request_access was refused with nobody at the machine, so the editor was driven over CDP (synthesised clicks on the inspector, preview pixels read from the canvas). cargo test -p openscreen-compositor --lib --tests: 407 passed, 0 failed, every_shader_entry_point_compiles (with the LAYER_MODELS library) and the macOS device-frame and 3D-cursor tests included; the tests/ render binaries are Windows/Linux only and ran 0. Export (openscreen export, MP4 1080p60): bench project plus four variants (laptop + 3D cursor + click impact + blur, phone dark + Prism Glow + Aurora gradient, monitor + click impact + colour, window + full blur), frames extracted with ffmpeg: device frames with their shadows, 3D cursor with its shadow, click impact, trail, rounded corners and backgrounds all present. Same projects exported losslessly (ffv1 candidate added locally, measurement only) with the rc.12 and main addons in one bundle: luma differs by 1/255 on anti-aliased edges, and by up to 20/255 on at most 159 px of a few frames along the 3D models' silhouettes (Metal fast math, models now in their own library); nothing missing. Preview: wallpaper image → gradient → colour → back, blur 50 % → 0 %, aspect 1:1 → Auto each return the exact starting frame; a static background holds still over 1 s of playback, Aurora animates; laptop and phone frames draw their shadows. Perf (screen-recorder-benchmark commons-upload): rc.12 1.134× floor, main 1.066×, figures in rendering-performance.md. Not covered: real recording, HUD, tray, computer-use input, the GUI export dialog, Escape, a 1.11.0-rc.1 run in the same session (discarded: a Parsec session went live mid-run), Windows (the D3D11 port is still unrun on a real GPU) and Linux. | | 2026-10-11 | dev helper from fix/os920-resample-continuity @ aa27e6fdf (audio fix commits 50a433168 cross-packet continuity + e29119445 starvation-resume restart; helper and test exe built from the committed tree with npm run build:native:win; test-exe sha256 bb6921d0…); historical evidence archived under former fork staging PR My-Denia/openscreen#23 for study issue #920 §24; current submission: upstream PR getopenscreen/openscreen#1113, open, with CI green on the head under review | Windows 11 build 26340.9616; both audio endpoints fixed 48 kHz shared mode (measured: single-valued 480-frame/10 ms packets, 0 non-aligned) | Partial — no defect found; two paths not real-machine testable on this host | Audio resampling continuity slice (#920 item 24), helper-level + real-machine hybrid. The defect: convertAudioWithGain's non-integer branch held no cross-packet state, so identical input produced different output depending on packet boundaries (48k→44.1k, 100-frame packets: 14700→14720 frames, drifting positions). The fix adds cumulative integer-rational output accounting (K(N)/C(N)), absolute-position interpolation with the legacy arithmetic preserved, bounded retention, and a finalizeAudioConversion tail used only by the stateless test-facing overload; a second commit restarts the interpolation stream at mixer starvation-resume (automated-review P2: the new state could otherwise complete ≤⌈T/S+½⌉ deferred frames across a real-time gap using pre-gap samples — a blend the pre-fix stateless converter never made; integer decimation and same-rate paths keep their pre-existing cross-gap behavior, pinned by controls). Evidence: assertion-based old-FAIL on the pristine pinned base (9 named failures) → new-PASS; full native suite 176/176 and focused 54/54 on that base; starvation fix proven behaviorally old-FAIL (50a433168: 1 failing mixer-path test — resumed queue 368 bytes vs cold 364, first frame −12594 = pre-gap/post-gap blend vs cold +16384) → PASS after e29119445. Submitted-tree native validation, separately attributed: pre-chain validation of the candidate tree that became 5764aa2b6e completed npm run build:native:win with 0 errors, full native suite 177/177, and focused suite 55/55, including os920-finalize-appends-after-whole-frames PASS. These are separate submitted-tree results, not a re-run or replacement of the historical pinned-base pass above. Real machine (matched-rate only): 6×60 s 48 kHz/48 kHz recordings (3 pre-fix/3 fixed helpers interleaved) + pause/resume takes — all valid, marker spacing 3.0000–3.001 s, old/new within each other's repeat envelope, zero clipping; captured real callback streams replayed through old/new converters 6/6 byte-identical; no regression observed. Mismatched-rate real capture (second pass, same day, owner-enabled): the default render endpoint (JM6PRO) was switched to 16-bit/44100 Hz and the encoder target forced to 48000 via the production OPENSCREEN_WGC_FORCE_AAC_SOURCE_RATE surface (the video-rate-alignment path); five 35 s takes ran old×2 / fixed×2 / instrumented-fixed with the same marker harness. Every real loopback packet was exactly 441 frames (3496/3496 — this hardware produces perfectly aligned 10 ms geometry, so the packet-boundary defect class is not reachable on it and stays covered by the deterministic suites); decoded old vs fixed output is time-aligned at zero offset with 97.4 % byte-identical frames and residual maxdiff 170 (AAC block-boundary quantization from the one-frame first-packet emission difference — no drift, no regression); each build repeated byte-identically; the mixer's startup starvation-padding path ran on real hardware (70 ms). Starvation/underrun beyond startup was not reproduced on real hardware — covered by the deterministic real-AudioMixer-path tests above; no full capture-to-export GUI cycle, editor/export/transcript sections, webcam, GIF or non-audio checks for this slice; macOS/Linux N/A (WGC-only code). Fork CI did not run at pass time (GitHub-side Actions suspension on the repo — settings and workflow states reported enabled/active while every dispatch returned 422; full evidence in the run); Actions was restored later the same day and CI has since run green on each exact head under review (runs 38119381114, 38121492689, and 38125479601 recorded for their respective exact heads — run evidence in the project archive). This CI has no native C++ job — the native evidence is the local MSVC build and suites above. Upstream PR #1113 is open, with CI green on the head under review; awaiting maintainer review. | | 2026-10-12 | installed v2.1.0 candidate, CI-built DMG Openscreen-macOS-Apple-Silicon-2.0.0.dmg from build run 38252812147 (openscreen-mac-arm64, release/v2.1.0 @ 9c55596e1), Developer ID signed and notarized (spctl: accepted, Notarized Developer ID, M4LK7C6S84), installed over the previous /Applications/Openscreen.app; About reports 2.0.0 (dmg) (the branch does not bump package.json), Electron 41.2.1, Chromium 146.0.7680.188, Node 24.14.1 | macOS 26.5 (25F71), Mac mini M1 (arm64), one 3840×2160 display at 2×, no camera (Scaleway MDM allowCamera=false), BlackHole 2ch as the only input | Partial, 6 defects | Real OS mouse and keyboard through computer-use, launched with open --env OPENSCREEN_DIAGNOSTIC=1 (no content-protection flag needed on 26), then OPENSCREEN_MAC_SOURCE_PICKER=legacy for every take after the first; tones into BlackHole (440 Hz) and the speakers (1 kHz) gave the mic and system-audio signals; every number below is ffprobe/ffmpeg on the files. Covered, post-2.0.0 first: recordings folder (row last, native picker, take lands in the custom folder, row survives restart, Reset to default only while custom and takes return to userData/recordings, deleted folder recreated by the next take, ejected volume gives "Can't reach the recordings folder", 298 MB volume refused from the HUD and the voice-over with "Not enough disk space ... (only 298 MB free)", from Record mode no take and no file but the toast was not captured, Restart on a volume filled under 500 MB gives "The new recording could not start. The previous one was deleted." with no file left); mic double-click stays off; Play past the last clip holds clip 2's own last frame and stops at the total (#997); duplicate clip carries zooms, speed and annotation (#1011); Edit clip trim-bar playhead, in-point frame, scrub with Apply disabled, grip pushes the playhead; zoom trails and their in/out/pan/trim/junction tooltips, trails follow a dragged trim, a trim between two zooms breaks the pan, end of timeline is no junction; Names and terms ("OpenScreen" spelled, opening kept, not Title Case; one undo restores field and transcript), transcript Backspace cut and Cmd+Z (#1012), Cmd+Z inside an edited word; agent: filler words plus a quoted phrase as one trim with no sliver, Screen turned left then 3D Orbit, stretch into a neighbour refused naming it; two conversations and the last reply survive a quit right after the reply, history is per project; MCP stdio (help lists mcp, listSources, startRecording on a window: 2944×1926 take in the custom folder, no Apple picker, 3 clicks 1.7 s apart grouped into one zoom; edit while closed, exportVideo 25.82 s = 33.82 − 8 s trim, getFrames, restoreCheckpoint back to the raw take, stale warning with projectId; edits refused on the project open in the editor and on a closed project without checkpoint); in-app server off by default, Running with port, token and snippets; GIF of 61 s at 20 fps Small in about 8 s; Glass Lens and 3D Glass Lens in preview and in a 720p export (refraction, shadow); Mouse clicks off by default, Click volume row, clicks in the export at the speed-mapped times (±20 ms). Covered, rest of the file: HUD launch, layout toggle, drag without drift or jump, language menu, tooltips on both bars idle and recording (a teleported pointer needs one small move, as a hand gives), gear lists inputs with a moving meter, "No camera found" and no Camera quality row, camera toggle refused with "Camera not found.", hide and tray refocus, tray menus idle and recording (only Stop Recording), quit and relaunch, single instance (open -n exits), Update Settings modes persisted in update-settings.json across restart; app menu items; About box; legacy selector; Record mode rows, Auto-zoom row only with Editable cursor, mic meter, hand-over to the HUD with the same settings, zooms placed after a take and none with Auto-zoom off; pause/resume, Restart, cancel without editor, stop from HUD and from the tray; gear does not open during a take; full-screen take fills 3840×2160 edge to edge (HiDPI), HUD absent, desktop icons absent while still on the desktop; mic and system audio each audible alone and together (−23.8 / −34.6 dB band levels); editor opens in Edit with the take and toast; New project, Import, Add to timeline toast, Media card name/duration/size, Export disabled with no asset; local transcription on Metal with French detected; Cmd+O, theme and language switches (UI and native menus follow, back to English); export dialog (formats, qualities with sizes, fps, no codec, GIF options and summary), MP4 progress, cancel with no partial and same options, retry, Saved to, Show in folder, GIF cancel keeps the existing destination, MP4 in QuickTime and GIF in Preview; imported audio pill named after the file at −18 dB with 1 s fades, a bed past the programme end is cut without pulling its fade-out forward; Save Diagnostics from the app menu writes a bundle; reopened project keeps clips, in/out, zooms, speeds and trims. Defects: (1) dragging a zoom into a neighbour of the same level passes into it instead of stopping flush (#1008): 23.0 s zoom dragged left landed at 20.055–21.984 s over 19.975–21.904 s, a 2 px nudge still overlaps by 40 ms; the pair draws as one pill and a level change on one half changed both, so #1017 could not be judged on exactly touching regions; medium. (2) MP4 export of the speech fixture (source −17.6 LUFS, +4.9 dBTP) measures −17.3 LUFS, sample peak −0.59 dBFS, true peak +0.5 dBTP against about −16 LUFS and at most −1 dBTP; medium. (3) agent title "Demo OpenScreen" with a switch to 9:16 in the same turn: box 95 %×9.33 % at size 64, only "Demo" visible in the preview and in the exported frame at 3.0 s; medium. (4) click sounds missing on a duplicated clip: clip 1 clicks at 6.04, 10.26, 14.44, 28.59, 36.91 s are in the export, the duplicate's click expected at 43.17 s is absent; low to medium. (5) HUD error toasts sit over the HUD's own controls, their × does not take the click (it reaches the window behind) and they stay while the pointer is over them; the mic toast at take start hid Pause, Restart and Cancel until the pointer left; medium. (6) a microphone picked as "BlackHole 2ch (Virtual)" is not resolved by the SCK helper ([native-sck] recording the default input; microphone was not resolved), so every take toasts "The chosen microphone could not be identified"; low. Observations, not filed: transcription shows "Transcribing 0%" for the whole job, 66.2 s of French in 101.6 s and 112 s (0.6–0.7× real time with the fr CTC aligner) against 61× on a speech-free take; the legacy Windows tab listed one window while the Claude window was on screen; the disk-space refusal comes after the 3-2-1 countdown; "Loop off" writes a NETSCAPE loop count of 1. Skipped: Apple's picker slice and the permissions window's first-run rows (operator asked for an unattended legacy run; Screen Recording already granted and not reset; one early take went through the picker and is not judged); every camera, webcam-quality, Full Camera and camera-background check (MDM blocks cameras); Windows, Windows on ARM, Linux and macOS 13/14 sections (other OS or version); Intel ONNX check (arm64 only); second display at another scale (one display); GIF speed against v2.0.0 (no v2.0.0 left installed to compare); in-app MCP edits and port change (no HTTP MCP client here without downloading one) and setPointerTrack; Escape checks