Skip to main content

parity ledger

Baseline: GNOME 51 as shipped in the TunaOS Marlin GNOME image (ghcr.io/tuna-os/marlin:gnome). Record the image digest used for each baseline capture in the evidence column. Rule: a row is pass only when a test or recorded review compares Tuna Desktop against baseline evidence. untested is not a pass. Undocumented gaps are not parity (see roadmap change control).

Status values: pass, partial, missing, deviation (deliberate, with rationale and owner), untested.

IDGNOME 51 behaviorTuna Desktop statusTestsEvidence / notesDeviation and owner
P-OV-01Super opens the overview with live window previews for the current workspacepassjourney:J-OV-OPEN, proof:G-OV-OPEN, proof:G-HOT-CORNER, proof:G-OV-PICK, cargo:spread_previews_never_overlap_and_stay_in_the_picker, cargo:activities_follow_live_primary_geometry_and_output_disconnect, cargo:hot_corners_use_exposed_secondary_corners_and_live_gnome_policy, cargo:hot_corner_regions_use_logical_pixels_at_fractional_output_scales, proof:G-OVERVIEW-EASING, cargo:opening_eases_out_sine_and_closing_eases_out_quad, proof:G-HOT-CORNER-FULLSCREEN, proof:V-HOT-CORNER-PRESSURE, proof:V-HOT-CORNER-CONSTRAINTS, cargo:native_pressure_barrier_uses_live_logical_scale_geometry, cargo:hot_corner_follows_gnomes_setting, cargo:hot_corner_opens_only_when_closedcompositor-drawn workspace card with scaled live previews; click picks (crates/compositor/src/overview.rs); recorded against GNOME 51 in docs/gnome-parity.md: overview with windows 1.3% off, a preview hovered 1.7%; GNOME's 250 ms ease-out transition: the card grows out of the desktop and windows glide into their previews; genuine fullscreen blocks the hot corner (G-HOT-CORNER-FULLSCREEN); on the Marlin image built in CI run 38878636595, real relative-mouse pressure opens and rearms the corner and pointer locks/confinements suppress it (V-HOT-CORNER-PRESSURE, V-HOT-CORNER-CONSTRAINTS; also passed on the published image, nightly 38844351544)—
P-OV-02Overview shows the workspace strip with thumbnails; drag or click switchespassproof:G-WS-THUMBS, proof:G-OV-DND, proof:G-WS-INSERT, cargo:thumbnails_strip_matches_gnome_51, cargo:two_workspaces_have_no_strip, cargo:thumbnail_decor_rounds_and_rings, cargo:a_dragged_preview_shrinks_fades_and_follows_the_pointer, cargo:drops_land_on_thumbnails_and_neighbor_cardsthree or more workspaces show GNOME 51's strip (43x26 thumbnails, an indicator in GNOME's accent-color, card shrunk to 849x509) and a click switches; a window preview drags (shrunk to 256px, faded) onto a thumbnail or neighbor card to move the window (overview.rs); parity frame 16-overview-workspaces is 1.4% off GNOMEnone: insertion gaps show the 18px marker, shift later thumbnails by 24px, and insert/move with dynamic workspace pruning; frame 16b thumbnail strip is 3.6% off, marker-only crop 0.0% off
P-OV-03Overview shows the dash with favorites and running appspassproof:G-OV-SHELL-UIGTK shell dash with favorites and GNOME's running dots (crates/shell-gtk/src/overview.rs); recorded against GNOME 51 in docs/gnome-parity.md: overview dash 0.0% off—
P-OV-04Typing in the overview searches apps and launches the top hit with Enterpass (nested)journey:J-OV-SEARCH, journey:J-OV-LAUNCH, proof:G-OV-SEARCH, proof:G-OV-LAUNCH, proof:G-SEARCH-PROVIDER, proof:G-SEARCH-ACTIVATE, proof:G-SEARCH-VIEW, proof:G-APP-INSTALLED, cargo:selection_follows_gnome_settingsapp search plus GNOME Shell D-Bus search providers with their settings keys (crates/shell-gtk/src/providers.rs); app results as GNOME's 145px tiles with the top hit highlighted under the focused entry, recorded against GNOME 51 in docs/gnome-parity.md: overview search 0.3% off—
P-OV-05App grid with pages and folderspassproof:G-OV-APPGRID, proof:G-APP-FOLDERS, cargo:folders_take_apps_by_id_and_category_minus_exclusions, proof:G-APP-GRID-PAGES, cargo:grid_modes_follow_the_area_like_gnome, proof:G-FOLDER-DND, cargo:folder_lists_change_like_gnome, proof:G-GRID-REORDER, cargo:grid_reorders_and_saves_like_gnome, proof:G-GRID-DRAG-PAGE, proof:G-GRID-HOVER, proof:G-GRID-PAGE-PEEK, proof:G-GRID-SHORT-PAGES, proof:G-GRID-FOLDER-WIDTH, cargo:moving_between_pages_preserves_the_short_source_page_on_rebuild, cargo:page_moves_spill_only_into_following_pages_and_deduplicate_saved_ids, proof:G-FOLDER-ZOOM, proof:G-GRID-PAGE-ANIM, proof:G-LAUNCH-ZOOM, proof:G-GRID-FADE, cargo:grid_animation::tests::curves_match_clutter_easing_modes, cargo:grid_animation::tests::gnome_51_durations, cargo:grid_animation::tests::reduced_motion_keeps_fades_and_off_keeps_nothingShow Apps opens the grid; folders come from GNOME's org.gnome.desktop.app-folders (ids, categories, exclusions, translated names); GNOME's pages (grid mode by shape, 8x3 on a wide screen), folder labels ellipsize within the tile instead of widening homogeneous cells beyond the viewport, page indicators and side arrows, wheel, swipe and PageUp/PageDown (overview.rs); GNOME's drag edits: an app dropped on another makes a folder named from their shared category, one dropped on a folder joins it, one dragged out of the folder dialog leaves it (the emptied folder is deleted) (folders.rs), with the client's drag icon drawn at the pointer; GNOME's order from app-picker-layout (saved items by page and position, the rest by name), a 200ms hover within 20px of a tile's side reflows live, drop saves and cancellation restores the layout; saved per-page boundaries preserve short pages across catalog rebuilds; while dragging, adjacent page icons slide into the 10% hints over 150ms and hints replace the arrows, an edge bump turns the page at once and a hovered hint after a second, repeating each second; GNOME 51's grid animations (grid_animation.rs, grid_pager.rs): the folder dialog zooms out of its icon and back over 200 ms as the icon fades, pages switch over 300 ms ease-out-cubic with the page dots following, swipes settle as GNOME's SwipeTracker, the arrows fade over 200 ms, launches zoom the icon and label to 3x over 250 ms, dragged icons scale and fade, new icons appear in place (GNOME 51 never calls AppIcon.scaleIn()), reflowed icons slide staggered 10 ms, and the grid fades in over 400 ms after 100 ms and out over 400 ms; all instant with animations off, fades only under Reduced Motion—
P-OV-06Escape returns to the previously focused windowpasscargo:overview_parks_keyboard_focus_and_restores_on_dismiss, journey:J-OV-ESCNative virtual Escape and Tuna Desktop Escape restore Beta at its original rectangle; paired frame 20-overview-dismiss-focus is 0.4% off, with both state records committed; see docs/gnome-parity.md—
P-PN-01Top panel: Activities, centered clock, system menu rightpassproof:G-PANEL, proof:A11Y-PANEL, proof:G-PANEL-FOCUS, cargo:escape_unwinds_menus_then_releases, cargo:arrows_walk_only_a_menu_free_barGTK shell panel, the default shell (ADR 0006, SHELL_BINARIES in crates/compositor/src/runtime.rs); recorded against GNOME 51 in docs/gnome-parity.md: top bar 0.7% off (the clock's minutes); Ctrl+Alt+Tab walks the bar, Enter opens its menus and Escape unwinds (G-PANEL-FOCUS)—
P-PN-02Clock opens calendar plus notification listpassproof:G-CAL-OPEN, proof:G-CAL-EVENTS, proof:G-CAL-LAUNCH, proof:G-NOTIFY-LIST, proof:G-NOTIFY-CLEAR, cargo:titles_follow_gnome_51, cargo:event_times_follow_gnome_51, cargo:overlap_sort_and_recurrence_follow_gnome, cargo:added_updated_and_removed_payloads_follow_the_calendar_server, cargo:malformed_payloads_change_nothing_and_never_panic, cargo:all_day_and_multi_day_events_land_on_their_days_across_dst, cargo:server_vanishing_clears_events_and_calendars, proof:V-CALENDARGTK shell calendar popover lists notifications with Clear (crates/shell-gtk/src/notify.rs); events from org.gnome.Shell.CalendarServer (crates/shell-gtk/src/events.rs) mark days with GNOME's dot and list in the events card, titled and timed as dateMenu.js does, the card hidden without calendars; the installed Marlin image's CalendarServer answers on the user bus on 5/5 boots (V-CALENDAR, CI run 38878636595 and published-image nightly 38844351544)—
P-PN-03Quick settings grid: Wi-Fi, Bluetooth, power mode, night light, dark style, volume, brightnesspassproof:G-QS-OPEN, proof:A11Y-QS, proof:G-QS-DIM-OPAQUE, proof:G-QS-EXTERNAL-SYNC, proof:G-QS-SERVICES, proof:G-QS-WIFI-MENU, proof:G-NM-AGENT, proof:G-NM-ENTERPRISE, proof:G-NM-VPN, proof:G-NM-WIRED, proof:G-QS-WIFI, proof:G-QS-BT-MENU, proof:G-QS-BLUETOOTH, proof:G-QS-POWER, proof:G-QS-DARK, proof:G-QS-BRIGHTNESS, proof:G-QS-SOUND-MENU, proof:G-QS-VOLUME, proof:G-QS-ABSENT, cargo:power_mode_follows_gnome_51, cargo:networks_group_and_sort_like_gnome, cargo:security_follows_nm_flags, cargo:signal_icons_follow_gnome, cargo:at_most_eight_networks_show, cargo:networks_dedupe_by_ssid_keeping_the_strongest_access_point, cargo:non_utf8_ssids_are_shown_and_grouped_by_their_bytes, cargo:signal_icons_switch_at_gnomes_thresholds, cargo:picking_a_network_activates_saved_creates_new_or_opens_settings, cargo:nm_device_states_decide_visibility_like_gnome, cargo:several_adapters_are_told_apart_and_the_toggle_acts_on_all, cargo:profile_selection_prefers_the_most_recent_then_the_first_by_name, cargo:devices_filter_and_sort_like_gnome, cargo:subtitle_and_icon_follow_gnome, cargo:secrets_follow_key_management_like_gnome, cargo:validators_follow_gnome, cargo:ports_share_a_sink_but_switch_distinct_routes_and_metadata_is_incremental, proof:G-QS-WIRED, cargo:wired_toggle_and_menu_follow_gnome, proof:G-QS-MIC, proof:G-QS-MIC-EVENT, proof:G-QS-SOUND-PORT, cargo:split_events_remove_capture_and_ignore_monitor_streams, proof:G-QS-NIGHT, proof:G-BATTERY, cargo:upower_battery_states_choose_icons_and_remaining_timeGTK shell grid drives NetworkManager, BlueZ, power-profiles-daemon, logind brightness, PipeWire volume, and the GSettings tiles; each tile hides without its daemon (crates/shell-gtk/src/services.rs); proven against stub daemons; GNOME's Wi-Fi menu lists, sorts and activates NetworkManager's networks, scanning while open (crates/shell-gtk/src/wifi.rs); GNOME's Bluetooth menu lists paired devices and connects them (crates/shell-gtk/src/bt_menu.rs); the shell is NetworkManager's secret agent, asking for Wi-Fi, 802.1X and external-UI VPN secrets, storing agent-owned user secrets through libsecret, and presenting each wired adapter in its own submenu (docs/network-agent.md); the volume slider's sound output menu (services.rs) enumerates PipeWire device routes and follows pw-dump --monitor recording/volume events before the next frame (stubbed route and event gates); the Night Light tile toggles night-light-enabled (G-QS-NIGHT; the tint itself is P-DS-01) and the UPower battery indicator follows hotplug, charge state and the percentage setting (G-BATTERY)—
P-NT-01Notification banners show top-center one at a time with actions and dismissalpasscargo:panel::tests::live::banners_surface_appears_and_destroys_with_queue, cargo:intake::tests::live_bus::stub_notify_shows_a_banner_and_close_notification_dismisses_it, proof:G-NOTIFY-BANNER, proof:G-NOTIFY-SEND, proof:G-NOTIFY-BLOCKED, proof:G-NOTIFY-EXPIRE, cargo:notify::timer_tests::banners_time_out_like_gnome, cargo:notify::timer_tests::one_banner_at_a_time_like_the_message_tray, cargo:notifications::tests::full_queue_leaves_new_arrivals_in_history_like_gnome, cargo:notifications::tests::notify_queues_banner_and_history_in_order, cargo:notifications::tests::high_urgency_is_held_by_dnd_and_persists, cargo:notify::timer_tests::an_updated_banner_on_screen_shows_again_in_place, cargo:transient::tests::an_updated_banner_pops_again_from_where_it_standsGTK shell banners top-center with actions, close, and GNOME's expiry (4 s after the banner has dropped in, or 2 s after an away user returns, through org.gnome.Mutter.IdleMonitor); GNOME's message tray queue: one banner on screen, critical ones next, at most three waiting with the one on screen, later arrivals only in the list, only an expired banner slides out, and a notification updated in place while its banner is up shows again on that banner (_updateShowingNotification) instead of queuing; GNOME's urgency levels: low never banners (list only), the shell's own Urgency.HIGH (wellbeing) queues ahead of normal but, unlike critical, is held by Do Not Disturb and expires; recorded against GNOME 51 in docs/gnome-parity.md: notification banner 0.4% off—
P-NT-02Do Not Disturb toggle and notification historypasscargo:notifications::tests::dnd_gates_banners_but_not_critical_or_history, proof:G-NOTIFY-DND, proof:G-NOTIFY-LIST, cargo:notify::group_tests::notifications_group_per_app_newest_first, cargo:notify::group_tests::collapsed_stack_peeks_like_gnome, cargo:group_animation::tests::notification_groups_expand_and_collapse_with_200_ms_ease_out_quad, proof:G-NOTIFY-GROUPDND in quick settings and the calendar holds banners back; history listed, grouped per app as GNOME 51 does (collapsed stack peeking 10 px then 7 px, 6 px inset per card, at most three; pressing expands under a header with Collapse; closing a collapsed group's top card closes the group); recorded against GNOME 51 in docs/gnome-parity.md: the date menu listing a notification 0.5% off; the quick settings holding the DND toggle 1.2% off—
P-WM-01Alt-Tab switcher with app icons and window previewspartialcargo:alt_tab_steps_commits_and_cancels_without_leaking_keys, cargo:cycle_chord_inside_an_open_switcher_cancels_only_on_escape, journey:J-SW-STEP, proof:G-SWITCHER, proof:G-SWITCHER-GROUP, proof:G-SWITCHER-FADE, cargo:strip_shrinks_to_fit_under_the_app_list_at_1280x800, cargo:switch_group_steps_through_the_apps_windows_like_gnome, cargo:switcher_keys_close_and_quit_like_gnome, cargo:window_backed_apps_stand_alone_like_gnome, cargo:toplevel_icon_assignment_is_double_buffered_and_survives_resource_destroy, proof:G-WINDOW-ICONSGTK shell draws GNOME's centered icon strip, one icon per app with GNOME's arrow under apps with several windows, and the selected title; Alt release focuses it; recorded against GNOME 51 in docs/gnome-parity.md: Alt+Tab switcher 0.5% off; an app's windows show as GNOME's thumbnails (compositor-drawn) after 500 ms or on Down/Alt+Above_Tab, shrunk to fit under the app row and scrolled when wider than the monitor; thumbnails fade in and the popup fades out over 100 ms, the switcher keeps the keyboard (arrows, Q, W, F4; a cycle chord such as GNOME's Alt+Escape does not cycle, and Escape or Tab cancel, switcherPopup.js:211-219)—
P-WM-02Super+Up maximizes, Super+Left/Right tiles halvespasscargo:super_arrows_drive_layouts_and_consumeNative and Tuna Desktop key captures produce identical work-area rectangles; frame 21-maximized is 0.3% off, frame 22-tiled-left 0.5%, frame 22b-tiled-right 0.3%; state records and comparison in docs/gnome-parity.md—
P-WM-03Dynamic workspaces, Super+PgUp/PgDn switch, Shift moves windowpassproof:G-WS-POPUP, cargo:dynamic_workspaces_end_with_one_empty, cargo:super_page_keys_switch_and_shift_moves_focused, journey:J-WS-NEXT, proof:G-WORKSPACE-SLIDE, cargo:a_key_switch_slides_both_workspaces_for_250_ms, cargo:a_second_press_retargets_from_the_drawn_offset, cargo:motion_off_switches_instantly_and_fade_crossfades_in_place, cargo:the_pill_fades_over_100_ms_ease_out_quad, cargo:the_pill_moves_between_dots_over_250_ms, cargo:added_dots_scale_in_and_removed_dots_scale_out_over_500_msGNOME's switch animation: both workspaces slide 250 ms ease-out-cubic as a render offset (no client reconfigure), a press mid-flight retargets from the drawn offset, animations off switch instantly (crates/compositor/src/workspace_slide.rs); the switcher popup fades 100 ms and the Activities dots follow the switch and scale in/out over 500 ms (crates/shell-gtk/src/ws_dots.rs); Native virtual Super+PageDown/PageUp and Tuna Desktop keys switch to workspace 1 and back; frame 23-workspace-switched 0.0% off, null focus in the empty workspace on both sides; recorded states in docs/gnome-parity.md—
P-WM-04Interactive drag-to-move and resize with pointerpasscargo:tile_preview_eases_from_the_window_in_200_ms, cargo:header_drag_moves_the_window, cargo:bottom_right_resize_grows_and_configures, cargo:super_drag_moves_without_a_client_request, cargo:dragging_to_an_edge_shows_gnomes_tile_preview, proof:G-APP-DRAG, proof:G-TILE-PREVIEWheader drag, edge resize, Super+drag, top-edge maximize, side-edge half tiling on drop, and GNOME's tile preview over the snap area while dragging (windows.rs, runtime.rs); parity frame 17-tile-preview is 0.5% off GNOME—
P-WM-06App menus and popovers open as xdg popups and close on outside clickpass (nested)cargo:popup_is_configured_where_the_positioner_asked, cargo:press_on_another_client_dismisses_the_grab_and_is_consumed, proof:G-APP-POPOVER, proof:G-CAL-OPENxdg popups with positioner constraints and the grab rule (crates/compositor/src/popup.rs); recorded against GNOME 51 in docs/gnome-parity.md: an app's popover 0.5% off, the window placed by Mutter's cascade (first free slot)—
P-WM-05Window close, minimize, maximize via server-side or client decorationspass (nested)proof:G-WINDOW-MENU, cargo:window_menu::tests::rows_follow_gnome_51, cargo:window_menu_actions_follow_gnomeclient decorations close, maximize and minimize (xdg set_minimized, Super+H); GNOME's window menu on a header-bar right click (Hide, Maximize/Restore, Move, Resize, Always on Top, Always on Visible Workspace, Move to Workspace, Close); recorded against GNOME 51 in docs/gnome-parity.md: three decorated windows 0.4% off, the window menu 0.2%—
P-WM-07Scrollable tiling (niri mode, Super+Shift+T): a strip of columns holding one or more windows, Super+R widths, Super+arrows along the strip and inside a column, Super+[ / Super+] / Super+, / Super+. consume and expel, pointer move and resize of columns and window heightspass (nested)proof:G-SCROLL, proof:G-SCROLL-COLUMNS, proof:G-SCROLL-WHEEL, proof:G-SCROLL-OVERVIEW, proof:G-SCROLL-WORKSPACES, proof:G-SCROLL-SWITCHER, proof:G-SCROLL-LAYOUTS, cargo:strip_columns_resize_on_primary_output_failover_without_losing_focus, cargo:new_columns_open_right_of_focus_and_the_view_follows_like_niri, cargo:strip_geometry_gaps_overflow_and_advertise, cargo:scroll_preset_shrink_reclamps_offset, cargo:strip_view_springs_to_target_while_geometry_jumps, cargo:settles_in_niri_time, cargo:strip_pointer_move_presents_actor_keeps_slots_and_commits_only_own_button, cargo:strip_pointer_move_escape_restores_order_view_and_floating_restore, cargo:strip_super_drag_second_button_preserves_initiator_and_balanced_client_events, cargo:strip_resize_configures_continuous_width_neighbors_and_preserves_floating_stash, cargo:strip_resize_cancel_restores_width_neighbors_view_and_ignores_unmapped_actor, cargo:strip_pointer_resize_clamps_width_and_escape_restores_preset, cargo:strip_pointer_width_and_order_survive_layout_round_trip, cargo:strip_consume_and_expel_follow_niri_columns_and_keep_widths, cargo:gnome_mode_forwards_column_chords_untouched, cargo:strip_columns_honour_client_min_and_max_sizes, cargo:strip_window_height_resize_inside_column_commits_and_cancels, cargo:strip_modifier_resize_button_resizes_from_window_thirds, cargo:strip_drop_joins_a_column_near_a_window_edge_and_a_gap_near_a_column_edge, cargo:strip_columns_and_heights_survive_mode_round_trip_close_and_workspace_move, proof:V-STRIP-INSTALLED, proof:V-STRIP-EDGE-STATIONARYa Tuna Desktop extra with niri 26.04 as its reference (docs/niri-parity.md): 16px gaps on every side, columns tiled on every edge (no CSD shadow), new columns right of the focused one, the view scrolling the least that shows the focused column whole (niri's default), niri's 4px #7fc8ff focus ring; the view glides to its target on niri's default view-movement spring (critically damped, stiffness 800, epsilon 0.0001; crates/compositor/src/spring.rs), settling in about 0.4 s, while clients are configured once at the target and only drawn positions animate, windows moving and changing height inside a column on the same spring; columns hold several windows: Super+[ / Super+] consume a lone window into the bottom of the neighboring column or expel a shared one into a new column on that side keeping the column's width, Super+, / Super+. take the next column's top window and send the bottom one out (niri's default binds; GNOME 51 binds none of these chords), Super+Up/Down move focus inside a shared column and Shift moves the window; windows share the column height by weight, minimums are met and one resized window keeps a fixed height (niri update_tile_sizes), and column widths clamp to the clients' min/max sizes, re-sizing when a client changes them; pointer move (Super+drag, or a client title drag through xdg move) carries the window as a drawn actor and drops it at the boundary nearest the pointer (niri insert_position): a column edge makes a new column, a window's top or bottom edge inside the column under the pointer joins that column, shown by an 8px bar; Escape, a VT switch, a shortcut-inhibitor lock and output, workspace or mode changes cancel back to the exact order and view; an xdg resize or the GNOME modifier resize on the pointer's outer third of a window (niri resize_edges_under; P-WM-09's mouse-button-modifier and resize-with-right-button choose the modifier and button, not niri's fixed Mod+RMB) changes the column width from one side, clamped to 120 px through 4× the work-area width, and a window's height from its top or bottom inside a shared column (not the topmost top edge), with Escape restoring every height; Super+R then steps to the next larger preset; columns, widths and heights survive a Super+Shift+T round trip while floating rectangles restore exactly, and a window changing workspace leaves its column; the Marlin image built from source in CI run 38878636595 drives the strip with real QMP input across 10 installed frames, including a grip resize, a title-drag reorder and a VT cancel, and a stationary edge (V-STRIP-INSTALLED, V-STRIP-EDGE-STATIONARY; scripts/lib/vm_strip_acceptance.py), and #812's exact-head run 40811403179 passed 11 frames with a compositor-owned Super+drag; V-STRIP-INSTALLED now aims its drags at nearest gaps and adds a real Super+] consume, a GTK grip height resize inside the shared column and a Super+] expel, which no cited run has passed yet; the published-image nightly does not run the strip, so the row stays nestedpaired niri 26.04 frames and measured differences are in docs/niri-parity.md (the paired frames cover single-window columns only); wheel/touchpad view scrolling remains direct; maximize/fullscreen/edge tiling keep strip membership rather than entering niri fullscreen; strip geometry uses the primary output rather than independent strips per output; measured post-animation transient timing differences remain explicit; a lone window stays full height (niri lets a vertical resize shorten it); the dragged window keeps its slot until the drop instead of leaving the layout; drop bars are 8px rather than niri's 300px hint areas; windows with attached dialogs only drop as new columns; in-column moves use Super+Shift+Up/Down rather than niri's Mod+Ctrl; no tabbed columns, window-height presets or height reset
