Desktop window restoration proof Environment: Linux X11, Electron 44.10.1, Openbox, two virtual 1920×1080 displays arranged horizontally, scale 1, light theme. Each side was built in its own checkout with `vp i` and `vp run build:desktop:bundle`. The Electron runtime was installed with `node apps/desktop/node_modules/electron/install.js`. The scenario runner explicitly excludes desktop shell behavior, so the real Electron application was launched through Playwright's Electron launcher under `headless-desktop`. The driver invokes actual BrowserWindow actions; it does not replace Electron APIs, emit synthetic window events, or modify persisted settings. Each run creates fresh isolated app and Chromium state with development seeding disabled, dismisses onboarding, and creates no project or agent turn. Setup a virtual X11 server with the attached `xorg-randr.conf`, enable DUMMY1 at 1920x1080 to the right of DUMMY0 using xrandr, and start Openbox on that display. Set PROOF_DISPLAY to that display when invoking the driver. Both outputs must appear separately in Electron's screen.getAllDisplays(). Run: `headless-desktop node proof-driver.cjs ` The same driver performs three sequences: 1. Move a normal window to display 2, maximize, quit, relaunch. 2. Maximize on display 1, quit and relaunch already maximized, move to display 2, quit, relaunch. 3. Maximize on display 1, quit and relaunch already maximized, move to display 2 and immediately minimize, quit while minimized, relaunch. It records the full two-display desktop at 12 fps, with display labels and action captions outside the captured desktop. Stills are saved only after identical PNG captures 500 ms apart. Observations record actual bounds, normal bounds, maximization, minimization and display bounds. No tests are used as evidence. Limit: this is a Linux reproduction of desktop shell behavior. It does not establish behavior on macOS or physical monitors with differing scales, sizes or arrangements. AI-assisted; verified with the tests and tools listed above.