and Ctrl/Shift-scroll (unreachable from computer-use on this Mac); Spaces switch (no system key combos granted); no-microphone refusal and unplugged mic (BlackHole cannot be removed here); not run for time: captions and caption placement, translation, annotation types (image, arrow, blur), v1.8.0 anchoring section, most audio-track checks, voice-over take, inspector facets, shortcuts dialog, editor window bounds, migrated-project persistence, chat rewind/compaction/rename/delete, export speed-region audio section. | | 2026-10-12 | installed v2.1.0 candidate after release/v2.1.0 was fast-forwarded to main @ bd04d3a7 (adds #1147, #1151, #1155, #1158, #1160, #1166, #1131), CI-built DMG Openscreen-macOS-Apple-Silicon-2.0.0.dmg from build run 38309126813 (openscreen-mac-arm64), Developer ID signed and notarized (spctl: accepted, Notarized Developer ID, M4LK7C6S84); the SCK helper is dated 23:22 that day and carries the new Chromium suffixes (Virtual, Bluetooth, Built-in) | macOS 26.5 (25F71), Mac mini M1 (arm64), one 3840×2160 display at 2×, no camera (MDM), BlackHole 2ch as the only input | Pass, scoped to the merged fixes | Targeted re-run of the six fixes from the pass above, real OS mouse and keyboard through computer-use, launched with open --env OPENSCREEN_DIAGNOSTIC=1 --env OPENSCREEN_MAC_SOURCE_PICKER=legacy; numbers are ffprobe/ffmpeg on the files. #1158 (HUD toasts): "Camera not found." now shows centred above the horizontal bar, clear of every control, and its × dismisses it with a real click. #1147 (microphone suffix): "BlackHole 2ch (Virtual)" picked explicitly in the gear, mic on, HUD take: no "could not be identified" toast, no microphone was not resolved line in the main log, and a 440 Hz tone sent into BlackHole is in the take at −23.7 dB band level; only one input exists here, so picking an input that is not the default was not exercised. #1151 (click sounds on a duplicated clip): re-export (720p 30 fps) of the two-clip project: clip 1 clicks at 6.04, 10.26, 14.44, 28.59, 36.91 s unchanged, and the duplicate's click is now present at 43.16 s (expected 43.17 s). #1166 (export loudness): the speech fixture exported at 1080p 30 fps measures true peak −1.5 dBTP (was +0.5), sample peak −1.52 dBFS, integrated −17.2 LUFS (the clipped source cannot reach −16 without re-clipping, as the PR states), AAC at 264 kb/s. #1160 (agent title fit): the chat agent, asked to switch to 16:9, add "Bonjour OpenScreen", then switch to 9:16, refits both titles (56 px and 61 px, one line each); the exported 1080×1920 frames show "Demo OpenScreen" and "Bonjour OpenScreen" whole, where the earlier build showed only "Demo". Smoke: launch, HUD, record, stop and editor open, Cmd+O, export dialog. Skipped: #1155 (Windows WGC helper only, needs a 10 min loaded take on Windows); the rest of the file (covered by the row above on the earlier build). | | 2026-10-12 | release/v2.1.0 @ 9c55596e1, CI artifact openscreen-windows-x64 from build run 38252812147 (NSIS sha512 matched update-info-win-x64.json), silent per-user install over 2.0.0, About reports 2.0.0 (win32 x64 · nsis, Electron 41.2.1, Chromium 146.0.7680.188, Node 24.14.1); package.json is still 2.0.0 on the branch, so only the run id tells this build apart | Windows 11 Home 26200 (x64), AMD Ryzen 7 5800X 8 cores, RTX 4070 Ti, 1920×1080 @ 100 %; Logitech StreamCam (sub-4K, MJPEG 1080p30 max), NVIDIA Broadcast camera (DirectShow-only) | Fail — 2 should-fix, 1 medium, 6 minor; post-2.0.0 sections first, rest partial | Real OS input through computer-use; grant openscreen.exe by executable, because Openscreen resolves to the Store 2.0.0 package (same trap as rc.3). Defects. (1) Should-fix, MCP stdio: an edit sent to a project open in the editor is accepted ("not open in the editor"), written to disk, then silently dropped by the editor's next autosave (trim added 17:07:25, gone at 17:07:44, listProjects reports open:false). (2) Should-fix, #945: 10:25 take, webcam + system audio, under a continuous CLI export loop: the screen track is in step with the wall clock until 5:00 (HUD-timer flips at 21.017 s and 301.017 s), then loses 217 ms by 10:00 (−0.033 @376, −0.083 @411, −0.150 @451, −0.200 @601); audio loses 43 ms, webcam none, so the screen ends 187 ms short of audio and 221 ms short of webcam. (3) Medium: any CLI/MCP process shares %APPDATA%\openscreen and takes Chromium's Local Storage lock (Restart Manager: holder electron … mcp), so the GUI runs on in-memory storage: the system-language prompt returns on every launch after "Keep current language", and the stdio server logs Unable to move the cache: Accès refusé. Minor: identical speed regions dragged into each other overlap (1.27 s) instead of stopping, then edit together; a project's first chat message shows a raw ISO UTC time; stopRecording gives 1 zoom for 2 clicks (first at 1.9 s) while its description promises one per click; MCP listSources lists 3 windows where the HUD lists 5 (Settings absent); the open-project search ignores Enter on a single match; HUD dropped near the top moves 68 px down when a popover opens and stays there. Doc drift: camera shape is Square/Original, not Rectangle/Square. Open question: main export 2451 frames at 30 fps against 2444.9 predicted (97.579 s − 10.686 s trims − 5.397 s speed), +6 frames over 13 segment boundaries, per-segment rounding suspected, not proven. Passed: HUD launch, layout toggle, drag, language menu (Esc closes it), camera/mic toggles, mic double-click cancels, camera quality 1080p persists, hand-edited 720p shows 4K, tray refocus and red recording icon with "Recording: Tout l'écran", single instance; record with no source opens the picker and starts; pause (62.5 s wall for 48.8 s file), restart (old take deleted), cancel, all-off stop → editor < 2 s, gear locked during a take, content protection ON keeps the HUD out of the file; screen High/no B-frames/bt709 tv, AAC 48 kHz, videoEncoderRuntime: "hardware"; webcam 1920×1080@30 = webcamFormat, 16 Mbit/s, last frame live; DirectShow camera resolved, file = negotiated 1080p30; Hide desktop icons (wallpaper in file, icons back after); recordings folder pick/persist/recreate/reset; zoom trails, hatched cut tooltip, pan chain and its break, flush drag and resize (#1008, different values), #1017 on touching regions, clip duplicate carries zoom/speed/annotation/Full Camera (#1011), Edit clip scrub (#997 last-frame hold too), Full Camera across an inner trim; names and terms fix the spelling and undo with the field, transcript Backspace + Ctrl+Z (#1012), word correction; chat: filler words (one trim), quoted phrase, screen turned left → 3D Orbit, 9:16 + fitted title, refused stretch naming the neighbour, one undo per turn, two conversations survive a quit, per-project history, project delete removes chat.json; stdio MCP: help lists mcp, listSources, window take → project in the editor's list, edit refused without checkpoint, restoreCheckpoint to the raw take, export, getFrames through trims and its stale warning, setPointerTrack draws the cursor and zooms on scripted clicks; Glass Lens 2D/3D in exports, click sound at 3.72 s for 3.727 s predicted after a 1.5× region, none in a trim, none with Show cursor off; GIF 13.4 s vs 127.4 s on v2.0.0 (9.5×, 8 cores, same 249 frames); export dialog v2, MP4/GIF cancel (no partial, destination hash unchanged), −15.9 LUFS / −1.6 dBFS, A/V within one frame; imported track −18 dB/1 s fades, ~10 dB duck under speech, continuous across the junction; captions + French translation; About/update dialog; theme and language; Media/Edit/Record; navigator pan/zoom; editor bounds and invalid-file fallback; persistence across restarts. **Skipped, with reason:** microphone checks (StreamCam mic delivers no samples even to ffmpeg, headset mic is digital silence), voice-over; in-app HTTP MCP client checks (copying the token was refused by the agent's classifier); tray right-click (shell granted at tier click) so Save Diagnostics, Update Settings and helperOutput (delivered=, no-visible-frame) are unread, and Alt never showed the Help menu (keyboard negative, unconfirmed by hand); Ctrl/Shift+wheel (wheel events never reach the editor through this driver); DPI scaling (changing OS settings is outside what the agent may do); 4K, no-30-fps, 640×480 and C920 cameras, covered lens, 96 kHz device, second display, removable drive and low-disk volume, notes window; Windows 10, ARM, macOS, Linux. | | 2026-10-13 | release/v2.1.0 @ 912b44800 (fast-forwarded to main: #1144, #1145, #1150, #1152, #1153, #1154, #1156, #1161, #1163, #1165), CI artifact openscreen-windows-x64 from build run 38370716184 (NSIS sha512 matched update-info-win-x64.json; the run's macOS x64 job failed once on a notarization-staple network error, NSURLErrorDomain -1005, and passed on re-run), silent per-user install, About still 2.0.0 (intermediate build) | Windows 11 Home 26200 (x64), Ryzen 7 5800X, RTX 4070 Ti, 1920×1080 @ 100 %; Logitech StreamCam | **Pass — scoped re-run of the fixes from the 2026-10-12 row, 1 new minor** | Real OS input through computer-use, measurements on the files. **#945 / #1165:** 10:57 full-screen take, StreamCam 1080p30 + system audio, while 44 CLI exports ran back to back: screen 657.535 s, audio 657.536 s, webcam 657.537 s (2026-10-12: screen −187 ms and −221 ms); HUD-timer flips filmed at 21, 301, 451, 601 and 651 s all sit at +0.033 to +0.050 s, so the screen holds the wall clock to one frame over the whole take (2026-10-12: −217 ms after 5:00); TTS markers drift 0 ms over 567 s. **#1163:** CLI export of the two-clip project at 60 fps gave 4886 frames against 4886.8 predicted from the clips' real video length (48.765 s each); the progress total said 4890, the residual the PR documents. **#1154:** with the project open in the editor, stdio listProjects reports it open: true, createCheckpoint + addTrim is refused ("open in the OpenScreen editor …") and the file is untouched; .open-in-editor.json carries the GUI's pid. **#1145:** an installed Openscreen.exe mcp locks its own cli-sessions\<pid>-<id>\Local Storage, never the GUI profile, with no cache error, and dead runs' folders are swept on the next run. The GUI-side symptom (language prompt on every launch) could not be re-tested: ten older dev electron … mcp servers from another worktree, built before the fix, still hold %APPDATA%\openscreen\Local Storage\leveldb\LOCK. **#1150/#1153:** listSources, startRecording's not-found error and stopRecording now say what is not listed and how clicks share zooms; the not-found error still carries a JavaScript stack trace. **#1152:** a project's first chat message shows 11:16:44, like the reply. **#1156:** Enter in the Open project search opens the highlighted single match. **#1161:** HUD dropped at y≈200, the language menu and Device settings each push the bar down while open and it returns to y≈200 once they close. **#1144:** two identical 1.5× regions, the second dragged into the first, stop flush at 25.173 s with the length kept (4.983 s). **New, minor:** adding a region with its key while the playhead is inside an existing region of the same kind creates an overlapping one, whatever the values (2× 20.190–25.173 then 1.5× 23.253–28.236), the creation counterpart of #1133. Not re-run: microphone (no working input on this machine), everything outside the fixed points. | | 2026-10-14 | main @ b668b2c8d (MCP unification #1174: #1175, #1176, #1179; editor parity #1180: #1181 to #1186), CI artifact openscreen-windows-x64 from build run 38538195250 (NSIS sha512 matched update-info-win-x64.json), silent per-user install, About still 2.0.0 (intermediate build) | Windows 11 Home 26200 (x64), Ryzen 7 5800X, RTX 4070 Ti, 1920×1080 @ 100 %; Logitech StreamCam | **Partial — MCP server section only, 2 defects** | Real OS input through computer-use; MCP calls from an SDK client (@modelcontextprotocol/sdk) to the installed Openscreen.exe mcp, with the app running unless stated. **Tools:** listTools lists all 51 tools (record, export, frames, settings, captions, transcription, clips, projects, media). **Record:** startRecording on a Notepad window with camera: "StreamCam" and system audio, then real clicks and stopRecording, gives an 890×596 60 fps take with a webcam track, a cursor sidecar and 1 zoom for 3 close clicks. cursor: "system" gives no zoom and the session says system. A take dies with its stdio server: a client that disconnects after startRecording leaves an 83-byte MP4 and an empty webcam file. **Closed project, app running:** restoreCheckpoint with the checkpoint from another client's stopRecording works, because it lives in the app. **Media, transcription, captions:** importMedia of a spoken 17.2 s MP4, then insertClip. transcribe finds 40 English words in about 15 s, and getCaptionTranslationUnits fr returns 6 units. putCaptionTranslation writes all 6, and setCaptionSettings {enabled, style: bold, language: fr} shows them. **Look:** setEditorSettings applies a laptop frame, a wallpaper, 16:9 and a camera blur cut-out, and turns format fill on as the editor does. The inactive reasons are read back. Also tested: addAnnotation for a 72 px text and a mosaic blur. **Export:** exportVideo at fps: 30 gives 1920×1080 at 30 fps, 1472 frames for 49.07 s. The frames at 5 and 35 s show the blur, the webcam cut-out top-right, the title and the French captions. gifLoop on an .mp4 is refused. **Open in the editor:** listProjects reports open: true. addZoom and a wallpaper change land live, and each comes off with one Ctrl+Z. restoreCheckpoint resets the take live, and one Ctrl+Z brings every edit back. **App quit and relaunch:** one client session spans an app quit and a relaunch. With the app gone, an edit asks for a checkpoint and lands in the file after createCheckpoint. Once the app is relaunched and the project reopened, the same session's edit shows live, with no client restart. **Other tools:** setProjectInfo, duplicateClip, a clip cropRegion, addAutoZooms (0 new, telemetry reported per asset) and importMedia of an MP3 work. removeMedia refuses an asset still on the timeline and names its clips, and createProject from a video returns a checkpoint. **CLI:** Openscreen.exe captions <project> --max-words 5 writes a 40-word transcript, turns captions on, and leaves no auto-caption annotation. **Defect 1 (P1): every GIF export over MCP fails.