P-LK-01Idle blanks then locks; password unlocks; nothing leaks while lockedpass (nested)cargo:lock_command_engages_and_unlock_restores, cargo:correct_password_dismisses_lock_intact, cargo:pam_unlock_accepts_the_right_password_and_nothing_else, cargo:modifier_releases_during_lock_do_not_turn_unlocked_clicks_into_moves, cargo:swallowed_accelerator_release_clears_physical_pressed_state, cargo:same_focused_window_click_restores_keyboard_after_lock_ownership, proof:G-SUSPEND-LOCK, proof:G-IDLE-FADE, proof:G-LOCK-BACKGROUND, proof:G-UNLOCK, proof:G-LOCK-FOCUS, cargo:consumed_release_is_absent_from_the_next_keyboard_enter, cargo:consumed_modifier_release_updates_the_next_forwarded_key_without_focus_change, cargo:idle_fade_then_activity_cancels_without_unlocking_a_lock, cargo:idle_lock_follows_gnome_keys, proof:G-LOCK-KEY, proof:G-LOCK-FIRST-PRINTABLE, proof:G-LOCK-WITH-MENU-OPEN, cargo:locking_dismisses_the_grab_and_unlock_never_returns_to_the_menu, proof:G-SCREENSAVER, proof:G-DBUS-LOCK, proof:G-MEDIA-LOCK, proof:G-POINTER-TRACE-PRIVATE, proof:G-IME-TRACE-PRIVATE, proof:V-LOCK, proof:V-AUTH-FAIL-CLOSED, proof:V-SLEEP-AUTH-FAIL-CLOSED, proof:V-SLEEP-FRESH-AUTH, proof:V-SUSPEND, proof:V-NORMAL-INPUT-FAIL-CLOSED, proof:V-CAPTURE-INPUT-FAIL-CLOSED, proof:V-VT-FAIL-CLOSED, proof:V-SHELL-FAIL-CLOSED, cargo:stationary_pointer_press_enters_the_lock_surface_before_delivery, cargo:lock_pointer_enter_never_steals_a_held_grab, cargo:logind_lock_and_unlock_reach_the_compositor_and_the_hint_follows_the_lock, cargo:logind_unlock_ends_the_lock_and_strands_the_attempt_in_flight, cargo:actual_runtime_grants_only_the_shell_lock_and_input_reaches_only_its_curtain, cargo:lock_takes_the_keyboard_at_once_and_unlock_hands_it_back, cargo:destroying_a_granted_lock_without_unlock_is_an_error_and_never_unlocks, cargo:a_finished_lock_does_not_reserve_its_output_for_the_next_lock, proof:V-LOCKED-HINTtests/lock.rs, tests/unlock.rs; unlock through PAM (tuna-lock service); locks before sleep; the idle fade and lock follow GNOME's screenShield.js (crates/compositor/src/lock.rs, G-IDLE-FADE); the lock curtain, unlock prompt and blurred background are recorded against GNOME 51 in docs/gnome-parity.md. A lock surface mapped under a pointer that had not moved since used to get no pointer focus, so the first click, drag or scroll on the lock screen was lost (or, with focus left on a window, sent behind the lock); the pointer now enters the lock surface before a press or scroll unless a held button keeps its grab (windows.rs lock_input, #502). Keyboard focus leaves the application as the session locks (the lock surface, or nobody without one), not at the first key, and returns to the focused window on unlock (#729). Locking with a shell menu open (the date menu, quick settings) once left the menu's popover grab alive behind the lock, and GTK sent every key and click of the lock screen to the hidden menu until Escape closed it (beta.3 qualification); as GNOME's lock does, the lock now sends popup_done to every grabbed popup, and the shell's lock windows sit in their own GTK window group, so no other window's grab can hold their input (State::dismiss_popup_grab_for_lock, lock.rs, G-LOCK-WITH-MENU-OPEN). The actual nested runtime refuses a non-shell client's ext-session-lock and engages the shell's, and real host input under Xvfb reaches only the lock surface, or nothing when the lock client lost it (tests/session_lock_runtime.rs, check job). On the Marlin image (marlin-vm, CI run 38878636595, and the published image, nightly 38844351544), Super+L locks and PAM unlocks (V-LOCK); real pam_deny, a stale pre-sleep PAM result, VT switching and a killed lock UI all stay locked, and ordinary keys, capture and virtual-keyboard clients are refused while locked (scripts/tuna-vm-lane, docs/vm-lifecycle.md). As GNOME 51's screenShield.js does, logind's LockedHint is yes while locked and no after unlock, and the session's logind Lock/Unlock signals (loginctl lock-session/unlock-session) lock and unlock, unless the lockdown disables locking (crates/compositor/src/logind_session.rs, V-LOCKED-HINT, #846)idle blank→lock and locking an unlocked desktop on PrepareForSleep are proven only in the nested session (the VM lane locks before its S3 and has no idle gate), so the row stays nested; 24-hour/hardware idle and suspend coverage remains in the release matrix
P-SY-01Session starts from the display manager on hardwarepartialproof:D-OUTPUT, proof:D-IDLE, proof:D-FRAMES, cargo:recycled_buffer_repaints_damage_from_intervening_frames, cargo:timing_only_frame_and_removed_elements_keep_correct_damage, cargo:global_drawing_pass_order_and_opacity_changes_reset_buffer_history, proof:D-SHELL, proof:V-DRM, proof:V-SHELL, proof:V-PANEL, cargo:pam_unlock_accepts_the_right_password_and_nothing_else, proof:D-ENDURANCE, proof:D-NOPANIC, proof:V-DISPLAY-MANAGER, proof:V-GREETER-LIST, proof:V-GDM-FALLBACK, proof:V-GDM-LOGIN, proof:V-NOPANIC, proof:V-LIFECYCLE-OWNER, proof:V-VT, proof:V-SUSPEND, proof:V-SUSPEND-GPU-PM, proof:V-SUSPEND-GPU-UPLOAD, proof:V-SHELL-FAIL-CLOSED, proof:V-LOGOUT, proof:V-LIFECYCLEDRM/KMS on vkms in CI (scripts/tuna-drm-smoke), with retained idle before/after frame-counter snapshots (at most one new submission in three seconds), bounded initial submissions over the longer idle window, 240 real changing-client presentations at both default and 200% scale carrying ordered vsync/hardware-completion feedback, and a 16-second endurance run without a panic; the marlin-vm and marlin-vm-orca51 jobs build the Marlin image from source, give it tunaOS's tuna-flavor GDM login path (greeter theme overlay, FallbackSession=tuna, every other session hidden; packaging/marlin/vm-lane/tunaos/), install it with bootc and boot it five times in QEMU/KVM, where the lane types the fixture password into stock GDM's user list and GDM starts the Tuna Desktop DRM/KMS session as a non-root gdm-password logind session with the GTK shell and black panel (V-GREETER-LIST, V-GDM-FALLBACK, V-GDM-LOGIN on all five boots of CI run 40977772289, PR #858; scripts/tuna-vm-lane, docs/vm-lifecycle.md); in that run the session also survives a VT switch, real S3 suspend/resume and a supervised shell replacement, and the product's Log Out returns to the GDM greeter; the published-image nightly runs the same GDM login code against marlin:tuna (docs/marlin-tuna-nightly.md) and passed with tuna-desktop 1.0.0beta.1-1 after tunaOS#4055 (runs 40967204592 on marlin@sha256:6b9b3413, 40977772364 on marlin@sha256:ae6363c1); its sixth pristine boot now runs a bounded 3000 s soak with the native frame workload and publishes a memory and frame-pacing report (nightly-only V-SOAK, docs/soak.md; run 41080213387 on marlin@sha256:ae6363c1: 51 samples, session PSS 548.6 MiB after warmup and 549.4 MiB at the end, fds and processes flat, 224951 frames at a 13.33 ms p99 interval), which is not the 24-hour soak (#203)VM only (virtio-gpu). No physical hardware yet (#68)
P-SY-02Multi-monitor with per-output panel and hotplugpass (nested)cargo:hotplug_remove_and_readd_returns_windows_to_their_output, cargo:moved_output_carries_its_windows, cargo:rearranged_and_scaled_outputs_carry_their_windows_past_the_next_tick, cargo:unplug_racing_a_wl_output_bind_keeps_the_client, cargo:hotplug_prefers_stored_then_complete_existing_then_linear, proof:V-INTROSPECT-SCREENtests/outputs.rs, tests/migration.rs; real DRM connector hotplug runs nightly in the nested stress workflow's hotplug job (vkms configfs, H-* gates, docs/testing.md), not yet a ledger proof input; virtio-gpu hotplug, rotation and unplug in the Marlin VM (V-INTROSPECT-SCREEN)real panels/EDIDs and colord registration on hotplug untested
P-SY-03Fractional scaling and mixed DPIpartialproof:S-LOGICAL, proof:S-PANEL, proof:S-WINDOW, proof:S-INPUT, proof:S-X11-WINDOW, proof:S-X11-INPUT, proof:D-SCALE, cargo:picks_the_arrangement_for_the_lit_connectors, cargo:windows_take_the_scale_of_the_output_they_overlap_most, cargo:new_windows_start_at_the_scale_of_the_output_under_the_pointerper-output scale and layout from GNOME's monitors.xml on hardware; TUNA_SCALE for nested sessions; the nested proof runs 100/125/150/200 percent with native and genuine X11 painted clients and physical-pointer input; logical layout, fractional-scale clients, scaled input and cursor; on mixed DPI each window and layer surface gets the preferred buffer and fractional scale of the output it overlaps most, re-sent every framemixed DPI not yet run on hardware (#68)
P-SY-04XWayland apps run by defaultpass (nested)proof:G-X11-APP, proof:G-X11-DEMAND, proof:G-X11-RESTART, proof:G-X11-CLEANUP, proof:G-X11-INPUT-FOCUS, cargo:on_by_default_when_xwayland_is_installed, cargo:running_disconnect_keeps_backoff_and_total_restart_budgetdefault builds reserve compositor-owned sockets and advertise DISPLAY; first X11 connection consumes that exact owner to start the pinned Smithay 0.7.0 server; bounded restart preserves native windows; the X server's input focus follows the focused X11 window (Mutter's meta_x11_display_set_input_focus), not the X pointergreen main CI run 38878636595 (e67f6a8b) passed first-client mapping, on-demand start, actual server restart with native-ID preservation and socket/lock cleanup in the nested session; no installed-image X11 gate exists, and physical GPU/VT qualification remains separate
P-SY-06Clients find the protocols Mutter offers (dmabuf, activation, viewporter, fractional scale, cursor shape, idle inhibit, presentation time, fifo, shortcuts inhibit, xdg-foreign, dialogs)partialcargo:advertised_globals_match_the_golden_list, cargo:dmabuf_is_advertised_only_with_renderer_formats, cargo:focused_client_can_hand_focus_with_a_token, cargo:background_clients_and_serial_less_tokens_cannot_steal_focus, cargo:presentation_feedback_reports_the_frame_that_drew_the_window, cargo:a_hidden_window_keeps_its_feedback_until_drawn, cargo:fifo_holds_a_waiting_commit_until_the_next_refresh, cargo:fifo_keeps_advancing_behind_the_lock_screen, cargo:a_timed_update_waits_for_the_refresh_whose_presentation_reaches_it, cargo:timed_updates_apply_in_commit_order_not_timestamp_order, cargo:a_timestamp_with_a_whole_second_of_nanoseconds_is_invalid, cargo:nested_presentation_is_timed_on_the_monotonic_clock_without_hardware_claims, cargo:an_inhibiting_window_gets_the_shortcuts_until_super_escape, cargo:denied_shortcut_inhibition_stays_denied_after_refocus, cargo:shortcut_consent_cannot_survive_a_lock_or_focus_change, cargo:modal_dialogs_keep_focus_from_their_parent, cargo:exported_windows_parent_another_clients_dialog, cargo:provider_exit_returns_focus_to_the_pickers_foreign_parent, cargo:provider_exit_after_focus_moved_does_not_steal_focus, cargo:caller_exit_then_provider_exit_never_focuses_the_dead_caller, cargo:the_system_bell_rings, cargo:windows_keep_their_toplevel_tag, proof:G-GL-WINDOW, proof:G-WAYLAND-INFO, proof:G-SHORTCUT-CONSENT, proof:V-PICKERdocs/protocols.md lists what is advertised and each gap. G-WAYLAND-INFO runs wayland-info on the nested session and compares it with a real GNOME 51 capture (tests/protocols/gnome51-wayland-info.txt: gnome-shell/mutter 51.0, Fedora 45 headless, fedora:45@sha256:dbb22055c0c1...). The headless capture's 40 globals plus 3 source-derived floors remain in tests/protocols/gnome51-globals.tsv. Native Marlin GNOME 51 capture 37196669447 records the same 40 versions plus DRM lease v1; tests/protocols/marlin-gnome51-globals.tsv represents all 41 captured globals plus two remaining source-only floors, with raw candidate/reference captures and digest/image/package/observer manifest retained. G-WAYLAND-INFO compares both lists; 27 of the 43 are offered at GNOME's version. Added for parity with Mutter (#89): wl_compositor v6; presentation-time v2 with feedback from the frame path (kernel vblank time and sequence on hardware), fifo and commit-timing released per refresh cycle, commit-timing tv_nsec validated as Mutter does (crates/compositor/src/frame_timing.rs); xdg-dialog (modal dialogs attached to their parent), xdg-foreign v2 (portal dialogs parented to the app; a provider that exits or restarts while its focused picker is open hands focus back to the caller, as Mutter's focus_ancestor_or_top_window does, never to a dead caller and never away from an app the user moved to, #432/#404), keyboard-shortcuts-inhibit with Super+Escape restoring, pointer-warp, xdg-system-bell (GNOME's bell sound), xdg-toplevel-tag Portal file choosers from real Wayland and X11 GTK callers are parented, centred and modal on a Marlin VM (V-PICKER).13 GNOME 51 globals missing, each with its reason in docs/protocols.md: gtk_shell1 by choice; not in Smithay 0.7 (toplevel drag, session management, color management and representation, background blur, wl_fixes, xdg-foreign v1); tablet (Smithay manager v1 only, tablet input not routed); hardware only (DRM lease, syncobj; #68); Xwayland keyboard grab (rootful Xwayland only). 3 at older versions because Smithay 0.7 implements no newer (xdg_wm_base 6/7, wl_seat 9/10, text-input 1/2). Shortcut inhibition uses the GTK shell's consent dialog and GNOME PermissionStore (G-SHORTCUT-CONSENT); configurable Xwayland exemption rules remain absent. Native reference uses the derived Marlin GNOME 51 baseline, four-vCPU/6-GiB/1280×800 KVM; Bochs still advertises dmabuf v3, so GPU-capable native-backend feedback remains unqualified
P-SY-07End session: Suspend, Restart…, Power Off…, Log Out with GNOME's 60 s confirmation dialogpartialproof:G-POWER-MENU, proof:G-POWER-DIALOG, proof:G-POWER-OFF, cargo:nested_sessions_only_offer_log_out, proof:G-POWER-ESCAPE, proof:G-SUSPEND-LOCK, cargo:countdown_text_matches_gnome, proof:V-SUSPEND, proof:V-LOGOUT, proof:V-RESTART-DIALOG, proof:V-POWER-OFF-DIALOG, proof:V-POWER-BUTTON, cargo:end_session_dialog_types_and_signals_match_gnome_51, cargo:end_of_session_routes_like_gnome_session_51, cargo:end_session_errors_carry_gnome_session_names, cargo:gsd_power_key_shutdown_opens_the_shells_dialog_and_follows_its_answer, proof:G-END-SESSION-RESTART, proof:G-END-SESSION-POWEROFF, proof:G-END-SESSION-LOGOUT, cargo:power_menu_requests_end_the_session_only_as_the_dialog_confirms, cargo:menu_rows_ask_the_session_manager_like_system_actions, cargo:countdown_steps_by_ten_until_the_last_ten_seconds_like_gnome, proof:G-END-SESSION-USERS, proof:G-END-SESSION-REALNAME, proof:G-END-SESSION-BATTERY, proof:G-END-SESSION-BOOT-OPTIONS, proof:G-END-SESSION-ASYNC, cargo:other_sessions_are_filtered_and_titled_like_gnome_51, cargo:log_out_names_the_user_by_real_name_like_gnome_51, cargo:low_battery_warning_follows_gnome_51, cargo:offline_updates_shape_the_dialog_like_gnome_51, cargo:boot_options_need_logind_yes_or_challengequick-settings power menu (crates/shell-gtk/src/power.rs) that, as GNOME Shell 51's systemActions.js, asks org.gnome.SessionManager (the compositor's crates/compositor/src/session_manager.rs, or gnome-session where it runs) for Suspend, Reboot, Shutdown and Logout(0); the manager suspends through logind's SuspendWithFlags(SD_LOGIND_SKIP_INHIBITORS) and routes the rest as gnome-session 51 does (disable-log-out → LockedDown, logout-prompt, logout inhibitors, logout modes) into the shell's org.gnome.SessionManager.EndSessionDialog (Open types 0/1/2, Close, Confirmed*/Canceled/Closed): GNOME's 60 s countdown shown in its ten-second steps, the applications holding logout inhibitors listed under "Some applications are busy or have unsaved work" (services and non-logout inhibitors left out), Cancel and Escape; for Restart and Power Off the other users' logind sessions under "Other users are logged in" (greeter, closing and this session left out, at most five, graphical ones by user name and console/remote ones by AccountsService real name, as GNOME 51 titles them); Log Out's title and countdown name the user by AccountsService's real name (an empty RealName falls back to the user name, and without AccountsService the plain "Log Out" strings stay); with gnome-software's org.gnome.Software.OfflineUpdates reporting an update prepared or scheduled, Restart and Power Off offer "Install pending software updates" (checked unless the battery is low; confirming sets gnome-software's post-update action, a Power Off it accepts reboots to install, and unchecking cancels the update), a scheduled one turns a Restart into GNOME's type-3 "Restart & Install Updates", and UPower's DisplayDevice discharging below 30 % shows "Low battery power: please plug in before installing updates" while updates are to install, live as the battery changes; where logind's CanRebootToBootLoaderMenu answers yes/challenge, holding Alt swaps Restart for "Boot Options", which sets SetRebootToBootLoaderMenu(60 s) before confirming the reboot. The manager, never the shell, carries the confirmation out (logind PowerOff(true)/Reboot(true), a shutdown the shell promoted to a reboot reboots, Log Out ends the compositor, the session leader, through its orderly shutdown as gnome-session's exit ends a GNOME session); Shutdown/Reboot answer with the dialog's outcome, Logout once it opens, and a newer request supersedes an older one still waiting (private-bus tests). The GTK proof drives the hardware menu against the compositor's manager and a logind stub on a private system bus that records the calling process: G-END-SESSION-RESTART/-POWEROFF/-LOGOUT see each row reach the manager (dbus-monitor), Open(type, 0, 60, [inhibitor]), the listed inhibiting app, the countdown stepping 60 → 50, Cancel leaving the session and logind untouched with the request failed as cancelled, Restart and Power Off reaching logind from the manager's process, and Log Out (last) ending the compositor gracefully; with the stub's sessions, AccountsService and UPower and a gnome-software stub, G-END-SESSION-USERS, -REALNAME, -BATTERY and -BOOT-OPTIONS see the sessions section, the real name and its fallback, the update box, warning and Restart & Install Updates (hiding once charging, none without updates), and Alt's Boot Options reaching logind from the shell before the manager's Reboot; Suspend reaches logind from the manager and locks on PrepareForSleep (G-SUSPEND-LOCK). The nested preview offers Log Out only, through its own compositor's manager or, inside a host desktop, a local dialog that ends only this compositor. Recorded against GNOME 51 in docs/gnome-parity.md (Log Out dialog 0.4% off); on the Marlin image built in CI run 38878636595 (and the published image, nightly 38844351544) logind suspends to a real QEMU S3 state and resumes locked (V-SUSPEND), and Log Out returns to the display manager's greeter (V-LOGOUT; GDM in the current-build lanes since CI run 40977772289; the current-build lane logs out from the shell's power menu by physical keys: Log Out… reaches the compositor's manager as Logout(0), its dialog opens counting down from 60 with Cancel focused, and the session ends only once the countdown confirms); in the same lane Restart… and Power Off… from the menu open Open(2, 0, 60, …) and Open(1, 0, 60, …), step to 50 with the VM, session and logind untouched, and Return on the focused Cancel fails the request as cancelled (V-RESTART-DIALOG, V-POWER-OFF-DIALOG). Log Out ends the compositor instead of calling logind's Session.Terminate: Terminate also signalled GDM's session worker, whose switch back to the login VT logind's VT restore could cancel, leaving no greeter (published nightly runs 40883859027, 40940329310); gsd-media-keys' power key under power-button-action interactive reaches the dialog in the Marlin VM (V-POWER-BUTTON)V-SUSPEND calls logind rather than pressing the menu; a confirmed restart or power off reaches only the logind stub (the installed image cancels both dialogs); no physical hardware (#68). Offline updates, sessions, AccountsService, the battery and the boot loader menu run only against stubs (Marlin ships no gnome-software offline updates); session avatars show AccountsService's IconFile or the default avatar, without GNOME's per-user live updates; the Log Out real name is read per dialog rather than followed live. The session manager has no client registration, so no QueryEndSession client is asked before the dialog
