Run pnpm verify:launch from apps/v10/shell. The command builds the native
binary, starts it, and fails if the process exits early or a machine-checkable
native behavior does not reconcile.
On macOS, the gate drives the production Rail and tray menus through System Events. It proves that:
- the release binary starts and exposes one main window;
- Dock Left and Dock Right move that window to opposite sides of the work area;
- Keep on Top changes its real checked-menu state;
- closing the window leaves the process alive; and
- the Show Panel tray action restores the same window.
The invoking terminal needs macOS Accessibility permission for System Events.
The script never enables autostart and restores Keep on Top to its starting
value before it exits. --skip-build reruns the probe against the existing
binary.
On Linux, the same command builds the custom-protocol binary and runs it for ten
seconds inside an isolated D-Bus/Xvfb session. It also validates the complete
versioned window-state.json written during setup. This headless path verifies
native process startup, webview creation, tray initialization, and persistence;
it does not claim macOS UI behavior.
What still needs eyes on a Mac#
Automation can read positions, process liveness, menu checkmarks, and window visibility. A release reviewer should still look at these presentation details:
- In both light and dark menu bars, confirm the template tray icon is legible, correctly sized, and not clipped.
- Confirm the narrow panel is visually flush with each screen edge and does not overlap the menu bar or Dock on every attached display.
- Put another application over The Rail and visually confirm Keep on Top changes the stacking behavior, including after switching Spaces.
- Open the tray menu and confirm its labels, separators, disabled glance rows, and checkmarks are readable at the active display scale.
- Close and restore the panel once, then confirm the rendered Rail session is unchanged and still interactive.
Record the macOS version, display arrangement/scales, light/dark appearance, command output, and any visual defects in the release evidence for the tested build. These checks are intentionally documented as human observations rather than represented as automated success.