** It fails with the app running and without it: exportVideo to .gif → "--quality is only supported for MP4 exports", because electron/mcp/media-tools.ts always sends quality: "good" and the export runner refuses a quality on a GIF. **Defect 2 (P2): the recording bar loses always-on-top** after the source picker or the language prompt closes. Its window style goes from 0x280028 to 0x200000 and it sinks behind other windows; Ctrl+Shift+O brings it forward, still not topmost. **Minor:** camera: true picks the first listed camera, a virtual "VCam Camera" here, which wrote a 0-byte webcam file that the take dropped without a word. The language prompt shows on every launch, the #1145 symptom: ten older dev mcp servers from another worktree still hold the profile. The first two MCP takes had silent system audio (−91 dB) although the narration played; a take from the bar and three later MCP takes, with and without StreamCam, all had it (−3 dB), so this is not attributed. Not run: the in-app HTTP server, because copying its access token is refused to the agent; the microphone, with no working input on this machine; everything outside the MCP section. | | 2026-10-14 | main @ 5bc559198 (adds #1192 GIF export over MCP, #1194 recording bar kept on top, #1197 GIF plays once), CI artifact openscreen-windows-x64 from build run 38564622152 (NSIS sha512 matched update-info-win-x64.json), silent per-user install, About still 2.0.0 (intermediate build) | Windows 11 Home 26200 (x64), Ryzen 7 5800X, RTX 4070 Ti, 1920×1080 @ 100 %; Logitech StreamCam | **Pass — MCP server section re-run, both earlier defects fixed** | Same scope and method as the earlier row of this day (run 38538195250). **#1194:** the recording bar keeps WS_EX_TOPMOST (0x280028) after the language prompt and after the source picker, and stays above a Notepad window brought to the front. **#1192 / #1197:** exportVideo to a .gif with gifSize: "small" gives 852×480 at 15 fps; with gifLoop: false the file carries no NETSCAPE block, so it plays once, and with looping on its count is 0. **Record:** startRecording on a Notepad window with camera: "StreamCam" and system audio gives a 40.6 s 1166×888 60 fps take with a webcam file, system audio at −3.6 dB and 1 zoom. **Transcript and captions:** transcribe finds 18 English words, and putCaptionTranslation writes 3 French units. **Look and annotations:** setCaptionSettings shows the French captions in the bold style. setEditorSettings applies a laptop frame, 16:9 and a camera blur cut-out, and addAnnotation adds a 72 px title and a mosaic blur. **MP4 export:** 1920×1080 at 30 fps, 1219 frames for 40.63 s; the 4 s frame shows the title, the French caption, the blur and the cut-out webcam. **Open project:** listProjects reports open: true. A zoom and a wallpaper change land live, and each comes off with one Ctrl+Z. **One client session across an app quit and relaunch:** with the app gone, an edit asks for a checkpoint and lands in the file after createCheckpoint. Back in the reopened editor, the next edit shows live. **Media, clips, checkpoints:** importMedia of an MP4 and an MP3, insertClip, addAudio with a fade, duplicateClip, a clip cropRegion and setProjectInfo work. removeMedia refuses an asset still on the timeline. restoreCheckpoint on the open project resets it, and one Ctrl+Z brings every edit back. **Without the app:** createProject from a video returns a checkpoint, and Openscreen.exe captions writes a 40-word transcript with captions on and no auto-caption annotation. Not run: the in-app HTTP server (copying its access token is refused to the agent), the microphone (no working input on this machine), everything outside the MCP section. | | 2026-10-14 | main @ 5bc559198, CI artifact openscreen-windows-x64 from build run 38564622152; then fix/mcp-native-take-start @ bdfa71641 (#1205, now merged), build run 38582471723. NSIS sha512 matched update-info-win-x64.json for both, silent per-user install, About still 2.0.0 (intermediate build) | Windows 11 Home 26200 (x64), Ryzen 7 5800X, RTX 4070 Ti, 1920×1080 @ 100 %; Logitech StreamCam | **Pass after #1205 — in-app MCP server; 1 defect found and fixed** | Real OS input through computer-use. Calls come from an SDK client over StreamableHTTPClientTransport to http://127.0.0.1:<port>/mcp, with the Bearer token copied from *AI settings* and read from the clipboard (the server is ephemeral in dev). **Gate:** with *MCP edits* off, addZoom returns consent_required, and its guidance names *MCP edits*; startRecording and exportVideo are refused, also naming it. **Live edit:** with *MCP edits* on, an addSpeed on the open project lands live and one Ctrl+Z takes it back. **Export:** exportVideo at fps: 30, quality: "medium" gives 900×720 at 30 fps. To a .gif with gifSize: "small" it gives 600×480 at 15 fps, and getFrames returns the asked frame. A relative path is refused ("The path must be absolute"). **Port:** set to 47822 and applied, the old port refuses connections, the new one serves the same token, and the Claude Code snippet shows the new port. **Defect (P1, #1204):** on main @ 5bc559198, startRecording on a window closes the editor and the recording bar records after its countdown, with only the app's wgc-capture.exe running, but the call waits 120 s and answers "capture had not started". stopRecording then finds no take, and the bar keeps recording until stopped by hand. Cause: the native capture paths never told the MCP recorder a take had started. **After #1205:** startRecording answers recording: true in 2 s, and stopRecording returns the 29.2 s take with its webcam and 1 zoom, open in the editor under the requested title. The bar's recording settings come back to what they were before the take. Not run: *Enable MCP server* and *MCP edits* off by default, because this profile had already enabled the server; the refusal while the editor has unsaved changes; the microphone, with no working input on this machine. | | 2026-10-14 | main @ 5bc559198, CI artifact openscreen-linux from build run 38564622152. The Openscreen-Linux-2.0.0.AppImage sha512 matched latest-linux.yml; it was extracted (--appimage-extract), because WSL has no libfuse2 and no passwordless sudo for the .deb. About not read, since the GUI was not driven. | WSL2 Ubuntu 24.04 (kernel 5.15.146.1) under WSLg on Windows 11 Home 26200, Ryzen 7 5800X, RTX 4070 Ti. Chromium's GPU process fails to start under WSLg and falls back; Mesa/ZINK warnings. | **Partial — headless only (MCP stdio + CLI), 1 defect** | WSLg draws the Linux windows through msrdc.exe. Granting it to computer-use was declined, so nothing in the GUI was driven: no HUD, editor, tray or settings. The app was launched under WSLg (--no-sandbox) while an SDK MCP client drove openscreen mcp with ELECTRON_DISABLE_SANDBOX=1. The CLI rejects --no-sandbox after a subcommand, and the SDK's stdio transport passes no DISPLAY unless given the full environment. **Works:** - listSources lists one display, no camera (WSL has none) and the RDP Source microphone; - createProject from a 17.2 s video; - setEditorSettings with a laptop frame, a wallpaper and 16:9; - setCaptionSettings; - addAnnotation with a 72 px title and a mosaic blur; - addZoom; - exportVideo at 30 fps: 1920×1080, 516 frames for 17.2 s, and the 4 s frame shows frame, title and blur; - a .gif with gifSize: "small" and gifLoop: false: 852×480 at 15 fps with no NETSCAPE block; - CLI export --fps 24: 413 frames for 17.21 s. **Defect (P1, Linux): transcription fails over MCP and the CLI.** The Linux package ships the ffmpeg libraries (helper-ffmpeg/) but no ffmpeg executable. So transcribe returns "stt:native-extraction-unavailable: no ffmpeg binary for native audio extraction", and getCaptionTranslationUnits has nothing to translate. Openscreen captions <project> falls back to the renderer decode, which refuses the path: "Transcription failed: File path is not approved or is not a supported video file". A system /usr/bin/ffmpeg is on PATH and is not used. Not run: everything in the GUI (grant declined); screen and window recording (no xdg-desktop-portal ScreenCast or PipeWire in WSLg); the webcam (none in WSL); the tray (not shown by WSLg); the in-app HTTP server (its token is read from the GUI); transcription in the editor, which may still work through its own fallback. | | 2026-10-14 | main @ 0bed36852 (adds #1192, #1194, #1197 over the 2026-10-14 Windows rows; c541fffaa, the recording-state fix, landed after this build and is not in it), CI artifact openscreen-mac-arm64 from build run 38581408372, DMG Openscreen-macOS-Apple-Silicon-2.0.0.dmg signed and notarized (spctl: accepted, Notarized Developer ID, M4LK7C6S84), installed over /Applications/Openscreen.app; About reports 2.0.0 (dmg), Electron 41.2.1, Chromium 146.0.7680.188, Node 24.14.1, darwin arm64 | macOS 26.5 (25F71), Mac mini M1 (arm64), one 3840×2160 display at 2× (UI 1920×1080), no camera (Scaleway MDM allowCamera=false), BlackHole 2ch as the only input | **Partial, 4 defects; MCP section and #1194 first, rest scoped** | Real OS mouse and keyboard through computer-use; MCP calls from @modelcontextprotocol/sdk 1.x (stdio to Openscreen mcp, StreamableHTTP to 127.0.0.1:<port>/mcp with the token copied from AI settings); numbers are ffprobe/ffmpeg on the files. **Pickers:** the first launch (from Terminal, no OPENSCREEN_MAC_SOURCE_PICKER) only served the permissions window; every take used the legacy selector (open --env OPENSCREEN_MAC_SOURCE_PICKER=legacy) or the MCP servers, none went through Apple's picker. **#1194:** keepOnTop is a no-op off Windows, so the bar stays at window level 3 (floating), under the Dock (20) and the menu bar (24); it stays above a frontmost Finder window after the source selector, Device settings and the language menu, keeps its tooltips and clicks, stops under the menu bar and above the Dock when dragged, is pushed down by the language menu near the top and comes back (#1161), and does not float over another app's full-screen Space, as macSpaces.ts documents. **MCP stdio:** help lists mcp; listTools 51 tools; listSources displays, windows, both BlackHole inputs, no camera; window take 1840×872 with no Apple picker, 4 clicks 5 s apart gave 1 zoom (the no-yo-yo rule of resolveShots, not a defect); edit to a closed project, restoreCheckpoint to the raw take, edit without checkpoint refused; exportVideo 30 fps after a 6 s trim, 1085 frames for 36.17 s as predicted; getFrames maps ruler 27 s to file 21 s and warns once the project changed; cursor: "system" take has the system pointer and 0 zooms; camera: true refused (MDM); setPointerTrack 391 samples, 1 zoom, cursor drawn in the export. **#1174:** in-app server off, project open: listProjects open: true, a stdio edit shows live and one Cmd+Z takes it back, an editor change keeps it, restoreCheckpoint lands live and one Cmd+Z restores every edit; app quit: the edit asks for a checkpoint, then lands in the file after createCheckpoint; relaunch and reopen: the same client session edits live again. **In-app server:** off by default with *MCP edits* off; *Running* with port, token (Show, Copy, Regenerate) and snippets; edit, startRecording and exportVideo refused naming *MCP edits*; once allowed, an edit lands live and undoes with Cmd+Z; port 47821 → 47822 then *Apply*: the old URL fails, a client reconnects on the new one. **#1180 tools, both servers:** createProject, setProjectInfo, transcribe (35 English words, about 10 s), getCaptions, setCaptionSettings (bold), getCaptionTranslationUnits fr (4 units), putCaptionTranslation with a French translation written by hand and shown, setEditorSettings (laptop or window frame, wallpaper, 16:9, webcamBackgroundMode: transparent stored with the reason "No clip on the timeline has a camera"), addAnnotation (72 px title, mosaic and smooth blurs), importMedia (MP4 and M4A), insertClip, duplicateClip, setClipRange cropRegion (cropped frame in the export), removeMedia refused while on the timeline and naming the clips, then removing an unused asset, addAutoZooms (0 added, telemetry per asset). **Exports:** MP4 at fps: 30, 1920×1080, 721 frames for 24.03 s, and 1280×720, 1687 frames for 56.23 s, both as predicted; GIF gifSize: "small", gifLoop: false: 852×480, 15 fps, no NETSCAPE2.0 block (#1197); gifLoop on an .mp4 refused; MP4 opened in QuickTime and GIF in Preview, judged by their windows only (the viewers' grant was refused). **CLI:** Openscreen captions <project> transcribes on Metal (11.7 s in 8.4 s), turns captions on, no auto-caption annotation. **App, legacy picker:** permissions window lists the 15.2+ rows (System audio, Accessibility, Microphone, Camera), then Screen & system audio (Required) first under legacy; Device settings lists the inputs and "No camera found" with no quality row; camera toggle refused with "Camera not found."; Record mode rows (Recordings folder last, Auto-zoom only with Editable cursor), *Start recording* hands over to the bar with the same settings; pause stops the timer and resume restarts it; stop opens Edit with 3 zooms; full-screen