P-SY-05Screen capture and sharing through portals with consentpartialproof:G-SCREENSHOT-DENY, proof:G-CAPTURE-DENY, proof:G-CAPTURE-LOCKED, proof:G-CAPTURE-REVOKE, proof:G-CAPTURE-LOCK-REVOKE, proof:G-PORTAL-CONSENT, proof:G-PORTAL-LOCKED, proof:G-SCREENSHOT-BUTTON, proof:G-SCREENSHOT-UI, proof:G-SCREENSHOT-KEYS, cargo:exclusive_overlay_receives_navigation_and_retains_popup_system_accelerators, cargo:exclusive_overview_keeps_overview_accelerators, proof:G-SCREENSHOT-UI-WINDOW, proof:G-SCREENCAST-UI, proof:G-SCREENCAST-DBUS, cargo:lock_admission_fails_closed_and_updates, cargo:sessions_are_owned_by_the_unique_creator_and_revoke_permanently, cargo:arbitrary_process_cannot_be_the_installed_portal, cargo:paths_follow_gnome, cargo:window_properties_follow_gnome_shell, cargo:the_window_selector_leaves_room_for_the_panel, cargo:the_window_selector_reads_the_compositors_answer, cargo:record_areas_land_on_their_monitor, cargo:areas_crop_out_of_frames, cargo:the_pipeline_is_gnomes_software_vp8, cargo:an_admitted_caller_reads_sandboxed_apps_and_windows_through_org_gnome_shell, cargo:an_unadmitted_caller_is_denied_through_org_gnome_shell, cargo:a_locked_session_denies_the_list_through_org_gnome_shell, cargo:compositor_change_signals_reach_subscribers_on_org_gnome_shell, proof:V-PORTAL, proof:V-PORTAL-GNOME, proof:V-PORTAL-SCREENSHOT, proof:V-PORTAL-SCREENCAST, proof:G-INTROSPECT-SIZE, proof:G-INTROSPECT-TYPES, cargo:a_wedged_bus_never_blocks_the_emitting_thread, cargo:emissions_arrive_in_submission_order, cargo:owner_checks_are_bounded_and_fail_closed_on_a_wedged_bus, cargo:publishing_never_waits_on_a_wedged_bus, cargo:queued_publication_arrives_in_orderscreenshots of the screen or the focused window (org.gnome.Shell.Screenshot), monitor, area and window screen casts (org.gnome.Mutter.ScreenCast with RecordArea and DisplayConfig, PipeWire), and the portal's window list (org.gnome.Shell.Introspect, portal callers only) — what xdg-desktop-portal-gnome calls, niri's shape; as in GNOME 51 the Introspect object also answers under org.gnome.Shell, where the portal's Background implementation reads running apps and sandboxed-app-id (the GTK shell relays it; the compositor admits the original caller with the same rule, #410); the screenshot UI's window selector (the active workspace's windows spread as in the overview, the chosen one ringed in the accent color, saved alone at full size); screen recording from the UI's video switch through org.gnome.Shell.Screencast (GNOME's software VP8/WebM pipeline into Videos/Screencasts) with the red panel indicator that stops it; on the Marlin image built in CI run 38878636595 (and the published image, nightly 38844351544) the portal frontend, the GNOME backend and the Screenshot/ScreenCast interfaces routed by tuna-portals.conf answer on the user bus on 5/5 boots (V-PORTAL*; availability only, not a consent flow), and that run's portal-security and remote-security jobs repeat the genuine GNOME 51 Screenshot/ScreenCast/RemoteDesktop consent, lock and revoke flows; ScreenCast stream/close signals and Introspect change signals leave the event loop through an ordered queue, as GDBus emission does, so a bus that stops reading cannot stall frames or input, and a bus that stops answering revokes capture owners within 5 s (fail closed) instead of stalling the revocation scan (#657)Qualified packaged evidence records all sixteen current-head gates and genuine GNOME 51 Screenshot Deny/Allow/locked completion, ScreenCast consent/frame/Close/disconnect/lock and independent libei Remote keyboard/pointer delivery; #61 acceptance is complete, while Remote touch/keysym/text/clipboard, EIS snapshot refresh and interactive portal screenshot area/color remain unsupported; the installed GNOME backend native-display bootstrap and its retained pre-picker failure are documented in portal-service-channel.md; Screenshot consent and locked completion use GNOME capture with GTK Access UI; admission now requires the supervised shell or root-owned installed GNOME portal executable plus its unique bus identity, sessions belong to their creator, and disconnect/lock revoke streaming; the Background portal's org.gnome.Shell path is covered on a private bus with the real relay and service, while its installed-portal VM qualification (Background.GetAppState) is pending (#410); dmabuf zero-copy and cursor metadata are missing, so neither casts nor recordings show the pointer (nor does the UI's Show Pointer); recordings run GStreamer through gst-launch-1.0 rather than in-process, without GNOME's hardware-encoder pipelines; the window selector shows every window on the primary monitor rather than one selector per monitor (#61)
P-A11Y-01Screen reader reads panel, overview, quick settingspassproof:A11Y-PANEL, proof:A11Y-QS, proof:A11Y-GOLDENLive GTK named controls/actions and committed GNOME 51 AT-SPI visible trees compared against Tuna Desktop goldens in tests/a11y/gnome51/comparison.md; roles, hierarchy, unnamed native controls and service-fixture differences are explicit; comparison checked in CISpoken Orca reading is gated separately (P-A11Y-06)
P-A11Y-02Every pointer action has a keyboard equivalentpartialcargo:panel::tests::live::key_only_run_opens_launches_and_activates, proof:G-WM-REBIND, cargo:gnome_51_defaults_without_the_schema, proof:G-SWITCHER-REBIND, cargo:rebound_switcher_keys_replace_the_builtin_chords, cargo:switcher_keys_default_and_parse_like_mutter, proof:G-PANEL-FOCUS, proof:G-KEYBINDINGS, proof:G-WM-KEYS, proof:G-OVERVIEW-STATE-KEYS, proof:G-CYCLE-KEYS, proof:G-WINDOW-SWITCHER, proof:G-NOTIFICATION-FOCUS-KEY, proof:G-A11Y-MOUSEKEYS, journey:J-OV-ESC, journey:J-SW-STEP, journey:J-WS-NEXT, proof:G-ORCA-SPEECH-PREVIEWS, proof:G-ORCA-SPEECH-GRID, proof:G-PANEL-MENU-KEYS, proof:G-QS-MENU-KEY, proof:G-NOTIFICATION-KEYS, proof:G-WINDOW-MENU-KEYS, proof:G-SCREENSHOT-CTRL-C, cargo:delete_and_backspace_close_messages_without_modifierskeymap, key-only run and the pointer-action audit matrix (docs/keymap.md); every GNOME chord rebinds, the switcher's included; the panel focus walk (Ctrl+Alt+Tab), Down into and Left/Right between panel menus, quick settings' Shift+F10/Menu, notification Delete/Escape and group expand/collapse, window-menu arrows, screenshot Ctrl+C, overview window-preview and app-grid keyboard navigation, overview state keys, window-manager keys, the notification focus key and Mouse Keys are proven in the nested GTK shell; open: app-grid Home/End, the Ctrl+Alt+Tab group popup, app-icon context menus (absent as pointer actions too), and no installed-image keyboard-only gate exists (#591)—
P-IN-01IME composition in GTK appspartialcargo:an_input_method_types_into_the_focused_text_field, proof:G-IME, cargo:surrounding_offsets_count_characters_not_bytes, cargo:ibus_deletions_become_byte_lengths_around_the_cursor, cargo:surrounding_text_is_packed_as_an_ibus_text, cargo:ibus_text_encoding_matches_the_libibus_wire_shape, cargo:raw_and_nested_ibus_text_decode_without_native_type_assertions, cargo:content_types_map_to_ibus_purposes_and_hints, cargo:content_type_is_set_as_the_input_context_property, cargo:the_bridge_hears_the_text_cursor_in_global_coordinates, proof:G-IME-SURROUNDING, proof:G-IME-CURSOR, proof:G-IME-CURSOR-MOVE, proof:G-IME-ESCAPE, proof:G-IME-SHELL-ESCAPE, proof:G-IME-XIM, cargo:a_serialized_lookup_table_is_read_like_libibus, cargo:the_page_holds_the_cursor_with_gnome_index_labels, cargo:later_pages_and_labels_and_horizontal_tables, cargo:the_popup_sits_under_the_cursor_and_stays_on_the_monitor, cargo:a_press_never_gives_the_keyboard_to_a_surface_that_wants_none, proof:G-IME-CANDIDATES, proof:G-IME-PROPERTIES, proof:G-IME-PROPERTIES-LIFECYCLE, proof:G-IME-VARIANT-TYPES, proof:G-IME-TRACE-PRIVATE, cargo:ibus_properties::tests::typed_modern_and_explicit_legacy_properties, cargo:ibus_properties::tests::radio_group_matches_visible_siblings_not_submenus_or_separators, cargo:ibus_properties::tests::same_name_engine_and_same_path_focus_edges_retire_actions, cargo:ibus_property_menu::tests::old_owner_replies_and_absent_or_replaced_owner_are_refusedtext-input v3 and input-method v2 relay commits; candidate popups tracked; IBus as in GNOME: tuna-ibus-bridge is IBus's client (GNOME Shell's part), started by the compositor on a private socket when IBus is installed, starting ibus-daemon --panel disable itself; preedit and commits reach the app, every key in a focused field goes through IBus; unhandled keys return through a virtual keyboard only the bridge may hold, with the physical keymap preserved for GTK popovers; ibus-x11 is supervised only against the private XWayland display; window moves refresh the global caret; the field's surrounding text (SetSurroundingText, in characters), content type (the input context's ContentType property, as libibus sets it, so engines leave password fields alone) and the engine's DeleteSurroundingText flow as GNOME Shell's do; the text cursor reaches SetCursorLocation in global coordinates (the bridge holds a bufferless input popup and the compositor re-reports the cursor rectangle to it, and only it, globally); the GTK shell is IBus's panel as GNOME Shell is (it owns org.freedesktop.IBus.Panel on IBus's bus, so the daemon started with --panel disable routes the lookup table, auxiliary text and cursor location to it) and draws GNOME 51's candidate popup (ibusCandidatePopup.js, _ibus-popup.scss): the page's candidates with index labels, the cursor's row highlighted, vertical or horizontal, page buttons, 6px under the cursor; a click is CandidateClicked (the popup never takes the keyboard), the page buttons PageUp/PageDown, the wheel CursorUp/CursorDown; the panel also consumes IBus engine properties as GNOME 51's keyboard.js does (crates/shell-gtk/src/ibus_property_menu.rs, crates/shell-host/src/ibus_properties.rs): toggles, radio groups, submenus and actions with hidden/disabled state, an InputMode symbol on the indicator, provenance-pinned daemon peers and actions retired on engine/context/owner/lock changes (G-IME-PROPERTIES, G-IME-PROPERTIES-LIFECYCLE; scope)IBus deletions not touching the cursor are dropped (input-method-v2 deletes only around it); the candidate popup has no on-screen-keyboard suggestions and ignores SetCursorLocationRelative (only IBus's own GTK module sends it; the bridge sends global coordinates); the panel's IBus menu shows engine properties but does not list or choose the configured XKB/IBus input sources (#592); an fcitx5-style input method takes over when it binds first; no installed-image or GNOME 51 capture
P-IN-02Touchpad gestures (three-finger overview, workspace swipe)partialproof:D-TOUCHPAD-SWIPE, cargo:three_or_more_fingers_are_the_shells_and_fewer_propagate, cargo:travel_under_the_drag_threshold_moves_nothing_and_is_dropped, cargo:an_exact_diagonal_commits_to_the_vertical_tracker, cargo:natural_scrolling_off_reverses_the_gestures, cargo:a_held_swipe_follows_the_fingers_under_every_motion_policy, cargo:a_swipe_point_belongs_to_the_monitor_under_it, cargo:the_transition_grows_the_card_and_glides_windows_like_gnome, cargo:neighboring_workspaces_and_thumbnails_follow_intermediate_swipe_progress, proof:G-SWIPE-PROGRESS, proof:G-SWIPE-DIRECTIONS, proof:G-SWIPE-REDUCED-MOTION, proof:G-SWIPE-FOLLOW, proof:G-SWIPE-VELOCITY, cargo:a_swipe_follows_the_fingers_then_finishes_on_its_velocity, cargo:a_flick_completes_faster_than_a_drag, cargo:velocity_is_the_last_150_ms_of_travel, proof:G-SWIPE-MOMENTUM, cargo:swipe_release_projects_velocity_like_gnome, cargo:a_released_swipe_keeps_its_momentum, cargo:right_to_left_reverses_only_the_horizontal_trackers, cargo:the_swipe_runs_from_the_desktop_through_the_window_picker_to_the_app_grid, cargo:a_swipe_up_from_the_window_picker_snaps_to_the_app_grid, cargo:a_held_app_grid_swipe_draws_between_the_picker_and_the_grid, cargo:an_overview_workspace_swipe_follows_and_settles_across_workspaces, cargo:overview_swipe_follows_workspace_progress_and_is_bounded, proof:G-SWIPE-APPGRID, proof:G-SWIPE-OVERVIEW-SCROLLthree-finger swipe up/down toggles the overview, sideways switches workspace; other swipes reach apps; the overview follows a vertical three-finger swipe and settles where GNOME's swipe tracker projects the release, finishing with an out-cubic whose duration follows the fingers' velocity Real libinput (proof:D-TOUCHPAD-SWIPE, drm-smoke on vkms): a uinput Synaptics TM3053-004 clickpad (the rmi4 driver's capabilities, libinput's litest-device-synaptics-rmi4.c) whose kernel-timestamped finger motion libinput classifies as a touchpad and turns into GESTURE_SWIPE events. Through it, three fingers up hold the overview part way (0.42, then 0.77) and open it eased; down closes it; left/right slide to the next workspace and back; a slow half swipe settles back (velocity 0); a short 8 mm flick completes from 0.21 at 3.2 units/ms; four fingers are the shell's too; two-finger scrolling and pointer motion reach a GTK 4 client, no three- or four-finger swipe does; natural-scroll off reverses scrolling and both gestures; reduced motion keeps following and the eased release; animations off follow and land at once. A long swipe up from the desktop stops at the window picker; from the window picker three fingers up draw the app grid part way (its shell surface maps and follows) and the release lands on the app grid, the shell's grid easing over the swipe's settle time; down comes back to the picker. In the open overview (four fixed workspaces) held fingers scroll the workspaces part way and past the next one with the active workspace unchanged, a held release settles on the nearest, and a fast flick crosses two or more in one ease. Compositor matches GNOME 51's TouchpadSwipeGesture (js/ui/swipeTracker.js:120-204 at 51.0): three or more fingers (:35, :127), unaccelerated deltas (:142), a 16-unit drag threshold that commits to one axis and drops the travel before it (:153-179), the natural-scroll sign (:188-189), libinput cancel landing as a release (:194-198), a still history sample at begin (:509-513), and a horizontal swipe over a non-primary monitor moving nothing while workspaces are only on the primary (workspaceAnimation.js:509-512, workspacesView.js:961-962). The vertical swipe drives GNOME's controls state across the desktop, the window picker and the app grid (ControlsState 0-2, overviewControls.js:33-37, 757-797; overview.js:355-397), kept within the snap points either side of where it began (swipeTracker.js:547-576, no long swipes for the overview), its release projected over those three points (:601-658); the workspaces lerp between the window picker and app grid layouts with the state (overviewControls.js:588-600) and the shell's grid follows it (wire 0.49 OverviewSwipe: shown above the window picker, _updateAppDisplayVisibility :576-586; Show Apps set on release, gestureEnd :781-786). In the open overview a sideways swipe drives the workspaces' scroll position live, every workspace a snap point with long swipes (workspacesView.js:876, 960-1002), settling with the projected velocity in an EASE_OUT_CUBIC; a swipe caught mid-ease starts from where it is drawn. Under an RTL shell text direction (InputSettings.right_to_left) the horizontal delta is negated (swipeTracker.js:593-595). The overview follows the fingers under every motion policy (overview.js:359-380, overviewControls.js:777-779) and its release takes animationUtils.js:23-33 durations; locking mid-swipe interrupts it (swipeTracker.js:487-491).Remains: genuine physical touchpad hardware and paired GNOME 51 captures (the emulated clickpad is libinput-real but not a physical sensor); the secondary-monitor primary-only rule is unit-tested only (vkms gives one output); the app grid's arrival is drawn with the shell's existing picker/grid fade driven by the swipe, where GNOME 51 slides the app display up from below the monitor (overviewControls.js:113-134, 221-239); under RTL only the swipe direction flips: Tuna's desktop workspace strip and overview cards are not mirrored as GNOME's are (workspaceAnimation.js:247, 287-317; workspacesView.js:178-180, 338-355), and the flip is unit-tested only (no RTL drm-smoke session)
P-ST-01Settings written by GNOME Settings take effect (wallpaper, clock, fonts)partialcargo:panel::tests::live::wallpaper_survives_session_restart, cargo:drop_follows_gnome_background_settings, cargo:settings_changes_republish_the_drop_and_failed_writes_keep_the_last, proof:G-QS-DARK, proof:G-QS-EXTERNAL-SYNC, proof:G-SETTINGS-IDLE, proof:G-SETTINGS-INPUT, proof:G-ANIMATIONS-OFF, proof:G-SETTINGS-CLOCK, proof:G-SETTINGS-FONT, proof:G-SETTINGS-NOTIFICATIONS, journey:J-SETTINGS-LIVE, journey:J-SETTINGS-RESTORE, proof:G-DISPLAY-SETTINGS, cargo:input_sources_become_one_keymap, cargo:idle_lock_follows_gnome_keys, proof:G-ACCEL, proof:G-KEYBINDINGS, proof:G-KEY-BRIGHTNESS, proof:G-WM-KEYS, proof:G-WM-REBIND, proof:G-SWITCHER-REBIND, proof:G-WINDOW-SWITCHER, proof:G-CYCLE-KEYS, proof:G-OVERVIEW-STATE-KEYS, proof:G-NOTIFICATION-FOCUS-KEY, proof:G-MONITOR-BRIGHTNESS-KEYS, cargo:window_switcher_keeps_same_app_windows_and_scopes_the_workspace, cargo:monitor_brightness_never_uses_a_different_connectors_backlight, proof:G-SET-WALLPAPER, proof:G-SET-CLOCK, proof:G-SET-FONTS, proof:G-WALLPAPER-URI, cargo:escaped_wallpaper_selects_real_pixels_without_double_decodingComplete inventory classifies all 428 GNOME 51 desktop keys with consumer/toolkit ownership or an ignored-key rationale; live settings map summarizes handlers; actual GTK proofs exercise font family/size/style, text scale and external notification policy, and the journey changes/restores the running clock; GNOME Shell's keybindings and the window-manager keybindings are honored and rebind live, the switcher's and Super+PageUp/PageDown included; G-SET-WALLPAPER proves picture-uri writes publish, identify, decode and visibly repaint (with file-replacement refresh, an ignored invalid URI and revert), G-WALLPAPER-URI proves URIs with spaces, Unicode and literal percent escapes select the real wallpaper pixels, G-SET-CLOCK proves clock-format/show-date render live with invalid-format handling and revert, and G-SET-FONTS proves document/monospace fonts reach GTK apps live with a visible repaint, bogus-family fallback and revert; installed image 2026-10-30 (Marlin sha256:ae6363c1…, Corral VM, evidence): stock GNOME Settings 51.1 with real pointer input changed 24-hour format, weekday and seconds and Text Size, each visible at once in the panel clock and apps, and all persisted across GDM log-out/log-in; font-name/monospace/document were set by gsettings because GNOME Settings 51 has no font UI — font-name repainted the shell, monospace reached new Ptyxis tabs (app-owned), document-font-name has no consumer installed on the imagenative volume keys mutate PipeWire through bounded wpctl calls; brightness is exercised against a fixture, with physical backlight validation pending; intentionally ignored keys are documented in the inventory; background placement, shading and slideshow scope is tracked in P-ST-08
P-TR-01AppIndicator tray items show icons and menuspass (nested)cargo:watcher::tests::live_bus::stub_item_serves_icon_and_menu_over_private_bus, cargo:registrations_resolve_like_the_appindicator_extension, cargo:menu_layout_builds_the_whole_tree_like_the_extension, cargo:accessible_name_and_icon_follow_status_like_the_extension, proof:G-TRAY, proof:G-TRAY-POINTER, proof:G-TRAY-AYATANA, proof:G-TRAY-X11Measured in the nested proof against the AppIndicator extension Ubuntu and Fedora ship (appindicatorsupport@rgcjonas.gmail.com, indicatorStatusIcon.js/dbusMenu.js): a StatusNotifierItem stub and a real libayatana-appindicator3 item, from a Wayland client and from an XWayland (GDK_BACKEND=x11) client, show their icon, label and tooltip; middle click is SecondaryActivate, the wheel Scroll, a double click Activate, single and secondary clicks the menu; the menu, submenu and check rows work from the keyboard (Ctrl+Alt+Tab, Enter, arrows, Right/Left); live layout, label, attention and Passive updates apply in place; items leave when their app exits. The host runs org.kde.StatusNotifierWatcher on a worker thread (crates/shell-host/src/watcher.rs, crates/shell-gtk/src/tray.rs)Stock GNOME 51 shows no tray; it needs that extension, which Tuna Desktop builds in (owner: shell). Not measured: no paired pixel capture against GNOME with the extension, and legacy XEmbed tray icons (the extension's TrayIconsManager) are not hosted
