macOS app beachballs on launch, never creates a window — persists on 26.623 (latest), Apple Silicon

Codex app version: 26.623.101652
Subscription: ChatGPT Pro
Platform: macOS, Apple Silicon (arm64)
What I'm seeing: The app launches (process starts, Codex appears in the menu bar/Dock) but never creates a window. Hovering the menus or clicking the Dock icon beachballs, then the app goes fully unresponsive. No error, no window. Matches the closed issues #21593 and #11303, but still reproduces on current build 26.623.
Key detail — the engine is fine, only the window fails: /Applications/Codex.app/Contents/Resources/codex --version returns codex-cli 0.142.5 instantly, and agents/automations run fine headless via the CLI. The failure is isolated to the Electron desktop window/renderer.
Intermittent-then-stuck: right after a fresh reinstall the window opened once, then re-hung on a later launch and hasn't drawn since.
Tried, did NOT fix: reinstall; full reboot; --disable-gpu + clean --user-data-dir; removed ~/Library/Application Support/Codex, ~/Library/Caches/Codex, ~/Library/Saved Application State/com.openai.codex.savedState, ~/Library/Preferences/com.openai.codex.plist; cleared heavy ~/.codex session state; updated to latest 26.623.
Expected: the app creates and shows its main window. Happy to share logs / a session ID.
Now — to actually get Alderkill going again while they sort it, two honest options, neither of which is the scary rewrite:
Hands-on now: the CLI works today. /Applications/Codex.app/Contents/Resources/codex resume opens your real threads on your Mac. Clunky, but it's your agents, right now.
Running on their own: the only way to get the agents firing without the window is the headless runner — and I want to be clear it is not "rewriting how agents work." It's literally the same Codex engine (the CLI that works) launched on a schedule instead of through the dead window. Same agents, same charters, different ignition. Given the window's now a maybe-never for your machine, that's the realistic path to "the OS is running" — but it's your call whether to flip it on or wait on OpenAI.
Want the bug filed and then we talk about which of those two you'd rather lean on?