take 3840×2160 edge to edge, bar and desktop icons absent, system audio −12.6 dB; zoom-in and zoom-out trail tooltips; Glass Lens in preview and in a 720p30 export (1234 frames, true peak −1.6 dBTP, −18.5 LUFS on the clipped fixture voice); *Mouse clicks* off by default, *Click volume* once on, a click at 34.76 s in silence; export dialog (formats, sizes, rates, no codec), progress, *Saved to*, *Show in folder*; Cmd+O lists MCP projects; #997 holds the last clip's last frame at 0:24.0; app menu and menu bar item menus; About; Save Diagnostics writes a bundle. **Defects. (1) P1, MCP in-app startRecording (same as Windows, fixed after this build by c541fffaa):** the bar records within 5 s but the call returns after 120 s with "capture had not started", and on macOS the take is then lost: restore() ends it, no file is left, no editor opens, and stopRecording says "No recording is running." **(2) P2, launch blocked by the stdio server:** while an Openscreen mcp client session runs, opening OpenScreen from Finder or open starts no GUI (exit 0, no window); LaunchServices activates the mcp process instead, which turns Foreground with its own Dock icon after its first take. open -n works around it. **(3) P3, getFrames cannot read a GIF:** with path to a GIF, or with projectId after a GIF export, both servers answer "Chromium could not decode the video (media error 4)", so the documented check after exportVideo fails for GIFs. **(4) P3, system audio on a window take:** an MCP window take with systemAudio: true is silent (−91 dB) while sound plays from another app, because SCContentFilter(desktopIndependentWindow:) keeps only the window's own app; a display take of the same sound reads −10.5 dB. Windows records the whole mix. **Minor:** the camera: true error still carries a JavaScript stack trace; the stop click gets a click sound at the end of the export (40.9 s) although the zoom planner discards it. **Skipped:** Apple's picker slice and the permissions window's prompt rows (operator chose the unattended legacy run; Screen Recording already granted); the system-language prompt (system in English, so it never shows); Spaces switch (one Space); every camera and Full Camera check (MDM); in-app startRecording/stopRecording end to end, its zooms, settings restore and the unsaved-changes refusal (blocked by defect 1, the bar's settings did come back after the timeout); microphone audio (no source plays into BlackHole); Windows, Linux, macOS 13/14, Intel; not run for time: chat, names and terms, transcript editing, clip anchoring, audio tracks, voice-over, caption placement, persistence, editor bounds, GIF timing against v2.0.0, export cancel. | | 2026-10-14 | main @ 2d98601c4 (includes c541fffaa, the recording-state fix), CI artifact openscreen-mac-arm64 from build run 38598557320, DMG signed and notarized (spctl: accepted, Notarized Developer ID, M4LK7C6S84), installed over the build of the row above; About 2.0.0 (dmg) | macOS 26.5 (25F71), Mac mini M1 (arm64), one 3840×2160 display at 2×, no camera (MDM) | **Pass, scoped to defect 1 of the row above** | Real OS input through computer-use, launched with open -n --env OPENSCREEN_MAC_SOURCE_PICKER=legacy (-n because another session's Openscreen mcp was running, #1214); MCP calls from an SDK StreamableHTTP client to the in-app server. **In-app startRecording (c541fffaa):** on a Finder window, with the microphone and the editable cursor requested while the bar had both off, the call returns in 4.9 s (was a 120 s timeout and a lost take); the editor closes, the bar records with the requested settings and one capture helper runs. stopRecording returns in 2.7 s with the projectId of the project the editor opens on, in Edit, one clip, no zoom (Record mode's Auto-zoom after recording is Off), take 1840×872, 26.9 s. The bar's microphone, cursor mode and last source are back to their values from before the take. addZoom on it shows live and Cmd+Z / Cmd+Shift+Z take it back and forth; exportVideo at 30 fps gives 806 frames for 26.87 s as predicted, and getFrames shows the zoom at 8 s. **Skipped:** the refusal while the editor has unsaved changes (the drag meant to hold the editor dirty never moved the pill, so the accepted startRecording that followed proves nothing either way); zooms after an in-app take with Auto-zoom On; the defects of the row above other than defect 1 (#1214, #1215, #1216 still open). | | 2026-10-14 | main @ 61fa43812 (includes #1227, #1228, #1229, the fixes for #1215, #1214, #1216), CI artifact openscreen-mac-arm64 from build run 38615516865, DMG signed and notarized (spctl: accepted, Notarized Developer ID, M4LK7C6S84); app.asar and the capture helper carry the new code; About 2.0.0 (dmg) | macOS 26.5 (25F71), Mac mini M1 (arm64), one 3840×2160 display at 2× | **Pass on the three fixes, 1 new defect, 1 leftover** | Real OS input through computer-use for the Finder launch; MCP calls from SDK clients (stdio to Openscreen mcp, StreamableHTTP to the in-app server); numbers are ffprobe/ffmpeg on the files. Every OpenScreen process was quit first, another session's Openscreen mcp of an older build included, so no old binary could take the launch. **#1214:** with an Openscreen mcp client session open and no GUI, open /Applications/Openscreen.app brings the recording bar up in about 1 s (was: exit 0 and nothing); two quick opens with the GUI running leave one GUI; after a quit, a double-click on OpenScreen in Finder's Applications opens the bar. **#1215:** a GIF export (gifSize: "small", 778×480) read by getFrames through the stdio server by projectId (1, 6, 11 s on the ruler) and by path, and through the in-app server by projectId, gives frames of distinct moments (was "media error 4"). **#1216:** a window take of a Finder window with systemAudio: true while afplay plays speech reads −10.0 dB mean through the stdio server and −10.0 dB through the in-app server under legacy (was −91 dB), video 1840×872, the window only. **New defect (P1 for the in-app server on macOS 15.2+):** on a default launch (no OPENSCREEN_MAC_SOURCE_PICKER), in-app startRecording with window ignores the window and opens Apple's picker ("Share a Window or Entire Screen"), which waits for a click, so the call fails after 120 s with "capture had not started"; the picker then stays on screen after the timeout, and even after the app quits, until someone presses Cancel. The stdio server and a legacy launch are not affected (4.3 s to start). **Leftover from #1214:** the Openscreen mcp process turns from UIElement to Foreground within 0.3 s of the launch LaunchServices hands it, and keeps a second OpenScreen icon in the Dock; a click on it is an activate, so it opens the GUI rather than nothing. A bare hidden Electron 41 with the same handler, with or without a hidden window, stays UIElement, so the cause is in the app's CLI process and not found yet. **Not covered:** the two failure paths of #1229's audio stream (ScreenCaptureKit was not made to refuse it). | | 2026-10-14 | main @ 4f4bb997c (includes #1237 and #1239, the fixes for #1232 and #1233), CI artifact openscreen-mac-arm64 from build run 38627769728, DMG signed and notarized (spctl: accepted, Notarized Developer ID, M4LK7C6S84); About 2.0.0 (dmg) | macOS 26.5 (25F71), Mac mini M1 (arm64), one 3840×2160 display at 2× | **Pass, scoped to #1232 and #1233** | Every OpenScreen process quit before the install, other sessions' Openscreen mcp included; MCP calls from SDK clients; numbers are ffprobe/ffmpeg on the files. **#1233:** with an Openscreen mcp client session open and no GUI, open /Applications/Openscreen.app starts the GUI and the mcp process stays UIElement at 0.5, 2.5 and 8.5 s (was Foreground within 0.3 s, with its own Dock icon). The two OpenScreen icons left in the Dock carry no running dot once the GUI quits: they are the Dock's recent-apps entries, not processes. **#1232:** on a default launch (no OPENSCREEN_MAC_SOURCE_PICKER, so Apple's picker owns the bar's sources), in-app startRecording with window and systemAudio: true returns in 5.0 s with no picker window on screen (was a 120 s timeout and a picker left behind); the take is the window only, 1840×872, system audio at −10.3 dB, and stopRecording returns in 2.7 s. **Not covered:** the refusal message when Screen Recording is denied (the grant was not revoked on this Mac). | | 2026-10-14 | dev, local branch test/shader-fx-macos-e2e: feat/shader-effects-macos (the Metal pass) with #1202 merged in locally, never pushed, plus a local-only line that writes each export's scene to a file. Compositor addon rebuilt from that tree (string probe no Metal device HIT, control wallpaper image HIT), npm run build-vite | macOS 26.5 (25F71), Mac mini M1, on mains power | Partial: agent-written shader effects only, pass | Headless, so no computer-use: electron . --user-data-dir=<test profile> mcp over stdio, and the addon loaded in Node for the live view (createView, setScene, setParam as pushAllNativeParams does, presentTime, readFrame). A 6 s 1920×1080 test pattern with a hand-written .cursor.json of three clicks. **Agent loop:** addEffect with a typo refused in 6 ms at the agent's line 10; the C.6 ripple accepted in 152 ms, real Metal compile included, validated: true, tier light; exportVideo then getFrames show the ripple after each click and nothing at 0.5 s. **Preview against export** (BT.709): 3.1/255 over the whole frame with the effect, 3.0 without; on the pixels the effect changes, 13 to 16/255 against the export with the effect and 14 to 16 against the one without on the same pixels with no effect, which is the codec on sharp edges (two exports of the same project decode to identical pixels). The live view held 29.7 fps against the 30 fps source with and without. **Export** of 180 frames at 1752×1080: 2469 ms without, 2428 with the ripple, 2375 with a heavier effect. **Driver note:** without the setParam calls the editor makes, the headless view draws the cursor at its default size, which reads as a preview/export mismatch. Not run: real capture, the editor UI (effects have no track until the UI PR), the in-app agent, a text-target effect over a real caption (an_annotation_stays_out_of_image_effects_and_inside_text_effects covers the routing of a figure annotation on the Metal device, not a caption). | | 2026-10-15 | installed v2.1.0 candidate, release/v2.1.0 @ 912b44800, CI-built DMG Openscreen-macOS-Apple-Silicon-2.0.0.dmg from build run 38370716184 (openscreen-mac-arm64), Developer ID signed and notarized (spctl: accepted, Notarized Developer ID, M4LK7C6S84), installed over /Applications/Openscreen.app. release/v2.1.0 has since moved to a6cd941f3 with the shader-effects integration, so this build is not its current head | macOS 26.5 (25F71), Mac mini M1 (arm64), one 3840×2160 display at 2×, no camera (MDM), BlackHole 2ch as the only input | Pass, scoped to the fixes listed; no new defect | **Targeted at the fixes that reached the release after bd04d3a7**, real OS mouse and keyboard through computer-use, launched with OPENSCREEN_DIAGNOSTIC=1 and OPENSCREEN_MAC_SOURCE_PICKER=legacy; numbers are ffprobe/ffmpeg on the files. **Passed:** #1136, a HUD dropped at the top (bar y≈60) moves down while the language menu is open and returns to y≈60 when it closes; #1135, typing in Open project until one result is left highlights it and Enter opens it; #1138, with the project open in the editor stdio listProjects reports open:true and an addTrim after createCheckpoint is refused ("this project is open in the OpenScreen editor ..."); #1133/#1144, an identical 1.80x zoom dragged into another stops flush (16947 ms against 16947 ms, length 1727 ms kept), and changing the level of one half changes only that one and splits the pill (#1017); #1134, the first message of a new conversation shows 11:46:52 AM, not a raw ISO time; #1137/#1145, the running Openscreen mcp holds its own cli-sessions/<pid>-.../Local Storage lock while only the GUI holds the GUI's, and a GUI switched to German then relaunched with that server still running comes back in German with no language prompt; #1140/#1150, startRecording on a missing window answers that minimized, hidden, other-Space and behind-full-screen windows are not listed; #1141, the shipped stopRecording description reads "Not one per click ..."; 2c978f3c, the 7-trim speech project exports 3651 frames at 60 fps against an exact 3652.7 for its kept 60.879 s (a per-span ceil would give 3657), audio and video within 16 ms. **Known, already fixed:** while Openscreen mcp ran, opening OpenScreen from the Dock or Finder (open -a) activated the headless MCP process (bare "Quit openscreen" menu, no HUD); that is #1214 and #1233, fixed by 394aebef9, a999816b3 and 4f4bb997c, which landed after 912b44800 and are in release/v2.1.0 @ a6cd941f3, not in this build. **Observations:** the agent's setZoom stretching one of two touching identical zooms into the other merges them (the twin is removed, one undo restores it), which reads as a merge of one pill rather than a defect; the startRecording refusal carries a JavaScript stack trace after its message; the export above is 1.7 frames (29 ms) short of the exact kept duration. **Skipped:** 13366eba and 85aeaf08 (Windows WGC helper only); the rest of the file (rows above and the 2026-10-14 macOS row on main). | | 2026-10-15 | release/v2.1.0 @ a6cd941f3 (advanced to main: shader effects integration #1210, MCP unification and editor parity #1175 to #1186, fixes #1192, #1194, #1197, #1205, #1222), CI artifact openscreen-windows-x64 from build run 38714540858. NSIS sha512 matched update-info-win-x64.json; installed silently per user over the previous build; About still 2.0.0 (intermediate