P-WM-08Multitasking preferences: fixed workspace count, workspaces on primary only, edge tiling, new-window centering and current-workspace-only app switchingpartialcargo:fixed_workspaces_keep_empty_slots_and_reject_out_of_range_commands, cargo:shrinking_fixed_workspaces_preserves_window_ids_titles_and_focus, cargo:fixed_count_clamps_to_schema_range_and_insertion_uses_last_valid_slot, cargo:removing_last_window_preserves_fixed_slots_and_dynamic_reenable_prunes_them, cargo:fixed_workspace_layout_has_exact_slots_and_no_insertion_placeholder, cargo:gaps_compact_without_changing_window_identity_contents_or_focus, cargo:dialog_grace_expires_at_one_second_and_has_no_past_due_deadline, cargo:greatest_overlap_selects_secondary_and_primary_wins_ties, cargo:primary_change_moves_workspace_affinity_without_moving_windows, cargo:all_workspace_window_state_survives_delta_wire, cargo:multitasking_preferences_survive_positional_wire, cargo:application_workspace_filter_scopes_groups_counts_and_thumbnails, cargo:window_workspace_filter_is_independent_and_changes_live, cargo:enabling_filter_reconciles_an_open_remote_selection, proof:G-TILE-PREVIEW, proof:G-WS-INSERT, cargo:secondary_output_windows_are_sticky_and_unstick_onto_the_active_workspace, cargo:edge_tiling_off_shows_no_preview_and_drops_floating, cargo:uncentred_placement_takes_mutters_first_fit_then_corner_cascade, cargo:center_new_windows_survives_positional_control_wire, proof:G-EDGE-TILING-OFF, proof:G-SWITCHER-WORKSPACE, proof:G-WS-FIXED, proof:G-CENTER-NEW-WINDOWSThe shell reads num-workspaces (1–36), dynamic-workspaces, workspaces-only-on-primary, edge-tiling and center-new-windows (protocol 0.41) live into InputSettings (crates/shell-gtk/src/main.rs); the compositor keeps fixed slots, moves windows from removed workspaces to the new last one, compacts dynamic ones with GNOME's 1 s dialog grace and departed-workspace keep-alive leases woken by the idle scheduler (state.rs, workspace_tracker.rs), shows secondary-output windows on every workspace and lands them on the active workspace when they leave that state (mutter meta_window_on_all_workspaces_changed), gates pointer edge tiling, and places new windows centred or by mutter's find_first_fit (windows.rs); the app and window switchers apply their independent current-workspace-only keys live (crates/shell-host/src/model.rs). Nested gates G-WS-FIXED, G-EDGE-TILING-OFF, G-SWITCHER-WORKSPACE and G-CENTER-NEW-WINDOWS prove each toggle's visible effect. Still missing: physical multiple-output qualification (nested proof has one output; workspaces-only-on-primary is proven by a two-output cargo test only), startup-sequence workspace leases (GNOME produces them only for launches onto a named workspace, a path Tuna's overview lacks), installed-session and GNOME 51 comparison (#337; scope)—
P-WM-09The full GNOME window-manager shortcut list and focus policy (focus mode, auto-raise, raise on click, mouse-button modifier)partialproof:G-WM-KEYS, proof:G-WINDOW-MENU, proof:G-CYCLE-KEYS, proof:G-WINDOW-SWITCHER, proof:G-WM-REBIND, proof:G-PANEL-FOCUS, cargo:workspace_targets_resolve_like_mutter, cargo:gnome_51_defaults_without_the_schema, proof:G-FOCUS-SLOPPY, proof:G-FOCUS-AUTORAISE, proof:G-FOCUS-MOUSE, proof:G-FOCUS-RAISE-ON-CLICK, proof:G-FOCUS-MODIFIER, cargo:sloppy_focus_follows_enter_without_raising_and_keeps_focus_on_the_desktop, cargo:mouse_focus_unfocuses_when_the_pointer_reaches_no_window, cargo:pointer_rest_focuses_only_once_the_pointer_stops, cargo:auto_raise_raises_after_its_delay_only_under_the_pointer, cargo:raise_on_click_off_focuses_without_raising_and_still_delivers, cargo:short_modifier_grab_raises_at_release_without_raise_on_click, cargo:activation_focus_stays_until_the_pointer_travels_fifty_pixels, cargo:token_activation_raises_only_with_raise_on_click, cargo:mouse_button_modifier_moves_the_drag_to_its_key, cargo:modifier_resize_and_menu_buttons_follow_resize_with_right_button, cargo:modifier_parses_like_meta_parse_modifier, cargo:actual_gio_watch_delivers_live_focus_mode_changes_on_the_owned_contextListed accelerators and the six switch/cycle bindings work (Alt+Space, Alt+F10, Super+End/Home, window cycling, switch-panels and live rebinding are proven in the nested GTK shell); numbered workspaces 2–12, monitor moves and fullscreen/above/sticky/show-desktop keys still have no registration or consumer (#347). Focus policy follows Mutter 51.0 (src/core/display.c, window.c, prefs.c, meta-accel-parse.c, compositor/meta-window-drag.c), read live by the compositor from org.gnome.desktop.wm.preferences on an owned GIO thread (crates/compositor/src/wm_preferences.rs, decisions in focus_policy.rs): focus-mode sloppy/mouse focus on enter without raising (mouse also unfocuses over the desktop and shell chrome), honoring org.gnome.mutter focus-change-on-pointer-rest (GNOME Shell's session default on when the schema is absent) and Mutter's 50 px sticky focus after an activation; auto-raise after auto-raise-delay only in sloppy/mouse; raise-on-click false focuses clicks and client (xdg-activation) activations without raising, grabs raise only when shorter than 8 px, shell activations still raise; mouse-button-modifier selects the move/resize/menu modifier (Mod2/Mod3/Mod5 contribute nothing, a deliberate narrowing) and resize-with-right-button swaps resize and menu. The scrollable strip keeps click focus (Tuna's own layout). Pointer-rest and sticky focus are unit-tested only; the nested proof drives sloppy, auto-raise, mouse, raise-on-click and modifier changes live—
P-SY-08Session save and restore (opt-in window restoration across logout and reboot)partialproof:G-SESSION-RESTORE, proof:G-SESSION-RESTORE-STATE, proof:G-SESSION-RESTORE-OPTIN, cargo:a_restarted_compositor_puts_session_windows_back, cargo:restoring_a_toplevel_already_mapped_but_not_committed_still_applies, cargo:windows_record_their_state_when_they_close_not_when_their_session_goes, cargo:unknown_ids_start_new_sessions_and_removal_forgets, cargo:another_client_takes_a_session_over_and_the_same_client_cannot_twice, cargo:session_requests_raise_mutters_protocol_errors, cargo:the_global_is_advertised_at_mutters_version, cargo:store_round_trips_mutter_keys_and_state_numbers, cargo:newer_session_data_restores_nothing_but_the_session_stays_known, cargo:wrongly_typed_fields_are_skipped_not_the_toplevel, cargo:restored_rectangles_stay_where_an_output_shows_them, cargo:bus_callers_inhibit_and_their_departure_releases_the_idle_lockGNOME 51 restores windows only for clients that opt in through Mutter 51's xdg_session_manager_v1 v1; gnome-session 51 has its app save/restore switched off (RestoreSupported false), so GNOME Settings hides "Reopen Windows After Login" and save-restore changes nothing. The compositor now does the same (crates/compositor/src/xdg_session.rs, docs/protocols.md): the global is advertised at v1, a session's toplevels record workspace, floating and maximized/tiled/fullscreen rectangles, minimized and output when their window closes while the session is live and at compositor exit, as Mutter records them (on_window_unmanaging; a destroyed or disconnected session detaches its toplevels first, xdg_session_destructor); like Mutter, minimized is written but never restored; the store ($XDG_DATA_HOME/tuna/window-sessions.json, Mutter's keys and state numbers) is written 3 s after a change and synchronously at exit; restore_toplevel before the first commit puts the window back with restored ahead of its first configure; org.gnome.SessionManager reports RestoreSupported false. G-SESSION-RESTORE (scripts/tuna-session-restore-proof, gtk-shell job) moves and resizes a session window with real pointer drags, sends it to workspace 2 and maximizes another with real keys, SIGTERMs the nested compositor, starts it again and checks the reconnecting client gets both back; the same client without a session is placed as usual. The unwired gnome-session relaunch facade (session_restore.rs, #654) was removed. Not yet proven: logout and reboot of the installed Marlin session (tuna-session stopping the compositor), multi-output restore on hardware, and a stock GNOME 51 app using the protocol (#338)Store is JSON at $XDG_DATA_HOME/tuna/ rather than Mutter's session.gvdb; also kept: output name and the floating rectangle under maximized/tiled/fullscreen; at most 256 sessions (least recently used go) of 256 toplevels; remove_toplevel forgets unheld names; already_added, invalid_reason and rename honoured (Mutter 51 does not); a rectangle no output shows moves onto one — owner #338
P-SY-09Clients find the remaining Mutter protocols and versions, including native DRM lease, explicit sync and dmabuf feedbackpartialRecorded deviations: xdg_wm_base 6 vs 7, wl_seat 9 vs 10, text-input v1 vs v2, xdg-foreign v1, toplevel drag, background effects, wl_fixes, DRM lease, explicit sync, dmabuf feedback (docs/protocols.md, #364); see P-SY-06 for what is advertised—
P-SY-10Screenshot UI keyboard selection: arrows resize, Alt moves, Shift/Ctrl coarse and fine steps, R resetspartialproof:G-SCREENSHOT-KEYS, cargo:resize_selects_an_edge_on_axis_change_and_preserves_orientation, cargo:movement_and_coarse_resize_stay_inside_the_output, cargo:odd_output_reset_matches_gnome_integer_endpoint_geometry, cargo:crossing_edges_and_reset_keep_selection_nonempty_and_bounded, proof:G-SCREENSHOT-KEYS-SCREEN, cargo:cursor_follows_selected_endpoint_across_edges_and_centres_moves, proof:G-SCREENSHOT-KEYS-BINDINGS, cargo:keys_match_gnome_bindings_exactly, cargo:arrows_take_gnomes_modifier_combinations_onlyThe actual exclusive-keyboard screenshot overlay handles arrows, Alt movement, Shift full-output increments, Ctrl one-pixel increments and R reset (crates/shell-gtk/src/screenshot_ui.rs, screenshot_selection.rs). Keys match gnome-shell 51.0's ScreenshotUI binding pool exactly (keyval plus held Shift/Ctrl/Alt/Super/Hyper, locks ignored, as Clutter's clutter_binding_pool_activate): Ctrl+C captures like Return and Space while C alone picks Screen; Shift+letters, Ctrl+S, KP_Enter and Ctrl+Shift or Super arrows have no binding. G-SCREENSHOT-KEYS checks the accessible rectangle and saved PNG dimensions for resize/move/edge/reset, while ensuring keys do not switch workspaces; G-SCREENSHOT-KEYS-SCREEN checks that modified Screen navigation leaves the pointer and workspace unchanged; G-SCREENSHOT-KEYS-BINDINGS saves a real Ctrl+C capture after unbound chords left the mode and area alone. Unit tests cover oriented endpoint crossings, modifier precedence, exact binding matches, odd output sizes and reset; paired GNOME 51 interactive capture, multi-monitor screen-selector navigation and hardware qualification remain missing (#363)—
P-DS-01Night Light: colour temperature, schedule, manual times and temporary disablepartialproof:G-QS-NIGHT, proof:G-QS-EXTERNAL-SYNC, cargo:defaults_match_the_gnome_schema, cargo:temperature_clamps_to_range, cargo:neutral_temperature_is_unity, cargo:tint_shader_warms_and_identity_is_exact, cargo:controller_never_programs_identity, cargo:automatic_edges_smear_around_sunset_and_sunrise, cargo:automatic_window_follows_local_dates, cargo:fixed_window_follows_local_hours, cargo:fixed_window_handles_wrap_and_full_day, cargo:temp_disable_suppresses_until_sunrise_or_24h, cargo:temp_disable_clears_at_the_local_to_hour, cargo:color_properties_round_trip_through_the_share, cargo:preview_forces_temperature_then_expires_to_schedule, cargo:preview_rejects_invalid_durations_without_mutation, cargo:actual_gio_memory_watch_updates_typed_cache_on_owned_context_without_main_native_calls, cargo:night_light_clear_and_temp_disable_wake_idle_loop, cargo:system_zone_drives_schedule_across_dst_and_runtime_changes, cargo:temperature_jumps_ease_over_five_seconds_like_gsd, cargo:temperature_property_applies_until_night_light_moves, cargo:sunrise_and_sunset_publish_local_hours_for_the_automatic_schedule, cargo:color_temperature_and_sun_properties_follow_gsd, cargo:color_changes_follow_gsd_notifications_only_on_change, cargo:share_changes_reach_the_wire_as_gsd_properties_changed, cargo:every_change_maps_to_its_gsd_property_and_typeThe compositor reads all six org.gnome.settings-daemon.plugins.color keys live through an owned GIO worker (crates/compositor/src/night_light_settings.rs); its controller applies GNOME 51's temperature clamp, automatic and manual schedules in the system's local time (re-reading the zone each second, so DST and runtime timezone changes apply; from>to wraps midnight), the one-hour edge smear, the 10 K hysteresis and gsd 51's 5 s / 50 ms jump smoothing (gsd_night_light_smooth_cb), and the result tints every output in GL each frame (night_light.rs, runtime.rs); DisplayConfig.NightLightSupported is true and the compositor serves gsd 51's full org.gnome.SettingsDaemon.Color surface — NightLightActive, read-write Temperature (1000–10000 K, held until Night Light's own temperature moves), read-write DisabledUntilTomorrow, Sunrise/Sunset (local hours, -1 until a location) and NightLightPreview, emitting one PropertiesChanged per actual change exactly where g-s-d notifies (mutter.rs). G-QS-NIGHT now drives NightLightPreview over the session bus, watches Temperature ease to the key and back to 6500 K, and requires the matching PropertiesChanged signals in gdbus monitor. Still missing: nothing feeds geoclue coordinates, so automatic mode uses the stored night-light-last-coordinates or the fixed window; Kelvin uses the Helland approximation rather than colord's blackbody and the hardware cursor is not tinted; no proof inspects the rendered tint pixels, and there is no GNOME 51 or installed-image comparison (#339)—
P-DS-02Display modes, refresh rate, rotation, mirroring and output disabling from GNOME Settingspartialproof:G-DISPLAY-SETTINGS, proof:D-MODES, cargo:display_settings_requests_are_checked, cargo:display_settings_apply_requested_modes_and_disabled_outputs, cargo:display_rotation_validates_mutter_transforms, cargo:display_config_signatures_match_mutter_xml, cargo:current_state_lists_every_mode_like_mutter, cargo:display_config_applies_and_announces_over_a_private_bus, cargo:verify_rejects_what_mutter_rejects, cargo:verify_accepts_modes_mirroring_and_disabling, cargo:persistent_changes_await_confirmation_then_save_or_revert, cargo:a_newer_apply_supersedes_the_pending_confirmation, cargo:connector_modes_sort_and_deduplicate_like_mutter, cargo:supported_scales_match_mutter_fractions, cargo:sessions_start_with_the_stored_modes_mirrors_and_disabled_monitors, cargo:modes_mirrors_and_disabled_monitors_round_trip_in_mutter_format, cargo:saving_replaces_only_the_same_monitor_set, cargo:rotation_round_trips_in_gnome_format, cargo:rotation_change_repaints_and_resets_damageorg.gnome.Mutter.DisplayConfig follows Mutter 51 (crates/compositor/src/display_config.rs, mutter.rs): GetCurrentState lists every connector mode deduplicated and sorted as meta_monitor_normal_generate_modes, with Mutter's mode ids, is-current/is-preferred/is-interlaced, supported scales (meta_monitor_calculate_supported_scales) and DPI preferred scale; ApplyMonitorsConfig checks the serial (stale = AccessDenied) and runs Mutter's verification; verify/temporary/persistent behave as Mutter, a persistent change awaits GNOME Shell's "Keep these display settings?" dialog (shell protocol 0.41) and reverts after 20 s or on Revert Settings, Keep Changes saves Mutter-format monitors.xml (modes, mirrors, <disabled>) to ~/.config/tuna/monitors.xml. On DRM, mode/refresh changes are atomic-tested before commit (all-or-nothing, failed commits roll back), outputs switch off and mirror; D-MODES switches the vkms CRTC and observes the timeout revert. Unproven/different: mirroring and disabling have no runtime gate (vkms has one connector); Mutter advertises a wl_output per mirrored monitor where Tuna advertises the leader only; Mutter's synthesized fallback modes for scaling mode connectors, VRR modes, physical layout mode, underscanning and colour modes are not offered; the nested output exposes only its window-size mode and cannot rotate (#340, rotation from #345)—
P-DS-03Colour management: ICC profiles, calibration, HDR and YUV colour conversionpartialproof:D-COLORD, proof:D-GAMMA, proof:D-CALIBRATE, proof:D-COLOR-RESTART, cargo:colord_client_registers_devices_profiles_and_follows_assignment, cargo:colord_client_reuses_an_existing_device, cargo:colord_absence_is_reported_not_fatal, cargo:white_point_skips_identity_until_something_was_programmed, cargo:white_point_falls_back_to_ctm_then_to_the_renderer, cargo:calibration_sessions_program_raw_ramps_at_the_default_temperature, cargo:calibration_session_admission_follows_mutter, cargo:uncalibrated_ramp_matches_mutter_integer_math, cargo:calibrated_ramp_scales_the_vcgt_by_the_blackbody, cargo:blackbody_follows_colords_planckian_table, cargo:device_profiles_follow_the_color_store_rules, cargo:device_properties_match_generate_color_device_props, cargo:edid_identity_matches_mutter_fields, cargo:srgb_profile_round_trips_through_the_decoder, cargo:vcgt_tables_and_formulas_decode_like_lcmsSoftware/KMS path after mutter 51.0 meta-color-{manager,device,profile,store,calibration-session}.c (crates/compositor/src/color_manager.rs, edid.rs, icc.rs, drm/color.rs): colord devices with Mutter's ids and metadata, EDID/sRGB device profiles in $XDG_DATA_HOME/icc, ICC-store import for GNOME Settings, the default profile's vcgt × colord's Planckian blackbody on GAMMA_LUT (CTM fallback), Night Light on the CRTC instead of the GL tint on hardware, org.gnome.Mutter.ColorManager calibration sessions. drm-smoke on vkms reads GAMMA_LUT back with KMS ioctls against a real colord on a private system bus (emulated device per ADR 0008 item 5). Remains open: wp_color_manager_v1 and wp_color_representation_manager_v1 (not advertised), HDR/BT.2100 colour modes, YUV conversion, EFI panel colour info, real-hardware and installed-image qualification (#341)Night Light applies without colord and the GL tint remains for CRTCs without GAMMA_LUT/CTM (Mutter applies nothing then: update_white_point) — colour owner
P-DS-04Touchscreens, touch output mapping and sensor-driven orientation with orientation lockpartialcargo:sensor_strings_follow_mutter_51, cargo:policy_drives_unlocked_builtin_panels_only, cargo:sensor_poll_never_panics_without_a_bus, cargo:touch_points_invert_the_panel_to_logical, cargo:transform_point_matches_smithay_on_every_pixel, cargo:flipped_transforms_match_smithay_pixel_centers, cargo:render_transforms_follow_smithay_damage_tracker, cargo:rotated_scanout_damage_maps_into_buffer_pixels, cargo:display_rotation_validates_mutter_transforms, cargo:touch_press_moves_then_presses_left, cargo:touch_release_balances_with_left_releaseThe compositor polls iio-sensor-proxy (net.hadess.SensorProxy) with Mutter 51's claim discipline and sensor-to-transform mapping, rotates unlocked built-in panels (eDP/LVDS/DSI) live without rewriting monitors.xml, and serves PanelOrientationManaged from that policy (crates/compositor/src/orientation.rs, runtime.rs, mutter.rs); on DRM, touch and absolute-pointer input is remapped through the primary output's transform (drm.rs), and touch drives pointer emulation. Still missing: nothing publishes org.gnome.settings-daemon.peripherals.touchscreen orientation-lock (InputSettings.orientation_lock has no shell producer); org.gnome.desktop.peripherals.touchscreen output is ignored, so touch always maps to the primary output; there is no emulated touchscreen or accelerometer proof (ADR 0008 item 5) and no rotated-output runtime gate (#345)—
P-IN-03Compositor cursor themes, sizes, requested cursor shapes and locate-pointerpass (nested)proof:G-CURSOR-SIZE, proof:G-CURSOR-SHAPE, proof:G-CURSOR-FRAME, proof:G-LOCATE-POINTER, cargo:themes_resolve_through_their_inheritance_chain_then_default, cargo:missing_themes_fall_back_to_default_then_the_builtin_arrow, cargo:selected_legacy_theme_beats_bundled_svg_and_corrupt_files_fall_back, cargo:cursors_load_at_the_ceiled_scale_and_draw_at_the_theme_size, cargo:scalable_frames_preserve_nominal_geometry_hotspots_delays_and_argb, cargo:shapes_map_to_theme_names_mutter_uses, cargo:grabs_beat_client_shapes_and_no_focus_shows_the_default, cargo:cursor_animation_obeys_each_delay_and_wraps_at_exact_boundaries, cargo:a_lone_ctrl_tap_locates_the_pointer_only_when_enabled, cargo:configured_locate_keys_and_invalid_names, cargo:ripples_follow_gnome_timing, cargo:cursor_settings_survive_the_positional_control_wireThe compositor draws its own cursor from GNOME's live cursor-theme/cursor-size (inherited Xcursor themes and the bundled GNOME Shell 51 SVG theme, loaded at the ceiled scale), renders wp_cursor_shape_v1 shapes and client cursor surfaces, and plays locatePointer.js ripples on a lone locate-pointer-key tap (crates/compositor/src/cursor.rs, cursor_svg.rs, qualification); in CI run 38878636595 the nested GTK proof checks actual 24/48 px legacy and SVG pixels and hotspots, the crosshair shape, missing-shape fallback, a 32 px client surface, animated frames and callback-only commits, and locate-pointer on/off through compositor state; the same gates also passed on the installed .deb (tuna-deb-installed-proof, Ubuntu)no Marlin VM gate or recorded GNOME 51 capture covers this, so the row is nested only; installed DRM, fractional/multi-output cursor pixels and ripple pixels are unqualified (#342)
