[cursor]

What changes

Upstream quattro moved past omacom#18674 (ac6cb9c71), which is what #131 merged. Six commits replayed onto current quattro, through 6baca275b (translations, omacom#18831). No merge commits. Nothing skipped. One conflict, in the power-profile test. One follow-up drops the fork's parent-directory translation watch, which the new watcher replaces. Fork delta vs 6baca275b is 253 paths. The 250 paths upstream did not touch keep the same added and removed lines as quattro vs ac6cb9c71, except shell/Commons/ParentDirReload.qml, which this branch deletes. I18n.qml and Translations.qml now match upstream. Face, Grok Bot, Cursor, linux / linux-headers, and the dropped translation-sync workflow stay.

• Snapdragon power profiles get a CPU floor (Give Snapdragon power profiles a CPU floor for bursty work omacom/omarchy#18745). schedutil raises a core's clock too slowly for a browser's bursts. A root service sets a utilisation floor (uclamp.min) on logged-in users' processes: 75% in balanced, 100% in performance, none otherwise. It applies the floor once it is listening, again when the profile changes, and again when someone logs in. If the D-Bus monitor stops, the service fails so systemd restarts it. Setup enables it only on Snapdragon (omarchy-hw-aarch64-qualcomm), and starts it on an image's first boot, where a newly enabled service would otherwise wait for the next boot. Migration 1794503281 does the same for an existing Snapdragon install. Other machines are untouched. List and set still read and write the profile through busctl. The floor reads the active profile with powerprofilesctl get.
• Wine tray icons answer a double-click and a right-click (Make Wine tray icons respond to double-click and right-click omacom/omarchy#18597). Double-click runs the item's activate action, which is how Battle.net comes back from the tray. Right-click runs omarchy-tray-context-menu, which asks the app over StatusNotifierItem.ContextMenu to draw its own menu at the pointer. The pointer is converted with xrandr, so xorg-xrandr joins the base packages and migration 1794518371 installs it. Two items with the same id, or one that did not answer and might be another Wine app with that id, open nothing. The Battle.net window rule only floats now: the tray menu shares the launcher's title, so the old 1280x800 size was stretching the menu, and the launcher's fullscreen request is ignored. Maximize stays on its own binding.
• Translation files are watched before they are reread (Watch translation files before rereading them, so no change is lost omacom/omarchy#18734). FileView.reload() reads the file and only then replaces its watcher, so a change in that gap was lost, and a watcher that is never reloaded stays on the inode it first saw. WatchedFile splits the two: a reader with no watch, and a watcher that is reloaded before the read. I18n and Translations read catalogs, the language selection, and the locale registry through it. The fork's ParentDirReload was covering the same gap by watching the parent directory and calling reload() on the file. After the replay that call landed on WatchedFile, which has no reload(). The follow-up removes ParentDirReload.
• Wi-Fi rows stay in step with NetworkManager (Keep Wi-Fi rows in step with NetworkManager omacom/omarchy#18702). The rows were snapshots, rebuilt when the list of networks changed, when the panel opened, or when one of its own actions finished. A network connected, dropped, saved, or forgotten by anything else kept its old row until the panel was opened again. The panel now binds a key over every listed network's state and rebuilds when it changes, including networks with no row on screen and changes while the panel is closed. Identical rows stay in place, so a passphrase being typed keeps focus, and the cursor stays on its network through the re-sort. A hidden network's empty name is no longer the cursor's "no row". An action whose network is already in the asked-for state settles before the call, except the enterprise script, which still brings a profile up. A failed connect's reason, and its passphrase prompt, clear once that network connects and settles.
• First boot asks what the installer asks, on the installer's screens (Ask at first boot what the installer asks, on the installer's screens omacom/omarchy#18783). Owner setup and the installer had drifted: the console font, gum's key hints, gum's colours without COLORTERM, and a first boot with no language step that asked for the account in another order and reviewed it in a table that split "English (US, Dvorak)" in two. The shared form now carries the language picker, the layout and timezone that follow from the language, the review table, the gum look, and a COLORTERM default. console-font.sh and console-font.py come over from the ISO, and terminus-font joins the base packages. The size is the kernel's own, 8x16 or 16x32, counted in whole 16x32 cells so asking again cannot flip it. A console left in the kernel's font stays in English, because that font cannot draw the translations. First boot asks language, keyboard, then timezone, username, password, and hostname. "No, change it" goes back to the keyboard. The language picked is written to /etc/omarchy/language once its locale is generated, which is what makes it the owner's desktop language. A locale that cannot be generated leaves the desktop in English, and a retry forgets what an earlier attempt seeded.
• Translations through Update translations omacom/omarchy#18831, including the new first-boot strings.

The power-profile test conflict kept list and set off powerprofilesctl (they go through busctl) and took the floor's get queue, so the new assertion can step balanced, performance, and power-saver.

Verification

• Cherry-pick of the six commits from 3f974d011 through 6baca275b applied onto quattro ab8aa8782. test/shell.d/powerprofiles-set-test.sh conflicted and was resolved as above. git diff --numstat of HEAD vs upstream 6baca275b is 253 paths. On the paths upstream did not touch, the added and removed line counts match quattro vs ac6cb9c71, apart from deleting ParentDirReload.qml. I18n.qml and Translations.qml match upstream. linux and linux-headers are still in the other-packages list. Face, Grok Bot, and Cursor are still in the menu. The translation-sync workflow is still absent. The English catalog still has the fork's Grok Bot and AUR-held strings. docs/i18n.md still says catalogs arrive with these syncs, and it takes the new first-boot sentence.
• ./test/cli passed (133).
• ./test/shell completed with the files this sync covers green: powerprofiles-set-test.sh (9, including the floor), setup-form-test.sh (31), i18n-test.sh (401, one structural skip), tray-test.sh (21). i18n-startup-test.sh, network-panel-runtime-test.sh, and tray-menu-test.sh skipped: no Quickshell and no Wayland compositor, so the live translation reload, the Wi-Fi rows, and the tray menu were not driven. luks-rekey-journal-test.sh skipped its volume runs (no cryptsetup).
• provision-owner-keyboard-test.sh stopped on "the Georgian layout (ge) gets an XKB layout". This machine's /usr/share/systemd/kbd-model-map is 72 lines and has no ge row. Current systemd's map does (ge to ge,us). With that row bound over the host file, the same test passed (26). One skip: no readable VT, so the console-font size choice was not built.
• Thirteen other files failed the same way on quattro ab8aa8782: missing mise, socat, or pacman-conf, Python compression, no systemd as init, systemd-tmpfiles without --inline, pidwait, the China timezone Asia/Chungking, the update lock, and no private XDG runtime. They are not part of this sync.
• This checkout has no Omarchy session, so the tray menu, the Wi-Fi rows, the first-boot screens, and the CPU floor were not checked on a machine. No other sync was open.

Screenshots

No visible change: this checkout has no Omarchy session, so nothing was drawn here. The Wine tray menu, the Wi-Fi rows, and the first-boot language screen are visible on a machine; the tests above cover the behavior that can run here.