build). | Windows 11 Home 26200 (x64), Ryzen 7 5800X, RTX 4070 Ti, 1920×1080 @ 100 %; Logitech StreamCam | **Pass — new features first (shader effects, MCP), plus a short regression slice; 1 minor** | Real OS input through computer-use. MCP calls come from an SDK client over stdio (Openscreen.exe mcp, forwarded to the running app) and over the in-app HTTP server, whose token was copied from *AI settings*. **Tools:** 55, including addEffect, setEffect, setEffectDefinition and getEffect. **Recording:** a stdio startRecording on a Notepad window, with camera: "StreamCam", system audio and real clicks, gives an 890×596 60 fps take of 35.8 s, with a webcam file, system audio at −3.6 dB and 1 zoom. **Shader effects over MCP:** - a WGSL grayscale with an amount parameter compiles, at cost tier light (1 frame read); - an effect overlapping another is refused ("one effect runs at a time"); - code that does not compile, in addEffect and in setEffectDefinition, is refused with naga's error at the column; - setEffect values apply; - a second region can reuse the effect by effectId; - with no export yet, getFrames renders from the document (#1230): grey inside the region, webcam included, colour outside; - a ctx.clicks tint is red 0.96 s after a click and absent 2.96 s after, as documented (the last 2 s only), both in getFrames and in an MP4 export, where the frame centre at 28 s reads RGB 135/5/19. **Shader effects in the editor:** the Effects lane shows each region. An effect added over MCP to the open project lands live and comes off with one Ctrl+Z. The effect inspector shows the parameter slider, the Image / Text / Both target, the cost and the code; moving its slider to the middle desaturates the preview by half, and getEffect reads it back as amount: 0.48. **Captions:** transcribe finds 25 English words; putCaptionTranslation writes 4 French units; setCaptionSettings (bold, fr) shows them live in the editor preview. **In-app server:** with *MCP edits* turned on, startRecording answers recording: true in 4 s through the recording bar, with its countdown and only the app's own wgc-capture.exe running (#1205). stopRecording returns the 24.1 s take with webcam and system audio, open in the editor under the requested title. **Regression:** the Export dialog's MP4 at 1080p and 30 fps, saved through the native save dialog, gives 1506×1080 at 30 fps, 722 frames for 24.07 s. The MCP export of the same project carries the click sound at both clicks. **Minor:** each getFrames rendered from the document logs Failed to read the click telemetry of <take> ReferenceError: window is not defined (from src/lib/ai-edition/clickSound.ts, run in the main process). The frames are right and the export still has its click sounds, so only the log is affected. Not run: the microphone (no working input on this machine); HUD-started takes, tray, settings, i18n and persistence sections of the checklist beyond the slice above; macOS and Linux. | | 2026-10-15 | v2.1.0-rc.1, release/v2.1.0 @ cd88be2, CI artifact openscreen-windows-x64 from build run 38735485661. NSIS sha512 matched update-info-win-x64.json; installed over the previous build; About reads 2.1.0-rc.1 (Electron 41.2.1, Chromium 146.0.7680.188, Node 24.14.1, win32 x64 · nsis). | Windows 11 Home 26200 (x64), Ryzen 7 5800X, RTX 4070 Ti, 1920×1080 @ 100 %; Logitech StreamCam | **Partial — whole checklist attempted for the release candidate; 5 P2 failures, plus 1 known P2 that predates the candidate** | Real OS input through computer-use; MCP checks from an SDK client over stdio and the in-app HTTP server; audio judged on exports with a 200→2600 Hz sweep as the track (frequency = track time). **Failures:** - *Hide desktop icons*, a known defect from before this candidate (P2, not a regression; desktop_icon_cover.cpp has not changed since 2.0.0): the icons come back at the first click on another window, so a take shows them from then on. A take with no click keeps them hidden. The cover is placed above the icon host once and never placed again. - Full Camera preview (P2, preview only): the webcam picture is torn (a washed-out band over the bottom third) at 31.6 s and 47.5 s, and smeared over one frame at a trim cut; getFrames and the webcam file are clean. - Voiceover export (P2): over a 2× speed region followed by a cut, the piece before the cut runs 0.9 s into the next one and both play at once. Also no fade-out at the take's end (P3). - AI chat rewind (P2): the rewound user message stays in the conversation, and the next turn (*Smart cuts*) executed it again, bringing the rewound annotation and zoom back. - AI chat (P2, 2 turns of 5): operations reported as "applied" never reached the document (an annotation; *Smart cuts*' "added 3 trims"). *Smart cuts* took 3.5 min with no stop control. - A text annotation keeps its box as a share of the frame (P2): after a frame-style change it wraps and is clipped, in the preview and the export render. **Passed:** - HUD: layout toggle, drag, language, camera and mic toggles, *Device settings*, *Camera quality* (4K default; 1080p persists; a hand-edited 720p shows 4K), hide and tray refocus, quit. - Source picker; record, pause, restart, cancel, stop. A webcam take on a sub-4K camera records 1920×1080 at 30 fps. - *Record mode* rows, hand-over, auto-zoom On/Off. *Recordings folder*: pick, persist, reset. - Editor: Media tab, clips (reorder, *Edit clip* grips, ratio lock and *Free*, trim-bar playhead, the rail's clip picker), copy/paste of a clip carrying four modifiers (#1011), transport, end-of-timeline hold (#997). - Regions: every kind; zoom level row on a 1080p take ("Zoom goes from 1× to 2.5×"); 3D camera presets; orbit offering rule; focus drag; *Automatic zooms* (one undo); delete; speed playback. - Modifiers under a trim, and a trim inside Full Camera. Zoom trails with trim and junction hatching, pan trail, #1017, #1008. Clip-anchored modifiers through reorder, split and rejoin, clip delete, *Edit clip* clamp, region copy/paste. - Imported audio: label, inspector, -18 dB / 1 s defaults and reset. Ghost and in → out / length readout on edge drags. Mute absent from the file, loop fills its span, ~10 dB duck under speech. A music bed plays through a cut, by design (#560). A voiceover cut removes its words. It still loads after a restart. - Transcript and captions: seek, cut and undo (#1012), silence trim and restore, word correction surviving *Regenerate*, *Names and terms*, styles, customize, translate and delete, anchors. Caption plate measured on a 1752×1080 export: left edge 174–176 px for 175.2 expected. The packaged-build insertion gate holds. - MCP: edit, record and export are refused without *MCP edits*; edits to the open project land live. - Export dialog. GIF: cancel without a partial file; 1222 frames at 25 fps; plays once. MP4: 1466 packets = ⌈(48.866 − 0.001) × 30⌉, audio in sync, true peak −2.0 dBFS. - Composition, camera layouts, camera background (Cutout/Blur/Custom, in the export), Glass Lens, theme, French UI. - Shortcuts dialog (rebind and reset), wordmark menu, About and Copy, Ctrl+S/Ctrl+O, persistence (document identical after reopen), editor bounds after restart. - Usage ping: usage-ping.json written once on the day despite six relaunches. CLI help lists mcp; a held mcp process owns no window. - Takes: H.264 High, no B-frames, bt709 tags, AAC 48 kHz, hardware encoder. **Not run:** - Every tray context-menu check, the usage-ping checkbox included: right-click on the notification area is refused to computer-use. - Preview audio by ear; the microphone with speech. - Ctrl/Shift+wheel; MP4 cancel (the export finished first); the in-app startRecording flow and stdio quit/relaunch (passed on build 38714540858). - Migrated projects, ONNX removal, data-testid checks. The macOS-only CLI menu fixes (c309edd, 40c01bf, 2d4ac75) have nothing to observe here. - Hardware-specific Windows rows: displays, 96 kHz, C920, Windows 10, ARM. Evidence and the full list: the run's notes. | | 2026-10-15 | installed v2.1.0-rc.1, release/v2.1.0 @ cd88be2 — .deb (`2.1.0rc.1) from build run 38735485661, artifact openscreen-linux(AppImage sha512 matcheslatest-linux.yml), installed over 2.0.0 with apt. The run's first attempt failed on macOS x64 DMG("Validate stapled DMG"); the second published the pre-release without re-running the Linux job, so the published Linux packages are the ones tested here | Ubuntu 24.04, GNOME Shell 46 **Wayland**, AMD Radeon 610M (Vulkan/VAAPI), 1920×1080 @ 100%, USB2.0 HD UVC webcam | **Partial — 6 defects** (all Linux/Wayland; 5 of 6 fixed for rc.2, and the countdown placement in (1) is still open) | **Driver:agent-sh/computer-use-linux 0.7.16 MCP, not the computer-use MCP** (no Claude Desktop computer-use on Linux). Pointer input goes through the RemoteDesktop portal (compositor-level, real OS events), screenshots through the screenshot portal. Its keyboard needs a *second* portal grant that was never answered, so **no key was sent all run**; its drag is press → one jump → release in 70 ms, so **HTML5 drag-and-drop cannot be driven**. Screenshots lag one frame, and **every portal screenshot flashes the GNOME screen white — flashes taken during a take are baked into the capture** (onsets measured in the source); the washed frames in the exports below are those, not an export defect. **Defects:** (1) **the countdown's last digit "1" is recorded in frame 0 of full-screen takes** (2/2 checked), and the overlay is placed by mutter, not centred (top-left on one take, lower-middle on another); window takes are clean. (2) **Tray → Open does not bring back a minimized HUD**: GNOME only shows "“openscreen” is ready" (hud-overlay-hide→minimize(), and xdg-shell has no unminimize); the dock icon restores it. (3) **Vertical HUD runs off the screen**: the window grows to 648×1080 but keeps its top-left (~150,186), so record sits under the dock and Studio/language/hide/quit are below the screen edge until dragged. (4) **HUD cannot be dragged into the top of the screen**: x follows the pointer exactly and the drop point holds, but the bar stops at y≈681 (horizontal) / 593 (vertical) — the transparent reserve is part of the window and mutter keeps the window's top under the top bar. (5) **Editor restore button does nothing** (2 tries; stays maximized) and editor-window.jsonis written with impossible normal bounds —{x:256,y:-10,1440×1121}on a 1080-tall display. (6) **Save Diagnostics bundle is nearly empty on Linux**:recentLogsandmainProcessLogs []despite plenty of main-process output,projectState: nullwith a project open,helperOutputhas onlywindows/mac. Minor: the first frame of a VAAPI take is soft/blocky in its lower half (I-frame 54 KB, crisp by frame 5) and it is the frame the editor opens on; window captures keep the CSD shadow margin as an opaque black border around the window; window takes write no .cursor.json(cursor facet absent). **Since the pass** (merged tomainand cherry-picked torelease/v2.1.0for rc.2, not yet re-verified on a build): (1) the digit in frame 0 → #1267; (2) and (3) → #1272 (on Linux the HUD is hidden instead of minimized, and the vertical bar is kept on screen); (5) → #1271 (works around electron/electron#53063 and stops the saved bounds drifting); (6) → #1268 (main-process logs from launch and the Linux helper's output; the open project stays out of the bundle on purpose); the soft first VAAPI frame → #1269 (QP capped). The shadow border and the missing cursor of window takes are documented as limits in #1270. Still open: (4), which needs a HUD design decision, and the countdown's placement under mutter from (1). **Passed:** single instance (second launch exits 0 in 0.7 s); layout toggle both ways; language menu; minimize keeps the app alive; quit clean; tray menu (Open / Check for Updates / Update Settings ▸ with the 3 modes / Send Anonymous Usage Count / About / Save Diagnostics / Quit), recording state = red icon + only "Stop Recording", Check for Updates back after the take; HUD device panel (mics, live level, camera + preview,Version 2.1.0-rc.1). **Capture:** no in-app source button; the portal picker comes before the countdown, from the HUD and from the Rec stage alike, and **every** start asks again (5×); screen take 1920×1080, 3008 frames / 50.42 s, 0 dropped, vaapi, AAC 48k stereo, cursor + session sidecars, webcam webm 1280×720 30 fps (−73 ms offset); **window take 866×628 = the window, not the monitor**; pause freezes the timer and the file (~66 s wall − ~16 s paused ≈ 50.4 s); restart discards and re-asks the portal; cancel discards the file, mints no project, opens no editor, and leaves no OpenScreen PipeWire screencast node (pw-dump); system-audio-only take carries exactly the 6.04 s test tone (silencedetect); mic-only take (role:microphoneonly) picks the tone up acoustically, +10.4 dB in the 1 kHz band; capture prefs survive a restart. **Editor:** modes, play/pause, ruler seek, playback skips a trim (0:17 → 0:27.1 over ~5 s), trim edge drags, zoom levels / 3D presets / manual focus drag, speed presets, text / blur (mosaic, smooth, oval) / arrow annotations, annotation drag persists, undo/redo, wallpaper / gradient / colour / 1:1 frame, Dual-frame layout, **AI camera Cutout works** (ONNX Runtime bundled), theme light↔dark, locale fr↔en, wordmark menu (About row2.1.0-rc.1, About box with Electron 41.2.1 / Chromium 146 / Node 24.14.1 / linux x64 · deb, Keyboard Shortcuts, AI settings with the provider list and MCP server section), Open project dialog, reopen keeps every modifier, a 2.0.0-era project opens without schema error. **Transcription** auto-runs after stop on whispercpp-vulkan(50.5 s in 6.0 s); a TTS take transcribed word-perfect with silence spans and durations; word