P-IN-04Mouse, touchpad, pointing-stick and trackball settings from GNOME Settingspartialproof:V-MOUSE-HANDEDNESS, proof:V-POINTER-LIVE, proof:V-POINTER-HOTPLUG, proof:G-SETTINGS-POINTER, cargo:drm::libinput::tests::pointer_policy_carries_the_mouse_keys, cargo:drm::libinput::tests::touchpad_policy_carries_the_touchpad_keys, cargo:drm::libinput::tests::trackball_policy_takes_mouse_motion_and_trackball_buttons, cargo:drm::libinput::tests::pointing_stick_policy_takes_its_own_speed_profile_and_scroll_method, cargo:drm::libinput::tests::later_mutter_classes_win_on_combined_capabilities, cargo:drm::libinput::tests::touchpad_scrolling_replays_mutters_two_finger_then_edge_sequence, cargo:drm::libinput::tests::scroll_buttons_map_like_meta_clutter_button_to_evdev, cargo:drm::libinput::tests::touchpad_enums_map_to_libinput_values, cargo:drm::libinput::tests::supported_profiles_and_unsupported_fallback_use_device_default, cargo:touchpad_handedness_overrides_or_tracks_mouse_changes, cargo:input_handedness_survives_the_positional_control_wire, cargo:pointer_acceleration_nicks_and_wire_roundtrip, cargo:device_setting_nicks_follow_gsettings_desktop_schemas_51, cargo:device_settings_default_to_gnome_51_and_survive_the_positional_wireEvery GNOME 51 touchpad, mouse, trackball and pointing-stick key Mutter 51 applies reaches libinput live and on hotplug, with Mutter's per-class rules and apply order (crates/compositor/src/drm/libinput.rs, docs/settings-desktop-inventory.md): handedness, natural scroll, speed (pointing sticks their own), per-class acceleration profile, touchpad tap-to-click, tap button map, tap-and-drag and its lock, disable-while-typing, send-events (libinput itself suspends a disabled-on-external-mouse touchpad while a mouse is plugged), click method and two-finger/edge scrolling (edge yields while any touchpad scrolls with two fingers; settings reapply when the last one leaves), middle-click emulation for mice, touchpads and trackballs, mouse/trackball scroll-wheel emulation and the pointing-stick scroll method. Each device reports its applied libinput state with per-option availability in input_devices. On the Marlin image built in CI run 38878636595 a live GNOME primary-button flip swaps left/right for a GTK client, mouse speed and natural scroll reach the libinput device and restore, and a hotplugged USB mouse arrives with the current settings (V-MOUSE-HANDEDNESS, V-POINTER-LIVE, V-POINTER-HOTPLUG; the published-image nightly 38844351544 runs V-MOUSE-HANDEDNESS only); V-POINTER-LIVE also checks the mouse acceleration profile (flat then adaptive), and on current builds additionally requires mouse middle-click emulation and scroll-wheel emulation (button 2 with lock, then off) in the live device readback, availability-gated (CI evidence for these follows this change); G-SETTINGS-POINTER proves the nested GSettings-to-compositor path for every new key. Gaps: the pinned libinput bindings (input-sys 1.19) cannot set custom acceleration curves (custom falls back to the device default), the disable-while-typing timeout (libinput's own timeout applies) or sticky drag lock (tap-and-drag-lock gets libinput's timeout drag lock; Mutter 51 asks for sticky); the snapshot reports each as unsupported. Touchpad, trackball and pointing-stick application is unit-tested only: the guest has only a USB mouse, so physical qualification remains (#343)—
P-IN-05Graphics tablets and GNOME Wacom settings (mapping, calibration, pressure, stylus and pad buttons)partialproof:D-TABLET, cargo:tablet_area_becomes_libinputs_calibration_matrix, cargo:keep_aspect_uses_the_monitor_ratio_only_for_absolute_mapping, cargo:automatic_external_tablets_span_the_desktop_and_a_set_output_picks_one, cargo:display_tablets_find_their_screen_by_name_or_size, cargo:edid_twins_are_told_apart_by_the_connector, cargo:rotated_monitors_rotate_the_tablet, cargo:monitors_cycle_through_the_span_unless_integrated, cargo:pressure_curves_scale_pressure_as_mutter_does, cargo:pressure_settings_are_validated_like_update_stylus_pressure, cargo:settings_paths_follow_mutters_lookup, cargo:pad_settings_paths_follow_mutter, cargo:per_device_paths_read_gnome_defaults_then_follow_live_changes, cargo:mismatched_schema_is_unavailable_and_keeps_the_defaults, cargo:accelerators_parse_like_mutter, cargo:keybindings_press_modifiers_first_and_release_them_last, cargo:pad_buttons_and_labels_follow_mutter, cargo:ring_and_strip_directions_follow_mutter, cargo:edition_settings_paths_follow_start_action_edition, cargo:edition_follows_up_on_the_next_direction, cargo:diagram_reads_labels_from_the_layout_like_init_labels, cargo:label_coords_follow_get_item_label_coords, cargo:element_ids_follow_pad_diagram_names, cargo:show_arguments_become_a_device_node_request, cargo:denied_requests_answer_invalid_params, cargo:tablet_osd_messages_decode_to_handled, cargo:pad_osd_commands_carry_fresh_request_idsEmulated (ADR 0008): zwp_tablet_manager_v2 with tools and pads, libinput tablet routing, libwacom identification, Mutter 51's org.gnome.desktop.peripherals.tablet/stylus settings (mapping, output, area, keep-aspect, pressure curve, stylus and pad button actions, keybindings, switch-monitor) and gnome-shell's pad OSD (org.gnome.Shell.Wacom.PadOsd) pass D-TABLET on vkms with a uinput Wacom Cintiq 22HD and a GTK stylus client, including relative mapping, hotplug and the locked session. Hardware-only and unproven: pad mode-switch groups (uinput cannot create the LED sysfs entries), pressure range and eraser-button settings (libinput 1.26/1.29), dials (libinput 1.26), left-handed (the 22HD is not reversible), cycling across several monitors (vkms has one output) and real tablets (#344)—
P-IN-06Keyboard input-source policy (per-window sources, MRU), remembered Num Lock and custom shortcutspartialproof:G-SETTINGS-INPUT, proof:G-A11Y-KEYMAP, cargo:input_sources_become_one_keymap, proof:G-INPUT-SOURCE-MRU, proof:G-INPUT-PER-WINDOW, proof:G-NUMLOCK, cargo:mru_seeds_from_settings_then_follows_activations_like_gnome, cargo:the_switcher_selects_the_second_most_recent_and_wraps, cargo:the_release_check_picks_once_the_bindings_modifier_is_up, cargo:per_window_restores_each_windows_source_and_starts_new_ones_first, cargo:steps_take_the_short_way_and_reloads_keep_the_layout, cargo:usable_sources_are_the_keymaps_xkb_layouts, cargo:names_come_from_xkeyboard_config_rules, cargo:keymap_reload_keeps_num_lock_and_the_active_layout, cargo:restore_and_save_follow_mutter, cargo:remembered_on_restores_then_changes_are_saved, cargo:nothing_remembered_restores_nothing_and_writes_nothing, cargo:a_toggle_before_the_restore_lands_winsInput sources, XKB options and repeat settings are live (G-SETTINGS-INPUT, G-A11Y-KEYMAP). The GTK shell runs gnome-shell 51.0's InputSourceManager (crates/shell-gtk/src/input_sources.rs): sources are kept most recent first, seeded from and saved to mru-sources; Super+Space opens GNOME's switcher popup (a quick tap flips to the second most recent source, held presses step, releasing the modifier picks — also when it is released before the popup holds the keyboard, which the popup checks each time it takes the keyboard; another shell popup or the lock screen steps by position); with per-window each window and the overview restore their own source and new windows start on the first (G-INPUT-SOURCE-MRU, G-INPUT-PER-WINDOW). The compositor keeps Num Lock, Caps Lock and the active layout across keymap reloads as mutter does (additive Smithay patch), restores numlock-state at startup while remember-numlock-state is set and saves every change (crates/compositor/src/numlock.rs; G-NUMLOCK covers saving and reloads, startup restore is unit-tested). Settings' custom shortcuts (org.gnome.settings-daemon.plugins.media-keys custom-keybindings) are run by gnome-settings-daemon 51's media-keys plugin, which the installed session starts and which grabs them through the shell's org.gnome.Shell GrabAccelerators as in GNOME; the GTK proof's G-GSD-LAUNCHER covers a gsd launcher key. Open: xkb-model needs an InputSettings.xkb_model wire field; IBus sources take no part in switching (the keymap holds XKB sources and the shell drives no IBus engine); the shell drives the compositor's layout with relative SwitchInputSource steps from its own record, so a shell restart or XKB group-toggle options can desynchronise it until the next source change; no panel input-source indicator; no gate presses a custom shortcut (#346)—
P-NT-03Per-app notification preferences and lock-screen notification privacypartialcargo:notification_policy::tests::app_policy_combines_global_and_app_gates, cargo:notification_policy::tests::disabled_apps_drop_even_critical_notifications, cargo:notification_policy::tests::sound_and_expansion_are_per_app, cargo:notification_policy::tests::lock_screen_counts_unseen_notifications_per_source_without_details, cargo:notification_policy::tests::lock_screen_details_are_per_app, cargo:notification_policy::tests::lock_screen_hides_sources_the_policy_keeps_off_it, cargo:notification_policy::tests::critical_notifications_get_no_lock_screen_exception, cargo:notification_policy::tests::lock_sources_stack_newest_started_first, cargo:notification_policy::tests::system_scoped_sources_show_details_without_the_app_key, cargo:notifications::tests::disabled_apps_are_hidden_and_every_app_registers, cargo:notifications::tests::disabling_an_app_hides_what_it_filed_and_enabling_restores_it, cargo:notifications::tests::an_update_from_an_enabled_app_brings_its_hidden_notification_back, cargo:notifications::tests::hidden_notifications_survive_a_restart_hidden, cargo:notifications::tests::per_app_banners_and_sounds_follow_settings_and_dnd, cargo:notifications::tests::lock_screen_follows_live_policy_and_acknowledgement, cargo:notifications::tests::app_identity_and_acknowledgement_survive_a_restart, cargo:intake::tests::hints_read_through_the_dbus_variant_layer, cargo:intake::tests::live_bus::notify_files_on_the_delivery_context_thread, cargo:sound_feedback::tests::notification_sounds_prefer_the_theme_name_like_gnome, cargo:lock_notifications::tests::detail_lines_escape_and_flatten_like_gnome, proof:G-NOTIFY-POLICY, proof:G-NOTIFY-LOCK, proof:G-NOTIFY-LOCK-PIXELS, proof:G-NOTIFY-LOCK-LIVE, proof:G-NOTIFY-SOUND, proof:G-NOTIFY-RESTOREPer-app enable, banners, sound and lock-screen keys plus application-children registration follow GNOME 51's NotificationApplicationPolicy (crates/shell-host/src/notification_policy.rs), read on the shell's GLib main loop and watched live: an app switched off hides what it filed at once and, switched back on, gets it all back in the date menu's list, seen or unseen as it was and with no banner (messageTray.js _onSourceEnableChanged; G-NOTIFY-RESTORE), and the lock screen re-reads its keys while locked. While locked no banner shows (sessionMode.hasNotifications); the lock screen draws GNOME's NotificationsBox inside the session-lock surface (crates/shell-gtk/src/lock_notifications.rs): per source with unseen notifications its name, icon and count, titles and bodies only with details-in-lock-screen or an all-x-gnome-privacy-scope=system source; critical ones get the critical style and nothing more. A redacted entry never carries a title or body, so neither reaches pixels or the accessible tree (G-NOTIFY-LOCK, G-NOTIFY-LOCK-PIXELS). Each banner plays its sound-name (else sound-file) through the shell's libcanberra event-sound player as it shows, by the live enable-sound-alerts and event-sounds (messageTray.js playSound; G-NOTIFY-SOUND); the bus advertises sound. Global show-banners and Do Not Disturb stay live (P-NT-02) (#348)Nested proofs only: installed-image, GNOME Settings panel writes, a paired GNOME 51 VM capture and audible hardware output are unmeasured. The lock screen lacks NotificationsBox's MPRIS media rows and its wake-up-screen on arrivals while the shield is blanked, scrolls a tall list below the clock instead of raising the clock, and shows bodies as plain text (as Tuna's banners do). force-expanded has no visible effect: the GTK banner always shows the whole body
P-A11Y-03Screen magnifier and on-screen keyboardpartialproof:M-OFF, proof:M-SETTINGS, proof:M-ZOOM, proof:M-TRACK, proof:M-FOCUS, proof:M-ATSPI-FOCUS, proof:M-LIVE, proof:M-CLICK, proof:M-KEYS, proof:M-CURSOR, proof:M-INVERT, proof:M-CROSSHAIRS, proof:M-HALF, proof:M-DISABLE, proof:M-LOCK, cargo:proportional_tracking_mirrors_gnome_math, cargo:push_tracking_scrolls_only_past_the_boundary, cargo:focus_tracking_follows_the_focused_window, cargo:caret_wins_when_the_pointer_rests, cargo:defaults_match_gnome_51, cargo:osk_layout::tests::every_vendored_layout_parses_except_ge, cargo:osk_layout::tests::ge_falls_back_to_us, cargo:osk_model::tests::typing_commits, cargo:osk_model::tests::shift_applies_once, cargo:osk_model::tests::long_press_shift_latches, cargo:osk_model::tests::symbol_levels, cargo:osk_model::tests::presence_opens_after_rest_time, cargo:osk_model::tests::presence_focus_out_hides_after_delay, cargo:osk::tests::the_lock_keyboard_only_edits_its_prompt, cargo:osk::tests::keysyms_pick_keys_and_levels_like_mutter, cargo:osk::tests::touch_mode_follows_mutter_and_the_last_device, cargo:osk_messages_roundtrip_and_sit_last, cargo:osk_model::tests::caps_lock_tap_unlatches_once_the_keys_are_redrawn, proof:G-OSK-SHOW, proof:G-OSK-TYPE, proof:G-OSK-SHIFT, proof:G-OSK-LOCKThe shell mirrors every org.gnome.desktop.a11y.magnifier key and binds Super+Alt+8 / Super+Alt+=/-; the compositor draws GNOME 51's ZoomRegion as a live render transform on every output frame (crates/compositor/src/magnifier.rs: zoom, tracking, cross-hairs, color effects, magnified pointer sprite, lock screen), captures and casts read back the same passes, and AT-SPI focus/caret events steer it (focus_caret.rs); proven live in scripts/tuna-magnifier-proof (nested). Accepted differences: touch mapped through the zoom, caret/window-centre fallback for non-AT-SPI apps, per-monitor regions instead of one region, shell popover focus untracked. Stays partial: color brightness/contrast/saturation pixels, live caret tracking and hardware multi-output are not yet proven (#349) On-screen keyboard (#349): the GTK shell ports gnome-shell 51.0 keyboard.js with GNOME's own osk-layouts JSON (levels, shift and its long-press latch, long-press extended keys, number/symbol levels, modifiers, emoji pager, language menu, hide key) on a layer surface that never takes the keyboard; it follows screen-keyboard-enabled, the compositor's touch-mode auto-show, the current input source and mutter-style input-method focus (text-input-v3 tracked by the compositor with or without IBus; a re-enable is the input-panel request, so a click into a GTK field opens it) and types through the compositor (OskInput, wire 0.43): commits to the focused text input, or keysyms via the active layout for X11 clients and modifiers. The lock screen embeds its own keyboard that edits only the unlock prompt; the compositor reports no focus and refuses OskInput while locked. scripts/tuna-osk-proof (nested, real pointer clicks on a GTK entry and on the keys the shell's AT-SPI tree names) proves the keyboard shows on a click into a GTK entry only with screen-keyboard-enabled, after the 300 ms rest time, stays across entries and hides after the 50 ms focus-out delay (G-OSK-SHOW), types into the entry as the client reads it (G-OSK-TYPE), shift and the caps latch (G-OSK-SHIFT), and on the lock screen edits only the unlock prompt with the session keyboard hidden and no text reaching session clients (G-OSK-LOCK). Stays partial: X11 typing is proven only at the key-path level (no nested XWayland proof); auto-capitalisation from surrounding text, IBus-routed OSK keys, word-wise delete, window shifting over the keyboard and the bottom-edge touch drag are not implemented.—
P-A11Y-04Keyboard input aids: Sticky, Slow, Bounce and Toggle Keys, Mouse Keys and the keyboard shortcuts that switch thempartialproof:G-A11Y-STICKY, proof:G-A11Y-SHORTCUTS, proof:G-A11Y-SLOW, proof:G-A11Y-BOUNCE, proof:G-A11Y-MOUSEKEYS, cargo:sticky_shift_latches_for_one_key_then_releases, cargo:sticky_second_tap_locks_and_third_unlocks, cargo:sticky_two_modifiers_together_latch_both_or_switch_the_aid_off, cargo:slow_keys_deliver_only_keys_held_past_the_delay, cargo:bounce_keys_drop_a_quick_repeat_of_the_same_key, cargo:mouse_keys_move_and_accelerate_until_release, cargo:shift_five_times_switches_sticky_keys_and_back, cargo:shift_held_eight_seconds_switches_slow_keysorg.gnome.desktop.a11y.keyboard is read live by the shell and applied in the compositor's input pipeline before keybindings, the lock screen and every Wayland or XWayland client (crates/compositor/src/input_aids.rs), following Mutter's keyboard accessibility: latches and locks, the two-modifier switch-off, slow-key acceptance, bounce rejection, Mouse Keys' acceleration curve and keypad buttons, the beeps, and Shift five times / held eight seconds with GNOME's confirmation dialog. Nested proofs drive real X key events into a recording GTK client. Not yet compared against a GNOME 51 capture or qualified on hardware; timeout-enable/disable-timeout are read by Mutter but never acted on, and are ignored here too (#350)—
P-A11Y-07Pointer input aids: simulated secondary click and hover (dwell) click with its click-type chooserpartialproof:G-A11Y-DWELL, proof:G-A11Y-SECONDARY, cargo:resting_pointer_clicks_once_after_the_dwell_time, cargo:moving_past_the_threshold_cancels_and_small_jitter_does_not, cargo:chooser_types_are_one_shot_and_drag_spans_two_dwells, cargo:gestures_pick_the_click_by_direction_and_click_where_it_dwelled, cargo:holding_primary_makes_a_secondary_click_on_release, cargo:short_presses_and_moves_cancel_the_secondary_clickorg.gnome.desktop.a11y.mouse is read live and applied in the compositor's input pipeline ahead of every client (crates/compositor/src/pointer_aids.rs), following Clutter's pointer accessibility: a 100 ms rest starts the dwell, the threshold cancels it, window mode clicks the type chosen in the panel's dwellClick chooser (one-shot types revert, drag spans two dwells), gesture mode picks the click by direction and clicks where the pointer dwelled, and a held primary button makes a secondary click on release. The shell draws GNOME's pie timer. Nested proofs drive real host pointer events into a recording GTK client; the chooser itself is unit-tested, not clicked by a proof. Not yet compared against a GNOME 51 capture or qualified on hardware; the pie timer and cursor follow the primary output only, and gesture mode does not switch to GNOME's crosshair cursor (#350)—
P-A11Y-05Reduced motion, keyboard focus visibility and the always-shown accessibility menupartialproof:G-ANIMATIONS-OFF, proof:G-A11Y-PREFS, proof:G-FOCUS-RING, proof:G-FOCUS-RING-TIMEOUT, proof:G-A11Y-MENU-RESTART, cargo:indicator_follows_gnome_enabled_or_requested_rule, cargo:focus_timeout_preserves_forever_and_toolkit_defaultMotion preferences, the live accessibility indicator/menu and shell GTK focus timeout now have consumers. In the nested shell, real keys light GNOME 51's panel focus fill (white at 17%, no ring) on exactly the focused control. The fill clears at the default 3 s, or at 2 s and 5 s when set, and never at 0. A menu row toggled with real keys persists across one deliberate shell restart, the only exit G-SHELL-STABLE allows. GNOME's St panel ignores the timeout, while Tuna's GTK panel applies it. The newer GTK property runtime, installed-image restart, external clients and paired GNOME qualification remain open (#351); see focused evidence and scope.—
P-A11Y-06Screen reader activation from Settings and complete spoken shell navigation with Orcapartialproof:V-ORCA-ACTIVATION, proof:V-ORCA-LOGOUT, proof:G-ORCA-SPEECH-TOGGLE, proof:G-ORCA-SPEECH-LEARN, proof:G-ORCA-SPEECH-PANEL, proof:G-ORCA-SPEECH-SEARCH, proof:G-ORCA-SPEECH-PREVIEWS, proof:G-ORCA-SPEECH-GRID, proof:G-ORCA-SPEECH-QS, proof:G-ORCA-SPEECH-SWITCHER, proof:G-ORCA-SPEECH-NOTIFY, proof:G-ORCA-SPEECH-LOCK, proof:G-ORCA-SPEECH-LOCK-DBUS, proof:G-ORCA-SPEECH-LOCK-SUPER-LAT-SPI trees match GNOME for selected surfaces (P-A11Y-01); on the Marlin image built in CI run 38878636595 the live screen-reader-enabled key starts and stops the distribution Orca under the compositor with verified bus owner, package hash and display, and hardware logout stops only the session's Orca (V-ORCA-ACTIVATION, V-ORCA-LOGOUT; V-ORCA-ACTIVATION also passed on the published image, nightly 38844351544). Spoken navigation (scripts/tuna-orca-speech-proof, CI lane orca-speech on Ubuntu 26.04 with Orca 50 and speech-dispatcher, in a nested session that owns the reader): real keys drive the shell and each gate asserts Orca's exact utterances from speech-dispatcher's log; the original ten gates passed with Orca 50.3 in CI run 41033267347 (#864). Super+Alt+S starts Orca with GNOME's KeyboardMonitor owned and stops it again; Orca's own keys pass through (learn mode); Ctrl+Alt+Tab reads Activities as a toggle button and the panel menus with GNOME's menu role (panelMenu.js); the overview reads its search entry and an app result, and Tab from the entry reaches the window previews by window title and the app grid by app name, as GNOME's focus chain does; quick settings, each Alt+Tab selection, a banner's arrival and its Super+N focus are read. The lock gate now covers D-Bus Lock after Escape and ordinary Super+L separately: each requires exactly one bullet per password character, no plaintext, and real PAM unlock. The owned trusted AT receives complete lock-screen KeyboardMonitor events and retains its grabs, matching Mutter 51.1; this does not grant any ordinary session client input. Privacy reasoning and immutable GNOME source links are in docs/accessibility-shell.md (#936). Per-character speech acknowledgment removes the fixed-delay cancellation race in the search typing gate (run 41148163143), retaining the ordered echo and exact bullet-count assertions. Partial: the utterances are checked against GNOME 51's source (panelMenu.js, windowPreview.js, appDisplay.js, messageList.js), not a recorded GNOME 51 Orca session; Ctrl+Alt+Tab has no CtrlAltTabPopup group switcher (held modifiers, the overview's Search, Windows/Apps and Dash groups); a notification message (banner or listed) is a GTK alert where GNOME's is an ATK notification; the dash, calendar and volume OSD (silent in GNOME too, osdWindow.js) are not gated (#352)—