click seeks monotonically (0:04.5 → 0:12.5); captions on, Bold, top position; **translation via the configured OpenAI-compatible provider** → "Ceci est un test de transcription locale", back to Original. **Export MP4 1080p60** viah264_vaapi(hardware, dmabuf): 1752×1080, 2613 frames / 43.533 s, AAC 43.550 s (Δ 16.7 ms = 1 frame @ 60), decodes clean, ~33 s for 43.5 s; progress predicted 2617 = (50.474667 clip − 5.584431 trim − 2.577 × ½ speed) × 60, and 2613 = the same with the video stream's 50.4167 s end — the clip is sized to the audio, 3.5 frames longer than the video. Frames show background, card, webcam, cursor, the dragged arrow and the 2.2× auto-focus zoom. **GIF** 20 fps Small, loop off: 778×480, 871 frames / 43.55 s, decoded by ffmpeg and PIL, no NETSCAPE loop block. **Not run:** everything that needs a key (Esc, shortcuts incl. Ctrl+S/Ctrl+O, renaming, custom values, transcript word edit/insert/Backspace, chat and chat-driven editing), HTML5 drag-and-drop (media → timeline, clip reorder, therefore the whole modifier-anchoring section), imported audio / voice-over, unsaved-changes prompts, Check for Updates flows, About → Copy (an unfocused process cannot read the Wayland clipboard), mic + system balance, the CPU-notice path (GPU present). Observations, not filed: no H.264/H.265 choice in the export dialog and no blur intensity / block size / freehand in the inspector (checklist drift?); AI settings shows "–" for the OpenAI-compatible model although the config names one;Ctrl+Shift+Ocannot register (GNOME 46 has no GlobalShortcuts portal). | | 2026-10-16 | installedv2.1.0-rc.1 (cd88be2b1), CI-built DMG Openscreen-macOS-Apple-Silicon-2.1.0-rc.1.dmg from build run 38735485661 (the macOS x64 job failed once on Apple's ticket service, error 68, and passed on rerun; the pre-release then published), Developer ID signed and notarized (spctl: accepted, Notarized Developer ID, M4LK7C6S84); About reports 2.1.0-rc.1 (dmg), Electron 41.2.1, Chromium 146.0.7680.188, Node 24.14.1, darwin arm64 | macOS 26.5 (25F71), Mac mini M1 (arm64), one 3840×2160 display at 2×, no camera (MDM), BlackHole 2ch as the only input | Partial, no defect | **Scoped to what the RC adds over the builds already passed** (rows 2026-10-12 and 2026-10-15), real OS mouse and keyboard through computer-use, launched with OPENSCREEN_DIAGNOSTIC=1andOPENSCREEN_MAC_SOURCE_PICKER=legacy; numbers are ffprobe/ffmpeg on the files. **Usage ping (b5ff5a348):** under DO_NOT_TRACK=1 no ping (usage-ping.json unchanged after 10 s) and no tray item; a normal launch pings within 2 s (daymoves to 2026-10-16, week and month unchanged); the tray's *Send Anonymous Usage Count* is checked by default, unchecking writesenabled:false, checking it again restores true. **Shader effects (Metal):** the chat agent adds a "Black & White" effect on 0:03–0:08, shown as its own pill with an inspector (Image/Text/Both, cost Light, Code, Delete region); the preview is grey at 5 s and in colour at 9.7 s; the 30 fps export measures SATAVG 0 inside the span and about 21.5 outside. **MCP:** with the project open in the editor, a stdio addTrimshows on the timeline at once and Cmd+Z removes it (#1174);createProjectfrom the speech fixture,transcribe(fr, 122 words),setCaptionSettingsbold,exportVideo 720p: the captions are burnt in ("Voilà, alors je fais un petit enregistrement," at 3 s). **Recording:** HUD take of the screen, stop, editor opens; window take of System Settings through Record mode with system audio: 1446×1920 H.264 + AAC 48 kHz stereo, a 1 kHz tone on the speakers is in the take at −34.6 dB band level (aa17b10a9). **Export:** GIF with Loop off carries no NETSCAPE block (173f615d0). The CLI menu, #1214 and #1233 were passed the day before on builds with the same CLI code. **Skipped:** the in-app HTTP MCP server's new record and export paths (no HTTP MCP client here); Apple's picker slice; camera checks (MDM); the rest of the file (rows above, including the 2026-10-14 full pass on main). **Checklist gap:** no section covers shader effects or the usage ping yet. | | 2026-10-16 | v2.1.0-rc.2, release/v2.1.0@c5bc8be(main + bump), CI artifactopenscreen-windows-x64from build run 38868631147. NSIS sha512 matchedupdate-info-win-x64.json; installed silently over rc.1. | Windows 11 Home 26200 (x64), Ryzen 7 5800X, RTX 4070 Ti, 1920×1080 @ 100 %; Logitech StreamCam | **Partial — targeted pass on the rc.1 fixes; all 7 checked items pass, preview smoothness measured with one known gap** | Real OS input through computer-use; MCP edits and takes from an SDK client over stdio. Each item was rerun on the scenario that failed on rc.1 (#1258). **Passed:** - *Hide desktop icons* with wgc-capture.exegranted to computer-use (#1263): a 77 s full-screen take with clicks and typing in Notepad shows no icon in any frame checked (5, 40, 58, 66, 72, 76 s). - Chat rewind (#1259): the rewound message leaves the conversation andchat.json. Its text goes back to the composer. The next turn ("add a zoom from 30 to 32 s") brings back nothing from the rewound request. - Chat operations (#1261): a two-operation turn (annotation + 2× speed) and *Smart cuts* on a 37 s TTS-narrated take. Every "applied" line is in the document: 4 trims announced, 4 trims present, none over a transcribed word. The *Smart cuts* turn took about 1 min. - Text annotation refit (#1262): "Refit check" stays on one line, unclipped, with the Screen frame style and with 9:16, in the preview and in the export render (getFrames). - Full Camera at a cut (#1266): three plays across a 1 s trim inside a Full Camera segment, 13 frames sampled, no trail or ghost. The export render at 27.00 and 27.02 s, just after the cut, is clean. Parked inside the trim's hole, the preview shows the webcam in its corner: the region is split at the trim, as #1266 describes. - getFramesrendered from the document (#1253): nowindow is not definedand no click-telemetry error in the main-process logs. - *Save Diagnostics* from the tray, clicked by Etienne (right-click is refused to computer-use): the bundle carries2.1.0-rc.2, OS and Electron versions, the Windows helper output and 54 main-process log lines. projectStateis null and the open project is absent. **Preview smoothness (#1189), measured with a reservation:** the perf bench was running its own OpenScreen instance at the same time, with the screen probably at 60 Hz instead of 280 Hz; treat these figures as indicative. DXGI capture of the preview area at 60 fps during playback, counting runs of identical frames. Reference: an ffplay 30 fps pattern, captured the same way, shows one 67 ms gap in 10 s. - Normal take (60 fps source, webcam, two zooms one in 3D, a shader effect): no gap above 83 ms. 22 gaps of 67–83 ms in 14 s, grouped around the zoom transitions (6.0 s and 11.5 s). - Held-frame take (webcam, shader effect, orbit zoom, Full Camera): during Full Camera, recurring 67–100 ms gaps. This matches the known webcam + held-frame defect slated for rc.3. Long freezes elsewhere fall on static content and cannot be judged this way. - Cursor recoil: not judged. **Also seen:** -startRecordingwithcamera: trueover MCP picked the system default camera, a virtual "VCam Camera", and the take came out with no webcam. - The torn webcam frames of #1284 are in the recorded webcam file itself (plain ffmpeg decode), not in the preview. The preview andgetFramespick different frames at the same second, which is why the export looked clean on rc.1. **Not run:** the EXDEV fallback (#1256, Store MSIX only); the doubled voiceover (#1195, not fixed); the rest of the checklist (rc.1 row). | | 2026-10-16 |v2.1.0-rc.3, release/v2.1.0@523a3e4, CI artifact openscreen-windows-x64from build run 38878886935. NSIS sha512 matchedupdate-info-win-x64.json; installed silently over rc.2. | Windows 11 Home 26200 (x64), Ryzen 7 5800X, RTX 4070 Ti, 1920×1080 at 279 Hz, then 60 Hz; Logitech StreamCam | **Partial — diff since rc.1 (brief items 1–8 and 15); all pass except preview smoothness at 60 Hz: the preview still steps back one frame** | Real OS input through computer-use; edits and takes from an SDK MCP client over stdio. Three takes: - full screen with webcam, the screen held still while the cursor moves; - a testsrc2window at 30 fps with the StreamCam; - a full-screen TTS-narrated take. **Smoothness (#1189, #1282, #1285)**, measured on a DXGI capture of the preview at 60 fps during playback. A step back is a frame identical to the one two or three frames earlier while the frame between differs. It is counted on full-resolution frames and checked by eye on the source's frame counter (664 → 663 → 664). A downscaled first pass also flagged false positives. An ffplaytestsrc2reference captured the same way shows none, at 279 and at 60 Hz. Each project has the webcam, two zooms (one 3D) and a shader effect. - At 279 Hz, normal take: no step back in two runs, no gap above 83 ms past playback start. - At 279 Hz, held-frame take: no step back. In the webcam area the preview changes as often as the webcam file itself and pauses when it pauses, so the camera keeps moving through held frames (#1282). - **At 60 Hz, step backs remain (#1285 does not cover them): normal take 7 confirmed in 12.8 s, held-frame take 10 in 20 s.** No gap above 67–100 ms on the normal take. **Passed:** - Rewind (#1259): the rewound request leaveschat.jsonand is not replayed by the next turn. - Chat operations (#1261): a two-operation turn lands both. *Smart cuts* announces 4 trims and places 4, none over a transcribed word. - Text annotation (#1262): "Refit check" stays on one line, unclipped, with the Screen frame at 9:16, in the preview and the export render. - Full Camera cut (#1266): no blur trail in the preview or the export render right after a 1 s trim. The torn webcam frame visible there is in the webcam file itself (the camera, see #1284). - Copy a pill (#1245): clicking half of a 2 s pill made of two touching 1 s speed regions and pasting gives a 2 s region. -getFramesfrom the document (#1253): nowindow is not definedand no click-telemetry error in the logs. - *Save Diagnostics* from the tray (clicked by Etienne):2.1.0-rc.3, versions, helper output, 27 main-process log lines, open project absent. projectStateis null on this path by design; the in-app path that attaches the document was not exercised. - *Hide desktop icons* (#1263):wgc-capture.exegranted, cursor moves and three clicks on the desktop; no icon in any frame checked from 3 to 31 s. **Not run:** - the Store MSIX (#1256), macOS (item 9) and Linux (items 10–14), handled elsewhere; - cursor recoil judged separately from the full-frame step-back count. | | 2026-10-16 |v2.1.0-rc.3, release/v2.1.0@523a3e4, CI artifact openscreen-linux from build run 38878886935 (Linux packagessucceeded; the run is red onmacOS x64 DMG, so Publish GitHub releasewas skipped).Openscreen-Linux-2.1.0-rc.3.AppImagesha512 matchedlatest-linux.yml; run directly, not installed (the .debneeds a sudo password; FUSE3 mounts it, AppRun adds--no-sandboxbecause Ubuntu 24.04 restricts unprivileged user namespaces). Save Diagnostics reports2.1.0-rc.3, channel: appimage. The rc.1 .debstayed installed for A/B runs | Ubuntu 24.04, GNOME Shell 46 **Wayland**, AMD Radeon 610M (Vulkan/VAAPI), 1920×1080 @ 60.001 Hz, USB2.0 HD UVC webcam | **Partial — targeted pass on the diff since rc.1: the 13 listed items pass on their scenario (item 1 only after a session reload); 2 pre-existing P2 defects found, neither a regression from rc.1** | **Driver:** thecomputer-use-linux0.7.16 MCP that Claude Code starts over SSH inheritsXDG_SESSION_TYPE=ttyand noWAYLAND_DISPLAY, so it has no portal input and its click falls back to a missing ydotool. The same plugin binary, started from the shell with XDG_SESSION_TYPE=wayland WAYLAND_DISPLAY=wayland-0behind a small socket bridge, got one RemoteDesktop grant (pointer and keyboard) for the whole run. Its click takes a portal screenshot first, so **the whole screen flashes white ~100 ms on every click** (scroll, drag and keys do not); no click fell inside a measured playback window. **Defects:** (a) **Chat rewind fails on a message sent in the current view** (P2, pre-existing): ↶ → Rewind shows the toast "Message not found in this session.", andchat.jsonand the document stay unchanged. 6/6 attempts across 2 conversations, by the driver and by Etienne's own click. Cause, read in the code:LeftPanel.tsxseeds the user message aslocal_u and never swaps in the main-process id (randomUUID(), chat-service.ts:397). Messages are only reloaded from the main process when the active session changes, so rewind works again after switching conversations. rc.1 has the same code. Also, clicking ↶ scrolls the conversation to the bottom (the scroll effect has no deps), so the confirmation hangs detached at the top, clipped by the window edge. (b) **Preview after a seek on a held-frame (VFR) take** (P2, pre-existing): seek to 30 s and play, and the preview updates 1–2×/s for the rest of the clip (stills of 0.5–1.7 s, 2 A-B-A step-backs at the start) while the clock runs normally. Same with GOP 600 and GOP 60, and identical on rc.1. The CFR take after the same seek is fine, and the VFR take played from 0 is fine. **Passed:** - #1259 rewind, after a session reload: ↶ → Rewind on a two-operation turn shows "Rewound to the start of that message", puts the text back in the composer, and removes the annotation, the speed region and the later zoom. chat.jsonis rewritten without the message, which stays out after an app restart. The next turn ("Add a zoom from 10 to 12 seconds") leaves only that zoom in the document. - #1261 Smart cuts (wand → Smart cuts) on a 32.4 s espeak-ng-narrated clip with a local Whisper transcript (63 words): 3 dead-air cuts announced (4.99–6.94, 12.16–14.59, 25.56–27.52), exactly those 3 trims in the document. A typed two-operation turn (text "Hello" 1–3 s + 2× speed 15–20 s): both "applied" lines are in the document. - #1262 "Refit check" (64 px bold) stays on one line, unclipped, through Style None → Screen (Auto 1920×1508) → 9:16, in the preview and ingetFrames(1080×1920). - #1266 Full Camera 23–30 s with a 1 s trim at 26–27 s, motion blur 0.2:getFramesat 25.95/27.00/27.02/27.05/28.00 s is clean, and so are paused frame steps across the cut and a 60 fps capture of the preview playing across it (488 frames, no ghost; the only large change is the region-end transition at 30 s). - #1245 Copy on the fragment after that cut, paste at 31 s: region 30.974–37.974 s = 7.0 s, the whole pill, where the fragment is 3 s. - #1253getFramesrendered from the document (no export) over stdioopenscreen mcp, 7 frames in 2 calls: no window is not defined in the CLI's or the app's log. - Save Diagnostics from the tray with a project open: 997 bytes before any take (mainProcessLogs3,helperOutput.linuxempty,projectStatenull,recentLogs[]). After a take in the same instance: 5428 bytes,helperOutput.linux1418 bytes (ready → capture-stopped, no cursor samples), 17 main-process lines including the helper's stderr,projectStatenull, no error (rc.1: 541 bytes). - Countdown: frame 0 of a full-screen take shows no digit (frames 0–5 identical at the centre; the first frame lands 2.7 s after Share). The overlay is still not centred ("3" at ≈(360,310)), the part known to be open. - HUD: "Hide recording bar" destroys the window (no dock icon, process alive) and tray → Open brings it back. Switched to vertical, the bar fits on screen (y 472–956). The rc.1 binary on the same machine still runs off the bottom. - Editor: opens maximized after a take; restore un-maximizes to 1320×964 (2/2, one after a maximized reopen).editor-window.jsonkeeps 1320×964 over 3 launches (normal close{96,54,1320×964,false}, maximized close {316,62,…,true}), so no growth. - VAAPI first keyframe: full-screen take I-frame 204 KB, QP 13–21. Terminal window take at 3.1 Mbit/s (the rc.1 case), 1920×982: I-frame 185 KB, QP 13–30 with the cap reached in the lower half, text legible. **Preview smoothness** (#1189, #1282, #1285). Method: OpenScreen's PipeWire helper run standalone as a 60 fps monitor capture (VAAPI dmabuf, cursor hidden), counting runs of unchanged frames and A-B-A step-backs in the preview canvas (748×460, maximized editor). Reference: ffplay playing a 1080p60 pattern, captured the same way, changes at 60 Hz. Source: a 40 s 1080p60 take of a moving test pattern with two 8 s frozen segments, webcam PiP, pointer moving throughout. Project: 1.8× zoom 8–11 s, 2.2× orbit (3D) zoom 15–19 s, WGSL grayscale effect (light) 1–39 s, motion blur 0.2. - Normal (CFR) take, 2 full plays: **30 distinct frames/s** on the 60 Hz display throughout. Stills: 34 ms ×931, 51 ms ×79, 68 ms ×6. The only longer ones are a 350 ms stall at play start and a 2.9 s span where nothing in frame moves (pointer outside the 2.2× zoom per its telemetry, frozen pattern, still webcam scene). 0 step-backs; no still above 68 ms around the zoom transitions. GPU busy 30–55 %, so the 30 fps is not GPU saturation. **rc.1 on the same project is identical** (30 fps; 34 ms ×938, 51 ms ×77, 68 ms ×5; 383 ms start; 0 step-backs): the cadence is not new and rc.3 does not regress it. - Held-frame take (the same take made VFR with mpdecimate: holds of 6.7 s and 8 s, last frame at 34.9 s, clip 40 s, silent AAC to 40 s), played from 0 on rc.3: the preview keeps changing 30×/s through both holds (cursor and webcam move); max still 83 ms (once); 0 step-backs. In the trailing 5 s hold the pointer keeps moving until 0:40.0 (02dbf13e). Without the audio track (container ends at 34.93 s) playback stops at 0:34.9. **Also seen, not filed:** - stdio listSourcesraises the GNOME Share Screen picker and returnsdisplays: []("Failed to get sources" if cancelled), so a followingstartRecordingraises a second picker. An MCP take has no countdown, so its first frame shows the picker fading out. - Closing the editor after a take quits the app ("the last window closed"), because the HUD is closed while the editor is up. - Tray → Open on a HUD that is open but buried shows "“openscreen” is ready" instead of raising it (once). - In 2 of 6 relaunches in vertical layout the HUD's last three buttons sat under the dock until it was dragged up. It still cannot go above y≈472 (rc.1 (4)). - Save Diagnostics leaves a Files window revealing the bundle. - Chat replies render markdown tables as raw text. - The app quit once on its own at 16:08:52 ("[app] quitting", clean, no crash) while text was being typed into the chat composer. Cause not established; Etienne was also at the screen. **Not covered:** Save Diagnostics from the menus (on Linux the editor is frameless with no native menu bar, and the OpenScreen menu has no such entry); the 64 KiBhelperOutputcap (not reached at 1418 bytes); the in-app HTTP MCP server; the.debinstall path; a full MP4 export; everything outside the 13 listed items (rc.1 row). Takes, projects, chat files, audio peaks, media-link entries and bundles from this pass were deleted, and~/.config/openscreenwas restored from a backup taken before the pass. | | 2026-10-16 |v2.1.0-rc.3, release/v2.1.0@523a3e4, DMG Openscreen-macOS-Apple-Silicon-2.1.0-rc.3.dmgfrom artifactopenscreen-mac-arm64of build run 38878886935 (attempt 1). The pre-release was not published: the macOS x64 job failed on "Validate stapled DMG". DMG copied over by scp, SHA-512 identical to the source;spctlaccepted, Notarized Developer ID (M4LK7C6S84);stapler validateOK.CFBundleShortVersionString2.1.0-rc.3. The Save Diagnostics bundle reports2.1.0-rc.3·dmg, Electron 41.2.1, Chromium 146.0.7680.188, Node 24.14.1, darwin arm64 (About window not opened) | macOS 27.0.1 (26A434), Apple M1 (arm64), 8 GB, one 1920×1080 display at 120 Hz (1×); OBS Virtual Camera (no physical camera), BlackHole 2ch | **Partial — targeted pass on the changes since rc.1; 7 of 9 items pass, 2 P2 defects: #1266 in the export, rewind in the open conversation** | Real clicks and keystrokes through computer-use. Launched with open -a, OPENSCREEN_MAC_SOURCE_PICKER=legacy, OPENSCREEN_DIAGNOSTIC=1and--remote-debugging-port, the last only to read the scripts/perf-ab/probe.jsprobe (no injected action). Measured with theffmpegbundled in the app. avfoundation screen captures of the preview rectangle (120 fps asked, 75–79 fps obtained). Two HUD takes with the OBS webcam, microphone and system audio. **Defects:** - **#1266 in the export (P2)**: Full Camera 8.01–14.06 s with a cut at 10.00–11.39 s. The preview is clean at the cut (capture read frame by frame). In the exported MP4 (1080×1920, 60 fps, 1544 frames, 25.73 s = 27.12 − 1.39), frame 600 (t = 10.000 s, right at the cut at 10.001 s) shows the trail: the full-frame camera is doubled into several copies sliding toward its PiP corner. Frames 597–599 and 601–603 are clean.getFramesrenders the trail at 10.000, 10.001, 10.017 and 10.033 s; 9.967 and 9.983 s are clean. It persists without the shader effect and in the Auto frame. On another project (Full Camera 9.95–17.95 s, cut at 13.93 s), the preview andgetFramesat 13.90, 13.95 and 14.00 s are clean, but no sample lands exactly on the cut. - **Rewind (P2, predates rc.1)**: rewinding a message sent in the open conversation fails with "Message not found in this session." (2 tries) and undoes nothing. The renderer keeps the optimisticlocal_u id (LeftPanel.tsxl. 604–612); the real ids are only reloaded on a conversation switch. After Conversation 2 → Conversation 1, rewind works. This path dates frombe479097, so not a regression (fixed since by #1292). **Passed:** - Rewind #1259, once the ids are reloaded: the message leaves the conversation and chat.json, its text comes back in the composer, and the "Rewind me" annotation disappears. The next turn ("zoom 20–22 s") adds only that zoom (zoomRanges20000–22000), nothing from the rewound message. - Operations #1261 (DeepSeekdeepseek-v4-pro): everything the agent reports "applied" is in the document. B&W effect 2000–9000 ms and 2× speed 20–23 s (speedRegions), B&W effect 5–12 s, zoom 20–22 s. *Smart cuts* (Auto-enhance menu): "added 4 trims", 4 trims present, none on a word (3.00–3.28, 11.83–12.46, 20.02–20.80, 24.30–24.54 s), about 2 min. - Annotation #1262: "Refit check rc3" at size 48 stays on one line, not clipped, in Auto then 9:16, in the preview (2 projects), the 9:16 MP4 export (frame 102, 1.70 s) and getFrames(0.5 s). - Pill copy #1245: a zoom cut by a clip boundary (second clip added from Media) gives two fragments, 23.866–36.330 and 36.330–45.080. Click on the 8.75 s fragment, Cmd+C, Cmd+V: the copy covers 50.089–71.303 = 21.214 s, the whole pill. A cut inside a single clip also gives 11.127 s = 11.127 s. - Save Diagnostics (OpenScreen app menu): bundle with2.1.0-rc.3, OS, versions, helperOutput.mac(811 chars) and 14 lines ofmainProcessLogssince launch.projectStateis null and the open project is absent. The brief's "editor document present" was removed on purpose bydad3cca0. - getFrames #1253 through a stdio client (Openscreen mcp) on the 2 projects, open and closed: frames rendered, no "window is not defined" or telemetry error in the main log or the MCP process's stderr. - macOS preview #1257: on Apple Silicon the Metal preview shows and moves (figures below). The Intel path is not testable here. **Preview smoothness (#1189, #1282, #1285), measured:** - Normal take (screen 1920×1080 at about 57 fps, webcam 30 fps, B&W 2–9 s, two 1.50× zooms, Full Camera, cut), playback 0–19 s. Probe: 587 paints in 19.8 s = 29.6/s; p50 33.3 ms, p95 34.8, p99 36.1; the only paint above 50 ms is the pause; 0 long task; all through readback (0 shared). Capture: 0 frame stepping back (A-B-A pattern); no gap above 45 ms around the zoom-in (5.15 s) and the zoom-out (16.28 s). Gaps of 67–123 ms appear only during Full Camera, where the OBS webcam source repeats its own frames (22.5 distinct frames/s, 26 gaps above 67 ms over 8–16 s). - Held-frame take (still screen: 8.7 distinct frames/s in the take, held up to 0.9 s; editable cursor; webcam; 1.80× zooms; B&W 5–12 s). Probe: 29.3/s, p95 34.5 ms. Capture: 28.6 visible changes/s, so the cursor and camera keep moving; 0 step back; largest gap 100 ms (twice), during a pointer pause where only the webcam moves. - Caveat: the preview paints 30 times a second on a 120 Hz display with a source at about 57 fps, and the capture tops out at 75–79 fps. **Also seen:** local transcription runs on whispercpp-metal(36.4 s of audio in 16.3 s). The export pipeline decodes H.264 in software.getFrames reports "clip #1 shortened by 0.113s": the clip added from Media lasts 36.33 s, the stream 36.217 s. **Not covered:** Apple's picker slice; the tray menu (no right-click); the About window; a physical camera; the Intel path; the rest of the checklist (earlier rows). The Mac restarted during setup; the whole pass ran after it. The test takes, projects and exports are in the Trash (/.Trash/openscreen-rc3-pass-1752/); the HUD's camera and microphone are back to Off. | | 2026-10-16 | v2.1.0-rc.4, release/v2.1.0@89cdff2, CI artifact openscreen-windows-x64from build run 38948496658. NSIS sha512 matchedupdate-info-win-x64.json; About reads 2.1.0-rc.4. | Windows 11 Home 26200 (x64), Ryzen 7 5800X, RTX 4070 Ti, 1920×1080 | **Pass — quick check of chat rewind (#1292) only** | Real OS input through computer-use, on a fresh 10 s take, with the chat provider deepseek-v4-flash. Each rewind ended with "Rewound to the start of that message": the message left the conversation and chat.json, its text returned to the composer and its edits were undone. No "Message not found" in the UI or the logs. - Rewind right after sending, with the agent still thinking: no edit arrived afterwards. - Rewind of a turn the agent had already answered (one zoom added): the zoom was removed. - Confirmation opened while the agent was answering and confirmed after its reply (three zooms added): all three were removed. **Not run:** - The hidden-rewind case (a turn with no checkpoint). With a document open, even a text-only turn ("pong") records a checkpoint, so its ↩ is shown and the rewind works. The UI offers no editor state without a document to test the no-checkpoint case. - The rest of the checklist (rc.3 row). | | 2026-10-17 | v2.1.0-rc.5, release/v2.1.0@3db4c7c, installer Openscreen.Setup.2.1.0-rc.5-x64.exefrom the GitHub release (build run 39003658928). sha512 matchedlatest.yml; installed silently over rc.4; About reads 2.1.0-rc.5. | Windows 11 Home 26200 (x64), Ryzen 7 5800X, RTX 4070 Ti, 1920×1080 at 60 Hz; Logitech StreamCam | **Partial — diff since rc.4 (#1294, #1296) and preview smoothness. #1294 still trails at one cut; #1296 passes; one step back left at 60 Hz** | Edits and exports from an SDK MCP client over stdio; real OS input through computer-use in the editor. **Full Camera at export (#1294), fail on one cut.** Project with a webcam, two clips (the second a duplicate) and five editor trims, with six trims added at inexact bounds. Two Full Camera spans cover seven cuts: four trims, one clip junction and two clip starts. The project was exported at 1080p60 and scanned frame by frame for a one-frame outlier. The first frames after six cuts are clean. The first frame after the trim ending at source 46.34986635 (a start that rounds down in f32) is ghosted: the camera smeared from its bottom-right PiP corner to full frame, the trail #1294 describes. An rc.4 export of the same project, run from the rc.4 installer's payload, is byte-identical to the rc.5 export, so the fix changes nothing here. Two other starts that round down in f32 (47.293 and the second clip's 9.334) were clean in both builds. **Pause on a held frame (#1296), pass.