P-SN-01Sound preferences: over-amplification, feedback sounds, visual and audible bellpartialproof:G-MEDIA-AMPLIFICATION, proof:V-AUDIO-AMPLIFICATION, proof:G-MEDIA-VOLUME, proof:V-MEDIA-VOLUME, proof:G-AUDIO-HELPER-EXIT, proof:G-GSD-MIC-MUTE, proof:G-GSD-VOLUME-ONCE, cargo:media_keys::tests::amplified_output_obeys_live_limit_and_preserves_unmute_level, cargo:media_keys::tests::unmute_cannot_bypass_disabled_amplification, cargo:logic::service_tests::amplified_slider_ceiling_matches_gnome_without_rounding_above_it, cargo:audio_state::tests::amplified_backend_level_is_not_silently_truncated, cargo:sound_feedback::tests::feedback_requires_enabled_changed_audible_idle_sink, cargo:audio_state::tests::feedback_player_requires_running_stream_on_default_sink, cargo🔔:tests::first_bell_is_silent_until_trusted_policy_arrives, cargo🔔:tests::disabling_visual_policy_cancels_inflight_flash, cargo🔔:tests::global_rate_limit_and_double_pulse_threshold, cargo🔔:tests::pulse_opacity_is_bounded_and_expires, cargo:the_system_bell_rings, cargo:runtime::scene_slot_tests::frame_bell_stays_under_higher_windows_and_only_in_its_slice, cargo:tests::live_bell_policy_survives_the_positional_control_wire, cargo:gsd_media_keys_owns_every_media_schema_key_and_nothing_elseIn the installed session the volume and mic-mute keys are gnome-settings-daemon 51's own: gsd-media-keys (started by tuna-session.target) grabs them through org.gnome.Shell and applies volume-step, the allow-volume-above-100-percent ceiling, its feedback sound and the OSD through ShowOSD, while the shell binds none of the media-keys schema's keys (crates/shell-gtk/src/keybindings.rs); G-GSD-MIC-MUTE and G-GSD-VOLUME-ONCE run the distribution gsd-media-keys against the nested shell (mic-mute with its OSD; one key press, one step). Without gsd-media-keys (the nested preview) the shell's native keys follow allow-volume-above-100-percent live up to GNOME's +11 dB ceiling like the slider, and disabling it clamps an already amplified level (crates/shell-gtk/src/media_keys.rs, services.rs); G-MEDIA-AMPLIFICATION proves it in the nested shell and V-AUDIO-AMPLIFICATION on the real HDA/PipeWire sink of the Marlin image (CI run 38878636595 and the published image, nightly 38844351544); volume-change feedback follows event-sounds/theme-name and GNOME 51's own-playback rule (sound_feedback.rs); the live audible-bell/visual-bell/visual-bell-type policy reaches the compositor's xdg-system-bell consumer, which uses Mutter 51's flash timings (crates/compositor/src/bell.rs). Feedback sounds and the bell are proven only by unit tests in CI: the delivered nine-case audio proof (native-sound-proof.md) runs outside CI. Still missing: a visual-bell pixel proof, the Xwayland bell, an input-feedback-sounds consumer, GNOME's drag-end slider feedback and physical-device qualification; the GTK proof's gsd-media-keys is the runner distribution's, not 51.0 (#353)—
P-ST-02Wellbeing: screen-time history and limits, grayscale and break reminderspartialproof:G-WELLBEING-HISTORY, proof:G-WELLBEING-LIMIT, proof:G-WELLBEING-BREAK, proof:G-WELLBEING-PIXEL-GRAY, proof:G-WELLBEING-PIXEL-FADE, proof:G-WELLBEING-PIXEL-BOTH, proof:G-WELLBEING-PIXEL-RESET, proof:G-WELLBEING-ACTION-TAKE-LOCK, proof:G-WELLBEING-ACTION-TAKE-FADE, proof:G-WELLBEING-ACTION-DELAY, proof:G-WELLBEING-ACTION-SKIP, proof:G-WELLBEING-NEGATIVE, cargo:history_round_trip_matches_gnome_field_names, cargo:disabling_history_deletes_the_file_and_stops_recording, cargo:limit_warns_at_exactly_ten_minutes_then_reaches_then_clears_next_day, cargo:limit_at_or_under_ten_minutes_gets_no_warning, cargo:turning_the_limit_off_removes_the_notice, cargo:dispatch_posts_limit_notices_with_gnome_strings_and_tracks_grayscale, cargo:break_notices_update_the_resident_in_place, cargo:scripted_day_warns_reaches_and_resets, cargo:break_upcoming_countdown_and_take_cycle, cargo:break_flows_from_due_through_overdue, cargo:delay_and_skip_push_the_due_time, cargo:idle_break_fades_completes_and_sounds, cargo:break_service_public_flow, cargo:entry_exit_and_motion_policy_match_effects, cargo:wellbeing_effects_are_atomic_bounded_and_strictly_decodedGTK main runs the wellbeing consumers (crates/shell-gtk/src/wellbeing.rs): the GNOME-format $XDG_DATA_HOME/gnome-shell/session-active-history.json, the daily-limit warning/reached as one resident Urgency.HIGH notification updated in place (GNOME's TimeLimitsNotificationSource), the warning posted only at exactly 600 s left (so a limit of 10 min or less gets none) and the notification removed when the limit is turned off, history is disabled or over 10 min remain again (day rollover); eyesight and movement scheduling with a countdown OSD, upcoming/due/overdue/in-progress notices as one resident notification updated in place with Take/Delay/Skip (GNOME's BreakNotificationSource), Urgency.HIGH with the break type's notify toggle on and LOW (list only, no banner) with it off, removed when the break completes or reminders are disabled; lock-screen and completion sound; grayscale and fade reach the compositor's wellbeing_effects through protocol 0.35 (3 s ease-out-quad entry, immediate exit, per GNOME 51 timeLimitsManager.js/breakManager.js; effects); in CI run 38878636595 the nested GTK proof passed the history, limit, break, gray/fade/combined/reset pixel, Take/Delay/Skip action and negative gates, and the native-wellbeing-proof and real-GTK countdown OSD steps were green (key dispositions)movement breaks lack end-to-end coverage; paired GNOME 51 captures and an installed Marlin/DRM gate are missing (#354)
P-ST-03Power panel: GNOME 51 brightness interface, power saving, idle dim/blank and power policiespartialproof:G-QS-BRIGHTNESS, proof:G-KEY-BRIGHTNESS, proof:G-MONITOR-BRIGHTNESS-KEYS, proof:G-BRIGHTNESS-CONTRACT, proof:G-POWER-SAVE-MODE, proof:G-QS-POWER, proof:G-SETTINGS-IDLE, proof:G-IDLE-FADE, proof:V-GSD-PLUGINS, proof:V-GSD-POWER-IDLE, proof:V-POWER-BUTTON, cargo:power_save_mode_round_trips_over_a_private_bus, cargo:gsd_blank_and_unblank_turn_outputs_off_and_on, cargo:standby_and_suspend_disable_outputs_like_off, cargo:unknown_is_never_entered_or_left, cargo:values_outside_mutters_switch_stop_scanout_without_disabling, cargo:the_session_target_starts_gnome_51s_plugins_except_those_tuna_owns, cargo:the_leader_fifo_hangs_up_for_the_monitor_when_the_session_ends, cargo:bus_callers_inhibit_and_their_departure_releases_the_idle_lock, cargo:inhibit_validates_like_gnome_session_and_unions_flags, cargo:gsd_power_key_shutdown_opens_the_shells_dialog_and_follows_its_answerShell backlight control through logind works from the slider and keys (G-QS-BRIGHTNESS, G-KEY-BRIGHTNESS, G-MONITOR-BRIGHTNESS-KEYS); the shell exports org.gnome.Shell.Brightness with GNOME 51's surface and dim/auto arithmetic against a controlled sysfs/logind fixture (G-BRIGHTNESS-CONTRACT, docs/brightness-contract.md); DisplayConfig PowerSaveMode is read-write with Mutter 51 semantics (crates/compositor/src/power_save.rs): 3/2/1 stop scanout and turn every CRTC off on hardware, 0 repaints and modesets back, -1 is ignored, frames complete on Mutter's 100 ms dummy flip and orientation tracking is inhibited; G-POWER-SAVE-MODE drives gsd 51's dim → blank → unblank → undim sequence nested (scanout stops, repaint and exact backlight baseline return). The Power Mode tile switches profiles (G-QS-POWER) and idle-delay fades and blanks before locking (G-SETTINGS-IDLE, G-IDLE-FADE). The installed session starts gnome-settings-daemon 51's power plugin with GNOME's own units (share/systemd/user/tuna-session.target, plugin table in settings-map.md, V-GSD-PLUGINS); it reads the compositor's IdleMonitor, the InhibitedActions and SessionIsActive of the compositor's org.gnome.SessionManager (crates/compositor/src/session_manager.rs) and the shell's ScreenSaver. V-GSD-POWER-IDLE observes, in the Marlin VM, gsd-power's own SetDimming(true), the idle lock, PowerSaveMode=3 in the compositor snapshot, and the wake to 0 with undimming. The session also starts gsd 51's media-keys plugin, which holds logind's power, suspend and hibernate key inhibitor and applies power-button-action: interactive opens the shell's end-session dialog through the compositor's org.gnome.SessionManager.Shutdown, suspend/hibernate go through logind, nothing does nothing, so logind's HandlePowerKey never powers the machine off; V-POWER-BUTTON checks the inhibitor and drives the ACPI power button from QMP in the Marlin VM with the hostnamed chassis set to a desktop for the fixture (gsd 51 powers off a vm chassis for every value but nothing). Open: physical panel dimming and the CRTC disable on real hardware; automatic suspend after sleep-inactive-*-timeout (gsd-power never suspends a VM), battery/AC, lid and ambient-sensor policies; physical power, suspend and hibernate keys; the brightness adapter follows idle-brightness live and fails positive operations without hardware where GNOME 51 reads the key once and accepts them; the unlock hint does not hide on power-save changes (#355)—
P-ST-04Privacy device controls: camera, microphone and sound-output disable, privacy screen and USB protectionpartialproof:G-PRIVACY-PORTAL, cargo:privacy::tests::live_pw::live_native_executor_mutes_restores_and_refuses_wrong_identity, cargo:lockdown::tests::live_bus::real_portal_denies_camera_while_disabled_and_readmits_on_restore, cargo:lockdown::tests::poll_syncs_all_three_switches_and_keeps_absent, cargo:privacy::tests::live_pw::live_enforcer_mutes_bypass_and_restores_devices, cargo:privacy::tests::live_pw::live_enforcer_destroys_cameras_existing_and_new, cargo:audio_state::tests::privacy_scan_destroys_cameras_but_not_screen_shares, cargo:audio_state::tests::privacy_scan_requires_mute_on_every_unmuted_device, cargo:privacy::tests::plan_restore_unmutes_only_tracked_nodes, cargo:privacy::tests::restore_cycle::disable_beyond_per_tick_cap_restores_every_node, cargo:privacy_screen::tests::the_setting_switches_unlocked_screens_without_writing_itself_back, cargo:privacy_screen::tests::the_hotkey_writes_the_setting_back, cargo:privacy_screen::tests::locked_and_absent_screens_are_never_written, cargo:mutter::tests::current_state_reports_privacy_screens_as_mutter_doesThe GTK shell publishes disable-camera, disable-microphone and disable-sound-output as the standard Lockdown portal backend (crates/shell-host/src/lockdown.rs, started by crates/shell-gtk/src/privacy_portal.rs), and the shipped packaging/marlin/tuna-portals.conf routes Lockdown to it, so xdg-desktop-portal denies new camera requests as GNOME 51 does; G-PRIVACY-PORTAL verifies this against the real portal in the nested proof and against the installed .deb (tuna-deb-installed-proof, CI run 38878636595). On the PipeWire graph the shell also mutes every source or sink while microphone or sound output is disabled, re-mutes bypasses and restores only what it muted, and destroys v4l2/libcamera camera nodes and their linked capture streams while leaving screen shares alone (crates/shell-gtk/src/privacy.rs, services.rs, privacy-enforcement.md); these are deliberate extensions where GNOME has no enforcer, and the CI-required native-executor proof covers source mute/restore and wrong-identity refusal. usb-protection and usb-protection-level are read by gnome-settings-daemon's usb-protection plugin, which the installed session now starts (V-GSD-PLUGINS); it acts only through a running USBGuard daemon and is unqualified. privacy-screen is consumed by the compositor's DRM backend as mutter 51 does: KMS privacy-screen sw-state is switched on unlocked connectors, hw-state changes (the hotkey) write the setting back, and GetCurrentState reports privacy-screen-state (bb) (crates/compositor/src/privacy_screen.rs); unit-tested against a fake connector, hardware privacy screens unmeasured. Corral on the published image marlin@sha256:ae6363c1… (2026-10-30, evidence): it ships tuna-desktop 1.0.0beta.1, which predates this enforcement; Settings' camera switch (real clicks) denies and readmits a real portal client through xdg-desktop-portal-gnome's Lockdown, microphone/sound output are unenforced, no gsd plugin runs (Settings' Empty Trash does nothing), and USBGuard is not installed. With main a3308583 sideloaded, Tuna's Lockdown backend serves the portal and the shell mutes/unmutes on the keys; the camera destroy path, restart persistence, USB protection at lock, purge ages, camera hardware and unclassified virtual camera providers remain open (#356)—
P-ST-05Administrative lockdown (org.gnome.desktop.lockdown): disable command line, log out, user switching, lock screen and otherspartialcargo:each_key_denies_exactly_its_action, cargo:policy_unlock_restores_allowed_actions, cargo:absent_keys_keep_current_values_never_reset, cargo:command_booleans_parse_and_hung_backend_times_out, cargo:inherited_writer_cannot_hold_the_post_exit_read_past_the_deadline, cargo:lock_command_denied_under_lockdown_policy, cargo:lockdown_policy_lift_restores_lock, cargo:hub_lockdown_policy_reaches_sessions_and_restores, cargo:live_lock_screen_key_reaches_sessions_and_keeps_the_other_keys, cargo:a_change_of_the_key_arrives_as_an_event_and_an_unchanged_key_costs_no_wake, cargo:of_several_changes_between_two_ticks_the_newest_wins, cargo:without_settings_nothing_is_published, cargo:a_stalled_settings_backend_never_delays_the_caller, cargo:each_shell_key_hides_exactly_its_entry_point, cargo:full_lockdown_hides_every_shell_entry_point, cargo:keys_without_a_gnome_shell_consumer_touch_no_shell_widget, cargo:policy_round_trips_through_memory_backend, cargo:widgets_apply_entry_visibility, cargo:disable_log_out_denies_every_session_ending_action_but_suspend, cargo:lockdown_hides_nested_menu_but_keeps_hardware_menu, cargo:save_to_disk_lockdown_refuses_only_the_portal, cargo:password_peek_follows_disable_show_password_live, cargo:history_walks_and_keeps_the_typed_line_like_gnome, cargo:completion_takes_the_last_word_and_the_common_prefix, cargo:failed_lines_fall_back_to_files_like_gnome, cargo:spawn_errors_read_as_gnome_shows_them, cargo:gnome_51_defaults_without_the_schema, proof:G-LOCKDOWN-LOG-OUT, proof:G-LOCKDOWN-LOCK, proof:G-LOCKDOWN-COMMAND-LINE, proof:G-LOCKDOWN-SAVE-TO-DISKEach key's consumer follows gnome-shell 51.0, gnome-session 51.0 and gnome-settings-daemon 51.0 (crates/shell-gtk/src/lockdown.rs, crates/compositor/src/lockdown.rs). disable-lock-screen: the shell hides Lock and refuses its button, Super+L, ScreenSaver.Lock, sleep and idle locks (screenShield.js lock()), and the compositor backstop refuses the Lock command, idle lock and the shell's session-lock request from its own live watch of the key (crates/compositor/src/lockdown_settings.rs); a locked session never unlocks from policy (G-LOCKDOWN-LOCK). disable-log-out: Log Out…, Restart… and Power Off… leave the menu and are refused at open and fire, as gnome-session answers CanShutdown/CanReboot unavailable and refuses Logout; Suspend stays (G-LOCKDOWN-LOG-OUT). disable-command-line: Alt+F2 opens GNOME's Run a Command dialog (crates/shell-gtk/src/run_dialog.rs: shell parsing on PATH, Ctrl+Return in the configured terminal, file fallback, Tab completion, command-history), refused per open while set (G-LOCKDOWN-COMMAND-LINE). disable-show-password: unlock prompt, polkit and network secret entries lose the peek icon and GTK's Show Text live and re-hide revealed text (StPasswordEntry). disable-save-to-disk: shell screenshots stay on the clipboard, screencasts refuse to start and the portal's org.gnome.Shell.Screenshot calls are refused (G-LOCKDOWN-SAVE-TO-DISK covers the screenshot key). disable-application-handlers has no GNOME 51 desktop consumer, so the shell no longer gates Settings/calendar launches on it. Unqualified: Switch User (Tuna ships none, so disable-user-switching gates nothing), the live portal refusal and screencast refusal (unit-tested only), the display-backed peek-icon test (CI xvfb), GNOME Shell's developer run commands (lg, rt, debugexit), and an administrator-locked dconf system database on an installed image (#357)org.gnome.ScreenSaver.Lock under disable-lock-screen fails at once with an administrator-policy error, where gnome-shell 51.0's LockAsync never replies (it waits for a lock that lock() refuses), so callers do not hang to their D-Bus timeout; an unauthenticated org.gnome.Shell.Screenshot caller gets AccessDenied before the lockdown check, where GNOME checks lockdown first. Owner #357
P-ST-06Remote desktop: GNOME Settings Desktop Sharing, Remote Control and remote login over RDPpartialcargo:only_the_installed_remote_desktop_daemon_may_create_sessions, cargo:session_clipboard_and_keysyms_are_owned_bounded_and_revocable, cargo:remote_desktop_daemon_identity_is_the_root_owned_installed_binary, cargo:linked_screencast_records_after_remote_start_and_serves_stream_start_stop, cargo:a_plain_session_streams_after_start_and_ends_with_its_stream, cargo:a_remote_grant_links_one_screencast_session_of_its_own_owner, cargo:linked_monitor_streams_carry_the_eis_mapping_id_grd_requires, cargo:an_empty_connector_records_the_primary_monitor_like_mutter, cargo:pump_copies_exactly_and_stops_on_limit_revocation_idle_and_closeAuthenticated portal keyboard and pointer input work through the genuine GNOME 51 RemoteDesktop portal, including libei delivery, FD revocation and lock revocation (docs/remote-desktop.md; remote-security job, CI run 38878636595); gnome-settings-daemon's sharing plugin runs in the installed session (V-GSD-PLUGINS, plugin table). The installed gnome-remote-desktop daemon is admitted by executable identity plus grd bus name; NotifyKeyboardKeysym, mutter's session clipboard, RecordMonitor "", the stream mapping-id, Record* after Start and ScreenCast Stream.Start/Stream.Stop follow mutter 51 (remote-desktop.md). Corral VM 2026-10-30 (marlin@sha256:ae6363c1…, grd 51.0 and the candidate compositor and shell sideloaded, real FreeRDP 3 client; evidence): the client shows the desktop frame; view-only refuses its input (also when control is switched off mid-session); Remote Control drives Settings by pointer and keyboard; clipboard text crosses both ways; disconnect/reconnect, lock (disconnects and refuses until unlock, as GNOME's remote-access inhibit does), Desktop Sharing off in Settings and grd restart all end the session cleanly and reconnect afterwards. The published image ships no gnome-remote-desktop, so Settings has no Remote Desktop row there; RecordVirtual (extend mode and remote login), cursor metadata, touch, smartcard/Kerberos and physical hardware remain (#358)—
P-ST-07Date & Time: calendar preferences, automatic time and time zone on the shipped imagepartialproof:G-CAL-PREFS, proof:G-DATETIME-NTP, proof:G-DATETIME-TZ, proof:G-DATETIME-TZ-REJECT, proof:G-DATETIME-RTC, proof:G-DATETIME-CLOCK, proof:G-DATETIME-CLOCK-TZ, proof:G-DATETIME-MANUAL, proof:G-DATETIME-CLOCK-SET, proof:G-DATETIME-POLKIT, proof:G-DATETIME-NTP-ABSENT, cargo:a_time_zone_change_redraws_the_clock_at_once, cargo:the_zone_is_the_one_localtime_names_nowWeek numbers and first weekday update the calendar live; timedate1 time-zone round trip and rejection, NTP, manual time (SetNTP/SetTime as the Settings panel does them, restored), the GNOME 51 polkit rule for the desktop user (admin allowed, non-admin refused), missing-NTP-provider behaviour and the panel clock following zone and clock changes are gated in the Marlin VM lane (datetime-proof.md); on the published image (Corral, marlin@sha256:ae6363c1, 2026-10-30) every gate but the clock following a zone change passed, the Settings panel changed zone and manual time with real input and the zone survived a reboot. Remaining: the shell's polkit agent cannot authenticate with polkit 128's non-setuid helper, so a non-admin cannot unlock the panel (#884); GeoClue automatic time zone awaits actual current-build service/UI qualification by the standalone official-static-source observer (the historical VM had no supplied usable source; this is not a GPS requirement); RTC hardware accuracy and hardware evidence are unmeasured (#359)—
P-ST-08Background placement modes, shading and slideshow (dynamic XML) wallpaperspartialcargo:static_modes_reveal_distinct_real_image_regions, cargo:gradients_and_transparency_compose_under_none_and_centered, cargo:spanned_maps_real_global_union_instead_of_repeating_each_monitor, cargo:tiles_share_global_phase_across_monitor_origins, cargo:static_metadata_and_file_replacement_change_all_render_keys, cargo:slideshow::tests::wall_clock_cycles_boundaries_and_negative_delta, cargo:slideshow::tests::variants_prefer_large_aspect_then_width_and_reverse_ties, cargo:slideshow::tests::bounded_invalid_timeline_does_not_expand_entities_or_accept_nan, cargo:xml_reference_replacement_changes_epoch_and_rejects_stale_worker, cargo:blend_places_each_image_before_transparent_underlay, cargo:slideshow_sample_due_on_interval_resume_and_clock_step_only, cargo:slideshow_refresh_resamples_after_suspend_and_clock_steps, cargo:xml_observation_follows_a_live_timezone_change, cargo:static_cache_persists_raster_decodes_but_never_failed_fallbacks, cargo:no_picture_desktop_and_lock_do_not_schedule_ignored_source_observations, cargo:missing_unreadable_and_undecodable_pictures_fall_back_to_the_colorThe GTK shell publishes live placement, shading and both colours; the compositor renders all seven placement modes, gradients, transparency and global monitor layout, samples GNOME XML slideshows on the wall clock with bounded worker parsing and premultiplied crossfades, re-samples at once after suspend or a wall-clock step and re-reads the local start time on a zone change (crates/compositor/src/slideshow.rs, wallpaper.rs), and renders SVG/SVGZ through the bounded tuna-wallpaper-svg Glycin stream helper (#458); a no-picture lock never observes its ignored URI. The stock GNOME 51 chooser drives six placements, two gradients and light/dark XML slideshows matching GnomeBG's reference (portal-security stock-backgrounds and stock-slideshow, CI run 38878636595; dynamic, static) and SVG selections (svg, CI run 39317849236). Corral VM 2026-10-30 on published Marlin sha256:ae6363c1… (ships tuna-desktop 1.0.0-beta.1): stock Settings selections of a stock picture and a light/dark pair pass on desktop pixels; XML and SVG fail there because beta.1 predates both. The same image with main a3308583 sideloaded passes stock-Settings XML, Add Picture… SVG, same-URI SVG/XML edits, malformed/oversized SVG fallback, lock pixels, real s2idle resume, clock steps, zone changes and restart persistence; it also showed beta.1 persisting colour fallbacks of failed decodes, which main then read instead of the SVG (fixed by static cache version 6, which persists only actual raster decodes). Open: the image ships no gnome-backgrounds, no shipped image carries the SVG/XML code yet, and physical multi-output/scale and paired GNOME captures are missing (#360, #452)—
P-ST-09Apps panel: permissions, background apps, global shortcuts, default handlers and removable-media autorunpartialproof:G-SHORTCUT-CONSENT, cargo:shortcut_consent_cannot_survive_a_lock_or_focus_change, cargo:buttons_escape_and_close_map_to_gnome_responses, cargo:missing_or_malformed_store_data_asks_and_never_grants, cargo:each_request_is_answered_once_and_superseded_ones_never, cargo:only_remembered_answers_for_known_apps_are_saved, cargo:automount_mounts_a_hotplugged_volume_with_the_mount_operation, cargo:automount_refuses_inactive_inhibited_mounted_and_locked, cargo:unlock_mounts_waiting_volumes_without_autorun, cargo:autorun_follows_the_first_content_types_setting, cargo:autorun_never_and_unallowed_mounts_do_nothing, cargo:wrong_passphrases_ask_again_other_failures_do_not, cargo:sniffer_counts_a_real_directory_tree, proof:V-INTROSPECT-SANDBOXShortcut-inhibit requests need consent and the decision persists per app in PermissionStore (G-SHORTCUT-CONSENT). The Background portal reads a sandboxed Flatpak app's state through Introspect on a Marlin VM (V-INTROSPECT-SANDBOX). The GTK shell now carries GNOME Shell 51's automount/autorun components (media-handling consumer, passphrase prompt, in-process hotplug sniffer; unit-tested with fake volumes). Corral VM on the published Marlin image (2026-10-30): shipped build leaves a hotplugged volume unmounted (no consumer, no org.gnome.SessionManager in that build); the Background portal honours PermissionStore yes/no; GlobalShortcuts CreateSession/BindShortcuts/ListShortcuts work through GNOME Settings' provider; the shipped (2026-10-15) build missed the chord's Activated, while this candidate sideloaded on the same image delivers Activated and Deactivated to a sandboxed client through xdg-desktop-portal-gnome's GrabAccelerators path. The candidate sideloaded on the published image (bootc usroverlay) mounts a hotplugged removable volume and posts the "connected" notification whose Open with Files action opens it, mounts silently under autorun-never, leaves the volume unmounted while an automount inhibitor is held, defers a lock-time plug to unlock without autorun, and re-asks a wrong LUKS2 passphrase in the same dialog; Settings > Default Apps switching changes xdg-mime and the shell's open path follows it. GNOME Settings' per-app list is empty on the image (libmalcontent without the malcontent package, a tunaOS image packaging gap), so per-app permission toggles are unqualified. Remaining: per-app permissions through Settings, org.gtk.MountOperationHandler, the candidate in a published image, and physical USB/optical media (#361)—
P-ST-10Service-backed Settings panels (network, Bluetooth, online accounts, printers, sharing, users, region, updates) on the shipped imagepartialproof:G-QS-PRINTERS, proof:G-QS-PRINTERS-ABSENT, cargo:client_lists_printers_over_http, cargo:client_lists_jobs_and_cancels, cargo:tile_subtitle_names_queue_printer_or_absence, cargo:row_label_names_printer_status_and_jobs, proof:G-QS-WWAN, proof:G-QS-WWAN-ABSENT, proof:G-QS-LOCATION, proof:G-QS-LOCATION-ABSENTPrinters: the quick-settings Printers tile and menu list the live CUPS queue over IPP, and the queue window shows GNOME's printer rows, jobs and actions, with Cancel and Test Page reaching the print server; with no print server the tile and menu say No Print Service (crates/shell-gtk/src/printers.rs, nested proof against a CUPS stub). Mobile broadband and location services run against live service shapes in the nested proof: the Mobile Network menu reads modems through NetworkManager Udi and ModemManager Modem3gpp (operator, signal, APN) and its tile flips WwanEnabled; the shell's GeoClue2 agent registers, InUse drives the panel indicator, and AuthorizeApp answers from the permission store under the Settings-owned switch and max-accuracy-level ceiling with first-use consent; without ModemManager or GeoClue the tile and indicator hide and grants stay gated (crates/shell-gtk/src/wwan.rs, location.rs). Network, Bluetooth, online accounts, sharing, users, region and updates panels on Marlin, printer discovery/add through cups-pk-helper, physical modem proof and GeoClue agent whitelisting/packaging on TunaOS remain unqualified (#362)—
P-ST-11Legacy and app-owned GNOME settings are classified without false parity claimspartialEvery one of the 428 desktop keys names its GNOME 51 enforcing owner in the inventory: 253 implemented (each citing a gate or cargo test that exists), 26 toolkit, 38 app-owned, 51 deprecated, 60 unimplemented (each linking an issue); scripts/tuna-settings-inventory-check and scripts/lib/settings-inventory-check-tests.py enforce categories, citations, id existence and that keys gnome-control-center 51.0 writes carry an observable gate or an open issue. Stale Ignored rows corrected (locate-pointer, switch-panels, bells, per-app notifications, tablets, background gradients, document/monospace/font-rendering). Key-count coverage is not behaviour coverage: 32 GNOME-Settings-exposed implemented keys still have only consumer tests (listed as observable proof pending), and 60 keys are unimplemented (#365)—
P-LK-02Authentication beyond one password: fingerprint, smartcard/PIN, multi-step and web login on the lock screenpartialcargo:actual_main_runtime_pam_conversation_relay, cargo:three_failed_scans_end_failed_like_pam_fprintd_max_tries, cargo:a_retry_then_match_reports_pam_fprintd_messages, cargo:timeouts_and_daemon_failures_end_unavailable_never_matched, cargo:interactive_attempt_reports_armed_and_retry_then_matches, cargo:interactive_cancel_ends_the_attempt_and_frees_the_claim, cargo:auth_conversation_is_appended_and_round_trips, cargo:answer_frames_match_the_generic_encoding_and_never_debug_the_answer, proof:G-LOCK-AUTH-MULTISTEP, proof:G-LOCK-AUTH-WRONG-CODE, proof:G-LOCK-AUTH-CANCEL, proof:G-LOCK-AUTH-FPRINT, proof:G-LOCK-AUTH-FPRINT-MAXTRIES, proof:G-LOCK-AUTH-NO-PRINTSThe lock screen runs GNOME 51's AuthPrompt flow (crates/shell-gtk/src/lock.rs, lock_auth.rs) over shell protocol 0.44 against a GDM-UserVerifier-shaped compositor consumer (crates/compositor/src/runtime/auth.rs, lock-conversations.md): every PAM prompt of the tuna-lock stack is relayed one question at a time (password, then a code or PIN, with Info/Error messages shown and announced, Back from step 2 soft-resets), and fingerprint runs beside it as pam_fprintd does for gdm-fingerprint (fprintd D-Bus, 3 scans, GNOME's hint and messages). Nested-only and emulated (ADR 0008): the controlled multi-prompt PAM module in the actual runtime, and the GTK proof scripts/tuna-lock-auth-proof with pam_wrapper (pam_matrix + one-time code) and a mock fprintd; no reader hardware, no installed published-image stack. Smartcard/PIN-pad (pam_pkcs11/pam_sss, GNOME's smartcard mechanism and auth menu) and web login (GDM eidp) have no code; greetd sessions keep the single-password Unlock (#366)—
P-LK-03The login screen is Tuna Desktop's own greeter on GDM's daemon, matching GDM 51's login screen: user list, password, fingerprint, smartcard and multi-step auth, accessibility and input sources before login, user switching and GDM policy keysmissingAdded to v1 by the owner on 2026-11-16 (ADR 0008 item 6, epic #1164). GNOME 51: gdm's greeter is gnome-shell --mode=gdm (js/gdm/loginDialog.js, authPrompt.js, util.js over libgdm's GdmClient/UserVerifier/Greeter), started as org.gnome.Shell@gdm.service in gnome-session@gnome-login.target. Every image still logs in through that GNOME Shell greeter (V-GREETER-LIST, V-GDM-LOGIN). The lock screen's AuthPrompt flow (P-LK-02) is the starting point for the greeter's authentication UI—
P-AN-01Motion preferences: animations off is instant, reduced motion keeps fades only, slow-down factor scales durationspartialproof:G-ANIMATIONS-OFF, proof:G-INTROSPECT-MOTION, proof:G-ANIMATION-CLOCK, proof:G-TRANSIENT-OFF, cargo:transient::tests::the_motion_policy_gates_fades_and_motion, cargo:transient::tests::reduced_motion_keeps_only_the_fadeA shared full / fade-only / off policy with the slow-down factor drives shell and compositor animations, shell transients included (G-TRANSIENT-OFF), and a deterministic animation clock samples them at fixed delays (#523); paired GNOME 51 captures at the same delays are still missing (#493, epic #492)—
P-AN-02Window open and close animations (150 ms map/destroy, 100 ms attached dialogs)partialcargo:gnome51_normal_map_pivot_and_exponential_curve, cargo:destroy_and_attached_dialog_follow_quad, cargo:reduced_motion_and_slowdown, cargo:animations_off_land_without_scale_or_fade, cargo:keyboard_workspace_slide_suppresses_lifecycle_and_resume_does_not_replay_map, cargo:interrupted_map_and_size_change_close_skip_raw_snapshot, cargo:expired_ghosts_release_budget_before_new_closeWired native map/destroy rendering and bounded retained textures; slides and interrupted map/resize close suppress effects. Paired GNOME/native GPU, X11 runtime, shadow/popup geometry and snapshot-budget overflow remain unqualified (#494); see implementation scope.—
P-AN-03Minimize and unminimize toward the dash icon (400 ms)pass (nested)cargo:minimize_400ms_expo_moves_scales_and_fades, cargo:unminimize_motion_stays_opaque_and_restores_geometry, cargo:icon_target_geometry_and_rtl_fallback_use_global_logical_monitor_bounds, cargo:off_is_immediate_and_slowdown_scales_duration, cargo:reduced_and_monitor_sized_use_only_opacity, cargo:clutter_guint_opacity_truncates_quarter_expo_samples, cargo:live_minimize_frame_uses_latest_buffer_geometry_and_owns_no_snapshot, cargo:dash_icon_targets_ignore_unknown_windows_and_retire_destroyed_window_targets, cargo:dash_icon_targets_roundtrip_and_reject_oversize_invalid_bounds, cargo:minimized_parent_hides_transients_and_child_activation_restores_ancestors, cargo:restoring_parent_preserves_independent_child_minimizeMinimize and restore use GNOME Shell 51's 400 ms ease-out-expo; restore stays opaque, reduced motion and monitor-sized windows only fade, Off is instant and the slow-down factor scales the duration (crates/compositor/src/minimize.rs, window_lifecycle.rs); the target is the window's dash icon, whose bounds the GTK overview dash publishes over the control wire (crates/shell-gtk/src/overview.rs), and without one the window goes to the monitor's leading corner (mirrored for RTL); attached dialogs hide and restore with their parent while keeping their own minimize flag (transient_family.rs); the source-matched native pixel proof (scripts/tuna-minimize-ci-proof, LTR and RTL, written to the native-minimize-proof artifact, which the ledger check does not read) passed in CI run 38878636595 alongside the unit tests (qualification)paired GNOME 51 frames, native GPU/DRM, multiple outputs and fractional scale, X11 at runtime, shadows and installed-image qualification remain missing, so the row is nested only (#495)
P-AN-04Maximize, unmaximize, tile and fullscreen size-change transition (250 ms crossfade)partialproof:G-SIZE-CHANGE-ANIM, cargo:waits_frozen_for_the_new_buffer_then_eases_for_250_ms, cargo:fade_only_puts_the_window_in_place_and_fades_the_old_frame_where_it_was, cargo:no_frame_stretches_new_content_beyond_its_final_sizeThe compositor snapshots the old frame, waits for the resized buffer and eases/crossfades over 250 ms (crates/compositor/src/size_change.rs); the nested GTK proof covers maximize, unmaximize and half-tiling at exact 0/125/250 ms, fade-only and animations off. Fullscreen requests use the size-change path, but the proof does not qualify their transition; paired GNOME 51 intermediate frames and native-GPU qualification remain missing (#496)—
P-AN-05Workspace switch slide, finger-following swipe, switcher popup fade and Activities dotspartialproof:G-WORKSPACE-SLIDE, proof:G-WS-POPUP-CLOCK, proof:G-WS-POPUP-POLICY, proof:G-WS-DOTS-CLOCK, proof:G-WS-DOTS-POLICY, proof:G-WS-DOTS-SWIPE, proof:G-WS-DOTS-SWIPE-POLICY, cargo:a_key_switch_slides_both_workspaces_for_250_ms, cargo:a_swipe_follows_the_fingers_then_finishes_on_its_velocity, cargo:a_second_press_retargets_from_the_drawn_offset, cargo:the_begin_sample_counts_the_first_movement_towards_the_velocity, cargo:a_swipe_cannot_drag_past_the_last_workspace, cargo:swipe_bounds_follow_gnomes_get_bounds, cargo:a_held_swipe_slides_under_every_motion_policy, cargo:swipe_phases_tell_the_dots_where_the_strip_is, cargo:workspace_progress_follows_seat_touch_mode_and_is_bounded, cargo:the_pill_fades_over_100_ms_ease_out_quad, cargo:reduced_motion_keeps_the_fade_and_animations_off_snap_it, cargo:the_slowdown_factor_stretches_the_fades_but_not_the_hold, cargo:the_fade_out_starts_600_ms_after_the_last_display, cargo:workspace_keys_at_the_ends_stay_put_for_the_popup, cargo:the_pill_moves_between_dots_over_250_ms, cargo:a_second_switch_retargets_from_the_drawn_pill, cargo:added_dots_scale_in_and_removed_dots_scale_out_over_500_ms, cargo:the_slowdown_factor_stretches_the_dots, cargo:without_motion_everything_lands_at_once, cargo:held_fingers_move_the_pill_and_the_release_settles_on_its_time, cargo:a_switch_arriving_before_the_release_is_retimed_by_it, cargo:held_fingers_move_the_pill_under_every_policy_and_releases_snap_without_motion, cargo:a_late_first_frame_samples_the_requested_fade_not_a_new_oneKeyboard switches slide for 250 ms with ease-out-cubic, rapid switches retarget without jumping, and swipes follow finger travel then settle using velocity (crates/compositor/src/workspace_slide.rs). Swipes snap only to existing workspaces (GNOME's trailing empty one included; no phantom workspace past it), bounds follow _getBounds for a swipe caught mid-switch, velocity history starts with the begin sample (swipeTracker.js:509-565), and a held swipe slides under every motion policy (animations off only zero the release; animationUtils.js:23-33). The popup (ws_popup.rs) and Activities dots (ws_dots.rs) run on the shell's proof clock: G-WS-POPUP-CLOCK samples the popup at 0/50/100 ms in, the 600 ms hold and 650/700 ms out, and a key on the first workspace redisplaying it opaque with a restarted hold (workspaceSwitcherPopup.js:93-121; windowManager.js:1869-1880); G-WS-DOTS-CLOCK samples the dots' panel.js geometry at 0/125/250 ms each way; G-WS-DOTS-SWIPE checks the dots follow a held swipe (compositor→shell WorkspaceProgress, workspaceAnimation.js:255-268) and settle on the strip's own release time and curve; popup fades start at their request rather than the first GTK tick, with a delayed-first-frame unit regression for normal/slow/off and fade-out; the policy gates cover Reduced Motion (popup fade kept, dots snap) and animations off. Remains: paired GNOME 51 intermediate-frame/swipe captures; multi-output (the nested proof has one output: the popup's primary-only/per-monitor placement, per-monitor slide distance and MonitorGroup clipping are unmeasured); Super+Page_Down past the dynamic trailing workspace still opens another (mutter clamps, workspace.c:1312-1321) and Shift+Super+Page_Up on the first workspace does not prepend one (windowManager.js:1824-1829); Alt+Tab does not dismiss the popup (windowManager.js:1733); Reduced Motion crossfades key switches (#523) where GNOME 51 slides (#497, #596)—
P-AN-06Parent window dims behind an attached modal dialog (500 ms in, 250 ms out)partialcargo:parent_dims_500ms_and_undims_250ms_quad, cargo:overlapping_modal_membership_does_not_restart_and_reversal_is_continuous, cargo:reduced_motion_preserves_fade_off_and_suppression_snap_and_parent_death_clears, cargo:parent_brightness_change_invalidates_native_repaint_and_damage_layout, cargo:live_attach_preference_gates_effect_without_restarting_logical_fadeNative xdg-dialog parent RGB shader and 500/250 ms fades wired; transparent alpha preserved, overlapping children keep parent dimmed. Paired GNOME, native GPU/DRM, X11 roles, overview previews, installed settings delivery and performance remain unqualified (#498); scope and focused proof.—
P-AN-07Overview transitions: open/close easing, gesture momentum, picker and app grid, hover, thumbnails, drag, search, dash and panelpartialproof:G-OVERVIEW-EASING, proof:G-OV-APPGRID, proof:G-DASH-MOTION, cargo:swipe_release_projects_velocity_like_gnome, cargo:the_app_grid_swap_morphs_out_sine_over_250_ms, cargo:hover_grows_over_200_ms_out_quad, cargo:drags_shrink_then_snap_back_or_revert_like_gnome, cargo:search_fades_the_workspaces_over_250_ms, cargo:dash_motion::tests::gnome_51_dash_durations, cargo:dash_motion::tests::items_zoom_and_fade_in_and_out_over_200_ms_ease_out_quad, cargo:dash_motion::tests::an_item_removed_while_zooming_in_shrinks_back_from_where_it_stands, cargo:dash_motion::tests::reduced_motion_keeps_the_fade_at_full_size_and_off_lands_at_once, cargo:dash_motion::tests::labels_fade_in_over_150_ms_and_out_over_100_ms_from_where_they_stand, cargo:dash_motion::tests::the_slow_down_factor_stretches_dash_motionOverview opening uses out-sine and closing out-quad; swipe release carries momentum; picker/app-grid morph, workspace scroll, thumbnail growth, hover, drag and search transitions are implemented (crates/compositor/src/overview.rs). Nested proof samples open/close easing and Show Apps; unit tests cover the remaining curves. The dash follows gnome-shell 51 dash.js (crates/shell-gtk/src/dash_motion.rs, overview.rs): items persist across redisplays; an app joining zooms and fades in and one leaving zooms and fades out over DASH_ANIMATION_TIME 200 ms ease-out-quad with its size following its scale (DashItemContainer.show/animateOutAndDestroy, :154-185, :82-98), not on the overview's initial fill (:784-798); the hovered item's label (GNOME's dash-label pill, 8 px above it) fades in over 150 ms after DASH_ITEM_HOVER_TIMEOUT 300 ms (at once while labels are showing) and out over 100 ms, hidden on click and when the overview hides (:100-141, :483-571); like GNOME's chrome St.Label it is a plain label to assistive technology while the item's button keeps the accessible name; the separator shows only between favorites and other apps (nFavorites > 0, :809-835). G-DASH-MOTION samples item add/remove, a removal mid-add, the label fade in and out, Reduced Motion and animations off on the frozen proof clock. Deviations under the shared motion policy and interruption rule: Reduced Motion keeps items at full size while they fade (GNOME 51's dash.js does not consult reduced-motion); a label shown again while fading out fades in from its drawn opacity where showLabel restarts from 0; an app re-added while animating out zooms back in where GNOME's _redisplay lets it finish leaving; a redisplay during the overview's own open/close animation still animates (GNOME checks animationInProgress). Not implemented: _adjustIconSize (the dash keeps 64 px icons, so no icon-size animation), drag-and-drop placeholders and the window-drag dash fade. Paired GNOME 51 intermediate-frame/gesture captures and panel transition qualification remain missing (#499)—