** Still-screen take with the cursor circling, webcam, two zooms (one 3D) and a shader effect. Two trials, each pausing on a still passage during playback: - paused at 0:07.7, then the wallpaper changed; - paused at 0:13.0, then the background blur changed, then the format (4:5). The playhead kept its time, and the cursor and the webcam did not move. On a 60 fps capture of the preview, the blur change only touched the background band. **Smoothness at 60 Hz, seek then play,** measured as for rc.3: full-resolution step backs on a DXGI capture at 60 fps, with frames decoded without duplication. Both projects carry the webcam, two zooms (one 3D) and a shader effect. - Held-frame take, 22 s: no step back, no gap of 83 ms or more (rc.3: 10 step backs in 20 s). - Normal take (testsrc2window at 30 fps), run 1: one whole-frame step back in 16.6 s during a zoom (frame 158 identical to frame 155). Two gaps, 100 and 83 ms. - Normal take, run 2: no step back. Gaps of 83 to 100 ms past 11.7 s. The take itself holds a frame for 50 ms about every 0.25 s. rc.3 had 7 step backs in 12.8 s. **Not run:** - the ffplay reference capture at 60 Hz (rc.3 showed none); - #1294 in the preview (export only); - the rest of the checklist. | | 2026-10-17 |v2.1.0-rc.5, release/v2.1.0@3db4c7c, DMG Openscreen-macOS-Apple-Silicon-2.1.0-rc.5.dmgfrom the GitHub pre-release (build run 39003658928): 140 610 214 bytes, sha256 matches the release asset's digest;spctlaccepted, Notarized Developer ID (M4LK7C6S84);stapler validateOK;CFBundleShortVersionString2.1.0-rc.5. Installed over rc.3 | macOS 27.0.1 (26A434), Apple M1 (arm64), 8 GB, one 1920×1080 display at 120 Hz (1×); OBS Virtual Camera (no physical camera), BlackHole 2ch | **Fail — targeted pass on the diff since rc.4 (#1294, #1296); 1 P2 defect: #1294 does not fix the Full Camera trail** | Real OS input through computer-use, launched withOPENSCREEN_MAC_SOURCE_PICKER=legacy, OPENSCREEN_DIAGNOSTIC=1and--remote-debugging-port(read only, for thescripts/perf-ab/probe.jspaint probe). Measured with the app's bundledffmpeg. Preview captured with avfoundation (120 fps requested, 75–77 fps achieved). Nobody else used the machine during the measures. Two HUD takes with the OBS webcam and microphone: a normal one (60 fps test pattern, 35.2 s, three automatic zooms) and a held-frame one (still screen: 9.3 distinct pictures/s, held up to 1.0 s; editable cursor moving with 3 s pauses; 24.3 s). **1. Full Camera at a cut (#1294) — fail.** Held take, Full Camera 2.961–21.167 s, four 1.24 s trims inside it, cuts at 5.043, 7.810, 10.616 and 13.396 s on the timeline. Exported MP4 1886×1080, 60 fps, 1157 frames, then compared each cut's first frame (303, 469, 637, 804) with its neighbours. Frames 469 and 804 show the trail: the full-frame camera in several copies sliding toward its PiP corner, differing from both neighbours (mean grey diff 31 and 30–33) while the neighbours differ by 5–9. Frames 303 and 637 are clean. The pattern fits millisecond rounding, not f32: timelineMap.tsemits each Full Camera piece withstartMs: Math.round(srcStartSec * 1000), while the clip keeps its exact start. When the rounding lands the piece's start after the clip's start (10.292957 → 10.293, 18.361809 → 18.362), the previous frame, held at the clip's start by 297dfeac/#1294, falls before the piece, where the camera is in its corner. When it lands before (6.284394 → 6.284, 14.340314 → 14.340), the cut is clean. This also fits the two rc.3 cuts. #1294's f32 explanation was a coincidence: 18.361809 rounds up in f32 and still ghosts. **2. Pause on a held frame (#1296) — pass.** Held take, played from 1.5 s, paused at 0:04.0 while the pointer was moving over a held screen frame, then lowered the padding (50 % → about 20 %), picked another wallpaper, then a solid orange background. After each change the playhead still reads 0:04.0 and the pointer keeps its position relative to the recording; nothing steps back. #1296 targets Linux, where the seek lands on the held frame; on macOS it lands at or after the target. **3. Preview smoothness after a seek — pass.** Normal take, seek to 5 s then play 16 s: probe 503 paints in 17.8 s = 28.2/s, p50 33.3 ms, p95 36.2, p99 39.5, readback only, 0 long tasks (the only interval over 70 ms is the wait between seek and play). Capture: 26.7 visible changes/s, no gap over 56 ms, 0 steps back. Held take, seek to 3 s then play 16 s: probe 28.3/s, p95 34.6 ms, nothing over 40 ms during playback. Capture: 0 steps back. Whole-frame gaps of 112–279 ms around 9–10 s fall on a pointer pause where only the small PiP webcam moves; measured on the PiP alone, 26.1 changes/s, max 78 ms. In line with the rc.3 measures; no new hitch. **Not covered:** Apple's picker; a physical camera; the Intel preview path; GIF export; the rest of the checklist. Takes, projects, export and DMG moved to the Trash; HUD camera and microphone set back to Off. | | 2026-10-17 | v2.1.0-rc.5, release/v2.1.0@3db4c7c, Openscreen-Linux-2.1.0-rc.5.AppImagefrom the GitHub release (build run 39003658928, all jobs green). sha512 matchedlatest-linux.yml; run directly, not installed. For A/B on the same machine and the same projects: the rc.4 AppImage from its release (sha512 matched) | Ubuntu 24.04, GNOME Shell 46 **Wayland**, AMD Radeon 610M, 1920×1080 @ 60 Hz, USB2.0 HD UVC webcam (10 fps in a dark room) | **Partial — diff since rc.4: #1296 passes (pause and seek on held frames); #1294 still trails at 3 of 5 cuts; no smoothness regression** | **Setup.** Driver: the computer-use-linuxbridge of the rc.3 row, one RemoteDesktop grant. Takes are real PipeWire takes recorded through stdioopenscreen mcp startRecordingwithcamera: true, with the app quit first. While the app runs, its video-capture service keeps /dev/video0open even with the HUD hidden, and a take comes out with no webcam file and no error. **Finding: a real Linux PipeWire take has no long held frame.** A still screen with a still pointer, for 22 s and then 13 s, still gave a packet every 17–33 ms (no gap above 50 ms; 3201 frames in 55.2 s): the helper encodes a frame at every tick. On Linux the held-frame path (#1296) is therefore reached only with VFR media, imported or recorded on Windows or macOS. The rc.3 "1–2 updates/s" was measured on such a file. **The held-frame take used here** is the real webcam take turned VFR with mpdecimate, built like a PipeWire take (no B-frames, GOP 30, fragmented MP4, silent AAC to the end): holds 0.04→17.44 s and 21.40→33.47 s, last screen frame at 37.42 s, clip 51.18 s. **1. Full Camera at export (#1294): fail at 3 of 5 cuts.** Real webcam take with Full Camera 8.006–40 s crossed by five trims: 10.001–11.388, 15.2–16.0833, 20.5–21.7177, 26.3–27.4567, 31.111–32.2229. Motion blur 0.35. Exported at 1080p60 with MCPexportVideo(2969 frames, 49.47 s) and scanned frame by frame for one-frame outliers. - rc.5: the first frame after cuts 3, 4 and 5 (frames 1094, 1369 and 1588; 3.6, 4.7 and 0.4 ms after the cut on the output clock) shows the full-frame camera smeared toward its PiP corner. Cuts 1 and 2 are clean, and there is no other outlier in the file. - rc.4, same project: outliers at frames 600, 1094, 1369 and 1588. - So #1294 fixes only the PR's case (clip start 11.388, frame 600). Of the three starts still trailing, 21.7177 and 27.4567 round down in f32 and 32.2229 rounds up: f32 rounding of the start does not explain them on its own. The Windows rc.5 row also has one cut left. **2. Pause on a held frame, then a setting change (#1296): pass.** Held-frame take: seek to 24 s, play, pause at 0:27.6 (inside the 21.40–33.47 hold), wait 2 s, change the wallpaper, then the background blur. The playhead stays at 0:27.6, and the cursor stays where the telemetry puts it at 27.6 s (cx 0.189). rc.4 with the same steps: from the settle pass on, the cursor sits at its 21.40 s position (cx 0.551) while the playhead reads 0:27.6. **3–4. Seek then play.** Method: a helper capture of the canvas at 60 fps, as in the rc.3 row, counting changes per second, stills and A-B-A step-backs. Each run seeks by a ruler click and then plays, twice: into a still stretch at 26 s with 9 s of play, then into the final still stretch at 45 s, playing to the end. - **Held-frame take:** rc.4 (2 runs) gives 22–28 changes/s, with 100–133 ms stills recurring every 0.5 s (14 in the 9 s window: the drift watch re-anchoring) and 6–8 after the 45 s seek. rc.5 (2 runs) gives 29–30/s: a single 133 ms still 0.3 s after play, then nothing above 68 ms, and no step-back. - **Seek past the last screen frame** (45 s > 37.42 s): rc.5 holds 30/s and the cursor follows to the clip's end. - The 200–390 ms stall at play start is the same in both builds. - **Real (CFR) take:** rc.4 (2 runs) gives 29–31/s, stills mostly 34 ms, 150–267 ms at play start, no step-back. rc.5 (4 runs, one of them a fresh launch) has the same profile. One run, the first after a launch, measured 19–26/s with 51–83 ms stills, and the next three did not reproduce it. One A-B-A at play start in 1 of 4 rc.5 runs, none in rc.4. - So point 3 shows no new hitch, and the only step-back is that one at play start; point 4 is smooth. The preview still runs at ~30 distinct frames/s on this iGPU at 60 Hz in both builds (rc.3 row). **Also seen:** the MCP takes began recording 14–28 s afterstartRecording with no portal picker visible on screen, and the webcam started up to 28 s before the screen (webcamOffsetMs−28151 compensates). **Not covered:** #1294 in the preview; Full Camera across clip junctions and clip starts (trims only); a held frame coming from a real Linux take (there is none, see above); the.deb; the rest of the checklist. Takes, projects, audio peaks and media-link entries from this pass were deleted, and /.config/openscreenwas restored from a backup taken before the pass. | | 2026-10-17 |v2.1.0-rc.6, tag @ 36b3814, installer Openscreen.Setup.2.1.0-rc.6-x64.exefrom the GitHub release (build run 39010637863). sha512 matchedlatest.yml; installed silently over rc.5. The uninstall entry and the MCP server report 2.1.0-rc.6. | Windows 11 Home 26200 (x64), Ryzen 7 5800X, RTX 4070 Ti, 1920×1080 at 279 Hz; Logitech StreamCam | **Partial — minimal pass: smoke test and the fixes since rc.4. All pass, no critical defect** | Real OS input through computer-use. Export checks also ran from an SDK MCP client over stdio. **Smoke:** - The HUD appeared; a left click on the tray icon brought the minimized editor back. The silent install landed in C:\Program Files\Openscreen, where an rc.6 per-machine install already existed, rather than the per-user folder; the per-user rc.5 was removed. - A 44 s take with webcam, microphone and system audio, with clicks in an Explorer window (wgc-capture.exegranted mid-take). Stop opened the editor with the take, the webcam and one auto zoom. The take's audio peaks at −8 dB. - Playback moved; on a 60 fps capture of the preview, no step back. - Export from the dialog, MP4 1080p60: 1752×1080, 44.35 s, H.264 and AAC, decoded without error, audio peaking at −1.7 dB. - After quitting, reopening from the HUD showed the project intact: take, webcam, auto zoom, changed wallpaper, rewound zoom absent. **Fixes:** - Full Camera trail (#1294, #1304): the rc.5 project re-exported with the same cuts at 1080p60. The frame that trailed on rc.5 (first frame after the trim ending at source 46.34986635) is sharp, and no one-frame outlier is left in the export. Frames before it are identical to the rc.5 export. - Pause on a held frame (#1296): paused at 0:32.1 on a still passage, then changed the wallpaper. The playhead, the cursor and the webcam did not move. - Chat rewind (#1292): "Add a zoom from 2 s to 4 s" added a 1.8× zoom. Rewinding that message showed "Rewound to the start of that message": the zoom was removed and the text returned to the composer. No "Message not found" in the UI or the logs. **Not run:** - smoothness at 60 Hz (rc.5 row, #1301); - the rest of the checklist. | | 2026-10-17 |v2.1.0-rc.6, release/v2.1.0@36b3814, DMG Openscreen-macOS-Apple-Silicon-2.1.0-rc.6.dmgfrom the GitHub pre-release (build run 39010637863): 140 603 025 bytes, sha256 matches the release asset's digest;spctlaccepted, Notarized Developer ID (M4LK7C6S84);stapler validateOK;CFBundleShortVersionString2.1.0-rc.6. Installed over rc.5 | macOS 27.0.1 (26A434), Apple M1 (arm64), 8 GB, one 1920×1080 display at 120 Hz (1×); OBS Virtual Camera (no physical camera), BlackHole 2ch | **Partial — minimal pass toward stable: no critical bug; all 8 checked items pass** | Real OS input through computer-use, launched withOPENSCREEN_MAC_SOURCE_PICKER=legacy, OPENSCREEN_DIAGNOSTIC=1and--remote-debugging-port(read only: thescripts/perf-ab/probe.jspaint probe, the playhead text and the editor's media elements). Measured with the app's bundledffmpeg. Nobody else used the machine during the measures. **Smoke:** 1. The HUD appears. The tray icon (the green one in the menu bar; the aperture icon beside it is OBS's) opens Open / Check for Updates / Update Settings / Send Anonymous Usage Count / About / Save Diagnostics / Permissions… / Quit. *Open* brings back the HUD after *Hide recording bar*. 2. HUD take, 34 s, OBS webcam and microphone, six clicks on a 60 fps test pattern: the editor opens on it with the webcam PiP and two automatic zooms (1.25×, 1.80×). The take is 1920×1080 H.264 at about 56 fps, plus AAC 48 kHz stereo (mean −21.4 dB, real signal). 3. Playback: the preview moves. Probe 28.3–29.0 paints/s, p50 33.4 ms, p95 36.6. The preview's