P-AN-08App grid animations: folder zoom, launch zoom, page switch curve, swipe settle, page dots and page arrows (GNOME 51 defines AppIcon.scaleIn() but never calls it, so new icons do not scale in)partialproof:G-GRID-PAGE-ANIM, proof:G-FOLDER-ZOOM, proof:G-LAUNCH-ZOOM, proof:G-GRID-ACTIVATE-RUNNING, proof:G-GRID-FADE, proof:G-GRID-PAGE-PEEK, cargo:grid_animation::tests::curves_match_clutter_easing_modes, cargo:grid_animation::tests::gnome_51_durations, cargo:grid_animation::tests::reduced_motion_keeps_fades_and_off_keeps_nothing, cargo:grid_animation::tests::slow_down_factor_stretches_durations_and_delays, cargo:grid_animation::tests::folder_dialog_zooms_from_its_icon, cargo:grid_animation::tests::launch_zoom_triples_and_stays_on_the_monitor, cargo:grid_animation::tests::page_switch_eases_out_cubic_over_300_ms, cargo:grid_animation::tests::page_dots_follow_the_drawn_position, cargo:grid_animation::tests::swipe_history_keeps_150_ms_of_moves, cargo:grid_animation::tests::a_swipe_goes_at_most_a_page_past_where_it_began, cargo:grid_animation::tests::released_swipes_settle_as_gnome_swipe_tracker, cargo:the_plan_follows_app_icon_activate, cargo:a_stopped_app_launches_with_the_zoom_even_with_ctrl, cargo:exact_running_app_activates_before_close_without_spawning_or_zooming, cargo:ctrl_or_middle_on_a_running_app_opens_a_new_window_with_the_zoom, cargo:ctrl_on_a_running_single_main_window_app_raises_it, cargo:failed_spawn_keeps_overview_open_without_the_zoom, cargo:can_open_new_window_reads_the_desktop_file_like_shell_app, cargo:the_new_window_action_is_parsed_like_exec, cargo:open_new_window_runs_the_action_else_launches_againMatched to gnome-shell 51.0 appDisplay.js, iconGrid.js, pageIndicators.js and swipeTracker.js (crates/shell-gtk/src/grid_animation.rs, grid_pager.rs, folder_dialog.rs, overview.rs). The folder dialog zooms 200 ms ease-out-expo out of its icon and back with the shade and icon fades (G-FOLDER-ZOOM, incl. animations-off and Reduced Motion). Grid, folder, dash and search result tiles share GNOME's AppIcon.activate (appDisplay.js:3053-3071; primary and middle button, :1854): a stopped app launches and its tile's last drawn icon (and label, in the grid) grows 3× while fading over 250 ms ease-out-quad, kept on the monitor (G-LAUNCH-ZOOM, from a folder); a running app's window is raised with no launch and no zoom (G-GRID-ACTIVATE-RUNNING clicks a running app's grid tile: its window takes focus, no new window, no relaunch, no zoom surface); Ctrl+click or a middle click on a running app opens a new window with the zoom when shell_app_can_open_new_window (src/shell-app.c:631-700) allows it from the desktop file (SingleMainWindow, then X-GNOME-SingleWindow, then a new-window desktop action, else yes), through the new-window desktop action or a relaunch (shell_app_open_new_window, :574-623; G-GRID-ACTIVATE-RUNNING Ctrl+clicks for the relaunch and its zoom). The dash zoom has no CI gate because the proof's dash favorites are fixed at shell start. goToPage eases 300 ms out-cubic from wherever the pages are drawn; G-GRID-PAGE-ANIM checks every animated frame's GDK frame-clock sample against GNOME's curve and every page-dot update against pageIndicators.js, and that animations off jump. Arrows fade 200 ms; swipes settle on SwipeTracker's projected page and duration (unit-tested only, no CI swipe gate). Not frozen-clock: GTK shell animations run on the GDK frame clock, not TUNA_ANIMATION_CLOCK. Reduced Motion keeps only fades per Tuna's shared motion policy (GNOME 51's appDisplay.js does not consult reduced-motion). Not matched in activation, for want of what ShellApp tracks: a running app's exported app.new-window D-Bus action and the unique-GtkApplication test are not asked (such apps count as able to open a window), and bus activation of app.new-window and of DBusActivatable apps is not used (Tuna runs the Exec lines); shell_app_activate_window raises the app's most recently used window (and its other windows on that workspace) where Tuna raises its lowest-id window; GNOME's STARTING state (a click on a launched app with no window yet does nothing) is not tracked, so it launches again; a failed launch keeps the overview open where GNOME hides it and notifies. Paired GNOME 51 clips are missing (#500)—
P-AN-09Menus, quick-settings submenus, OSD, notification banners and list, Alt+Tab, modal dialogs and lightboxpartialproof:G-SWITCHER-FADE, proof:G-TRANSIENT-SETTLE, proof:G-TRANSIENT-OFF, cargo:transient::tests::gnome_51_durations, cargo:transient::tests::banners_drop_in_scaled_and_slide_back_up, cargo:transient::tests::list_items_zoom_in_and_out, cargo:transient::tests::submenus_grow_then_fade_and_fade_then_shrink, cargo:transient::tests::fades_ease_out_quad_from_where_they_stand, cargo:transient::tests::easing_curves_match_clutter, cargo:transient::tests::progress_is_linear_and_instant_when_animations_are_off, cargo:transient::tests::panel_menus_fade_rise_and_grow_open, cargo:transient::tests::a_menu_taking_over_from_another_opens_fade_only, cargo:menu_close::tests::closes_with_gnome_51_boxpointer_curve, cargo:menu_close::tests::fade_only_and_reduced_motion_drop_the_rise, cargo:menu_close::tests::shrinks_about_the_top_centre_and_rises, proof:G-TRANSIENT-CLOSE-CLOCK, proof:G-NM-WIRED, cargo:transient::tests::ease_out_expo_matches_clutter, cargo:transient::tests::popup_submenus_grow_and_turn_their_arrow_over_250_ms_out_expo, cargo:transient::tests::a_reversed_popup_submenu_starts_from_its_drawn_frame, cargo:transient::tests::popup_submenus_skip_the_animation_when_the_menu_scrolls, cargo:transient::tests::popup_submenus_keep_their_motion_under_reduced_motion, cargo:transient::tests::menus_close_as_boxpointer_from_where_they_stand, cargo:transient::tests::a_hidden_buttons_menu_closes_unanimated, cargo:transient::gtk_menu_tests::actual_gtk_popup_submenus_ease_open_and_closed, cargo:transient::gtk_menu_tests::actual_gtk_menus_close_at_once_and_switch_fade_only, cargo:menu_close::tests::a_hidden_buttons_close_pairs_whichever_arrives_first, cargo:menu_close::tests::an_unanimated_request_waits_for_its_popup_and_is_recorded, cargo:runtime::tests::switcher_thumbnails_draw_their_window_at_the_frames_opacity, cargo:switcher::tests::window_images_take_their_frames_opacity_as_clutter_does, cargo:switcher_fade_and_menu_close_wire_tests::switcher_thumbnail_opacity_rides_last_and_round_trips, cargo:switcher_fade_and_menu_close_wire_tests::close_menu_unanimated_follows_pad_osd_closed, cargo:control::switcher_thumbnail_tests::a_departed_shell_takes_its_switcher_thumbnails_with_itNotification groups animate expansion; the Alt+Tab popup fades out, thumbnails fade in/out, and the focused thumbnail strip scrolls over 100 ms ease-out-quad (crates/shell-gtk/src/switcher.rs, G-SWITCHER-FADE), and the compositor draws the thumbnails' window images at their frame's opacity (wire SwitcherThumbnail.opacity), so they fade in with the strip and out with the popup instead of popping (G-SWITCHER-FADE samples the published alphas at 0/50/100 ms in and 50 ms out on the proof clock); a shell that disconnects, crashes or is replaced while they are drawn takes them with it, and none draw under the lock. The OSD and its level fade over 100 ms, and a show during its fade-out lets the fade-out finish and hide it as osdWindow.js does (actual-GTK X11 test in CI), banners drop in and slide out over 200 ms, list items zoom over 100 ms, polkit, end-session and network-secret modals fade with their lightbox over 100 ms, panel menus (date, quick settings, accessibility, tray, input method, dwell) open as GNOME's BoxPointer (fade in, rise 6px and grow from 96% about the top centre over 150 ms ease-out quad; fade-only when taking over from another menu) and close as BoxPointer.close drawn by the compositor from the menu's last frame (fade out, rise 6px, shrink to 96%; fade-only on a takeover) while GTK closes them logically, for the accessible tree and input, at once (crates/compositor/src/menu_close.rs, G-TRANSIENT-CLOSE-CLOCK samples 0/75/150 ms on the proof clock); a panel button hidden under its open menu closes it unanimated as PanelMenu.Button.vfunc_hide does (panelMenu.js:149-154; CloseMenuUnanimated, G-TRANSIENT-CLOSE-CLOCK hides the Accessibility button) and the menu stays closed when the button returns. The window menu opens with BoxPointer's fade alone and closes with its full close over 150 ms (windowMenu.js:239, crates/shell-gtk/src/window_menu.rs, G-TRANSIENT-SETTLE). Popup submenus (wired adapters' NMSectionItem, IBus property menus) open and close as PopupSubMenu (popupMenu.js:1265-1342): height and arrow (0→90°, −90° right to left) ease out expo over 250 ms from where they stand, no animation when the menu must scroll, Left closes back to the item, the item carries the expanded state and a labelled group holds item and rows; GNOME animates the height under Reduced Motion too, so only animations off lands it (crates/shell-gtk/src/transient.rs PopupSubmenu, actual-GTK X11 test in CI; G-NM-WIRED opens a second adapter's submenu but checks its rows, not its frames). Quick-settings submenus grow then fade over 400 ms; Reduced Motion keeps only the fades and animations off lands them at once (crates/shell-gtk/src/transient.rs, G-TRANSIENT-SETTLE, G-TRANSIENT-OFF). The menu close is sampled as compositor state rather than pixels, the OSD's fade, level ease and the menu rise are proven by trace timings and unit frames rather than pixel captures, popup submenu frames are proven under X11 GTK only and no proof drives an IBus property submenu, the on-screen keyboard's extended-key and language popups (keyboard.js:289,1462, FULL BoxPointer) are still unanimated, the lightbox is the modal window's own background, message-list and lightbox timing are covered by unit tests only, and paired GNOME 51 intermediate-frame captures remain missing (#501)—
P-AN-10Lock curtain slide, wallpaper crossfade, screenshot flash, hot-corner ripples, startup and monitor-change transitionspartialcargo:screenshot_flash::tests::white_then_quarter_opacity_at_half_time_and_absent_at_end, cargo:screenshot_flash::tests::concurrent_feedback_is_bounded_and_lock_clear_is_immediate, cargo:screenshot_flash::tests::disabled_motion_and_invalid_areas_create_no_feedback, proof:G-LOCK-SLIDE, proof:G-STARTUP-REVEAL, proof:G-WALLPAPER-CROSSFADE, proof:G-HOT-CORNER-RIPPLES, proof:G-MONITOR-TRANSITION, cargo:startup_and_monitor_change_fade_on_gnome_timing, cargo:startup_holds_until_the_shell_is_ready_or_the_cap, proof:G-LOCK-CURTAIN-CLOCK, proof:G-LOCK-CURTAIN-POLICY, proof:G-LOCK-CURTAIN-DRAG, proof:G-LOCK-CURTAIN-HINT, proof:G-LOCK-CURTAIN-TOUCHPAD, proof:G-LOCK-SWIPE-SECURITY, cargo:lock_curtain::tests::pages_trade_places_like_unlock_dialog_51, cargo:lock_curtain::tests::reduced_motion_keeps_only_the_fade, cargo:lock_curtain::tests::crossfade_is_300_ms_ease_out_quad_from_the_current_value, cargo:lock_curtain::tests::a_drag_moves_nothing_until_16_px_then_follows_each_event, cargo:lock_curtain::tests::pan_velocity_is_clutters_last_150_ms, cargo:lock_curtain::tests::slow_releases_settle_on_the_nearer_page, cargo:lock_curtain::tests::a_flick_reaches_the_other_page_from_a_short_drag, cargo:lock_curtain::tests::a_cancelled_swipe_returns_to_where_it_began, cargo:lock_curtain::tests::touchpad_releases_use_the_touchpad_thresholds, cargo:lock_curtain::tests::touchpad_history_is_gnomes_event_history, cargo:lock_curtain::tests::touchpad_swipe_follows_gnomes_three_finger_gesture, cargo:lock_curtain::tests::touchpad_scroll_follows_and_the_stop_ends_it, cargo:lock_curtain::tests::the_hint_follows_touch_mode, cargo:lock_curtain::tests::the_hint_fades_in_on_clutters_default_ease_and_hides_at_once, cargo:lock_curtain::gtk_curtain_tests::actual_gtk_curtain_pages_follow_progress_and_policy, cargo:anim_clock::tests::a_file_freezes_steps_and_releases_the_clock, cargo:a_finished_finger_scroll_sends_axis_stop, cargo:lock_surface_alone_hears_touchpad_swipes_and_the_scroll_stop, cargo:hub_broadcasts_seat_touch_mode_changes_and_hands_newcomers_the_valueIdle shield fade and unlock crossfade animate (crates/compositor/src/lock.rs); the compositor's curtain lowers on lock and lifts after PAM (crates/compositor/src/transitions.rs, G-LOCK-SLIDE). The unlock dialog moves between its clock curtain and prompt on unlockDialog.js 51's single progress (crates/shell-gtk/src/lock_curtain.rs): the clock fades, rises 200 px and shrinks toward 30% while the prompt fades in from 200 px below; a key (a printable one becoming the first password character), click, wheel step or Escape eases it over 300 ms ease-out quad, and a vertical pointer drag (touch arrives as the primary button) follows past Clutter's 16 px and settles on swipeTracker.js's velocity projection in 100-400 ms ease-out cubic; Reduced Motion keeps the fade only, animations off switch at once, and the leaving prompt drops focus and is cleared once the clock is back. Touchpad: while locked the compositor hands the lock surface (and nothing else) touchpad swipes as pointer gestures, and every finger scroll's end as wl_pointer.axis_stop (mutter meta-seat-impl.c:2254-2277, meta-wayland-pointer.c:927-967; unlocked clients get the stop too); a three-finger swipe under way at lock ends cancelled, so no gesture opens the overview or switches workspace while locked (G-LOCK-SWIPE-SECURITY asserts it from the state document). The dialog follows swipeTracker.js's TouchpadSwipeGesture (3+ fingers, 16 px orientation lock, natural-scroll) and ScrollGesture (ended by the stop) over TOUCHPAD_BASE_HEIGHT with EventHistory velocity and the touchpad thresholds; G-LOCK-CURTAIN-TOUCHPAD injects both through the nested swipe-input file and checks each release against GNOME's formulas. The hint is hidden until the IdleMonitor's 4 s HINT_TIMEOUT watch fires, then fades in over 300 ms on Clutter's default ease-out cubic; a PowerSaveMode change hides it at once; its text follows the seat's touch-mode ("Swipe up" / "Click or press a key", new wire SeatTouchMode); G-LOCK-CURTAIN-HINT samples 0/150/300 ms on the frozen clock under full motion and Reduced Motion and checks animations off. The shell samples the compositor's proof clock, so G-LOCK-CURTAIN-CLOCK/-POLICY check frames at exact delays and G-LOCK-CURTAIN-DRAG drives real xdotool drags, a flick and a PAM unlock. Screenshot methods pass their flash request to an output-only white fade (crates/compositor/src/screenshot_flash.rs). Wallpaper replacement crossfades over 1000 ms ease-out quad (background.js:119,897-900; crates/compositor/src/wallpaper.rs), the hot corner draws ripples.js's three rings, and the startup cover and monitor-change transition (layout.js:21-28 timings, ScreenTransition) run in transitions.rs; their gates pass on main CI. Not matched: touchpad gestures are proven with nested synthetic input only, never a hardware touchpad; the swipe reaches the lock surface with accelerated deltas (GNOME reads unaccelerated ones); the touch-mode text is unit-tested only (nested sessions have no touchscreen); no auth-mechanism hint, notifications box, auth-method indicator or other-user button; Tuna starts in the overview like GNOME, but its startup is the compositor's black cover fade (GNOME's !hasOverview path) instead of overviewControls.js:798-860 runStartupAnimation (controls fade in, search falls, dash rises after 500 ms), and the uiGroup 0.75 scale is not drawn; the screenshot-flash pixel proof is not run in CI; paired GNOME frames for every transition are missing (#502)—
P-SY-11Clipboard contents stay pasteable after the copying app closespartialcargo:clipboard_survives_owner_disconnect, cargo:clipboard_survives_owner_null_selection, cargo:retained_clipboard_keeps_the_most_preferred_type, cargo:new_client_selection_replaces_retained_copy, cargo:unsupported_clipboard_type_is_not_retainedImplemented in #1056 (crates/compositor/src/clipboard_manager.rs) following mutter 51 meta-clipboard-manager.c: the most preferred supported type is copied (text 4 MiB, images 200 MiB) and served after the owner clears, destroys its source or quits. Proven with real socketpair Wayland clients only; no nested or installed proof with real applications. A RemoteDesktop session's selection is not copied (recorded divergence). (#1033)—
P-WM-10X11 windows without their own decorations get a server-side framemissingUntracked until the 2026-11-10 audit (main 8cd7dd75). GNOME 51: mutter src/frames/ (mutter-x11-frames). No frame code exists in crates/compositor/src; X11 windows are mapped bare (for example docs/walkthrough/x11.png). (#1034)—
P-WM-11An app that stops responding gets the "not responding" dialog with Force Quit and Waitpartialcargo:unanswered_ping_shows_the_close_dialog_until_a_late_pong, cargo:answering_window_is_never_reported, cargo:force_quit_disconnects_the_unresponsive_client, cargo:zero_check_alive_timeout_disables_the_check, cargo:lock_hides_the_close_dialog_and_stops_pinging, cargo:strings_match_gnome_shell_51Implemented in #1060 (crates/compositor/src/unresponsive.rs, windows/alive.rs, crates/shell-gtk/src/close_dialog.rs, wire 0.50) following mutter 51 ping/check-alive-timeout and gnome-shell closeDialog.js. X11 windows are not pinged (Smithay 0.7 does not report _NET_WM_PING replies); no nested or installed proof with a hung application and no paired GNOME frame. (#1035)—
P-PN-04Calendar popover shows World Clocks and Weatherpartialproof:G-DATEMENU-WORLD-CLOCKS, proof:G-DATEMENU-WEATHER, cargo:offsets_read_as_gnome_writes_them, cargo:clocks_sort_by_standard_offset_and_keep_ties_in_settings_order, cargo:clocks_own_setting_yields_each_location, cargo:time_only_follows_format_time, cargo:auto_location_needs_the_request_the_switch_and_the_grant, cargo:a_recent_valid_forecast_refreshes_without_loading, cargo:fixture_states_map_to_the_client_statesGTK shell date menu shows GNOME 51's World Clocks and Weather cards under the events card in a scrolling, edge-fading displays section (crates/shell-gtk/src/world_clocks.rs, weather.rs, weather_section.rs, app_integration.rs; libgweather 4, 4.4.2 or later). Nested at 1280×800 (scripts/tuna-gtk-shell-proof --datemenu-only): cards follow the apps' desktop entries; clocks follow the setting, a running Clocks' exported Locations, clock-format and the minute, with GNOME's times and offsets in UTC-offset order; Weather shows gnome-shell 51.0's prompt, Loading…, offline and unavailable lines and five hourly forecasts; the menu stays on the output and the section scrolls; pointer, accessible action and Shift+Tab + Enter open the apps. Forecasts come from a fixture in the proof; a live MET Norway forecast was observed by hand. Not proven: GeoClue automatic location, the real Clocks and Weather apps, paired GNOME frames. (#1036)The section fades over GTK's 40px undershoot where St's .vfade uses 68px; the User-Agent names org.tuna.Shell
P-NT-04Media player controls (MPRIS) in the notification listmissingUntracked until the 2026-11-10 audit (main 8cd7dd75). GNOME 51: gnome-shell js/ui/mpris.js. No MPRIS code exists on main. Open PR #926 adds media controls to the lock screen only. (#1037)—
P-PN-05Quick settings Airplane Mode and keyboard backlight togglespartialcargo:airplane_toggle_hidden_without_should_show, cargo:keyboard_levels_follow_gnome_steps, cargo:keyboard_click_turns_off_or_fullImplemented in #1053 (crates/shell-gtk/src/rfkill.rs, kbd_backlight.rs) following gnome-shell 51 rfkill.js and backlight.js. Unit-tested logic only; no nested or installed proof and no paired GNOME frames. (#1038)—
P-SY-12Switch User from the system menupartialcargo:switch_user_needs_multiple_users_and_multi_session, cargo:lockdown_hides_switch_user, cargo:an_existing_greeter_session_is_activated_before_creating_oneImplemented in #1055 (crates/shell-gtk/src/user_switch.rs) following systemActions.js and libgdm. Hardware sessions only; never exercised against a real GDM with two accounts. The unlock-dialog user switch is still unimplemented (#365). (#1039)—
P-SY-13InputCapture portal backend (org.gnome.Mutter.InputCapture)partialcargo:barriers_validate_as_mutters_check_barrier, cargo:router_captures_from_the_crossing_until_release, cargo:lock_and_pointer_constraints_never_capture, cargo:only_the_installed_portal_may_create_input_capture_sessions, cargo:an_input_capture_session_follows_mutter_from_barrier_to_releaseImplemented in #1062 (crates/compositor/src/mutter/input_capture.rs, capture_eis.rs) following mutter 51 meta-input-capture*.c; private-bus tests use a real reis receiver. Capture is wired on the DRM input path only and was never run with the real portal and an Input Leap/Deskflow client; cursor hiding, the capture-session clipboard and keymap updates during a connection are not implemented. (#1040)—
P-SY-14Thunderbolt device authorisation notificationspartialcargo:bolt_absent_does_nothing, cargo:auto_policy_enrols_when_the_user_may, cargo:unauthorized_device_while_locked_notifies_without_enrolling, cargo:failed_enrolment_stops_auto_enrolment_until_restartImplemented in #1058 (crates/shell-gtk/src/thunderbolt.rs) following thunderbolt.js. No boltd or Thunderbolt hardware was involved; ADR 0008 allows labelled emulated-device evidence. (#1041)—
P-ST-12Captive-portal login window for hotspot networkspartialcargo:portal_state_opens_the_helper_for_the_primary_connection, cargo:full_connectivity_closes_an_open_helper, cargo:done_with_retry_rechecks_connectivityImplemented in #1054 (crates/shell-gtk/src/captive_portal.rs) following network.js CaptivePortalHandler. Offers sign-in as a notification as GNOME does; needs the gnome-shell portal helper installed. Never run against a real captive portal. (#1042)—

Session-notification qualification scope​

For P-LK-01 and #556, scripts/tuna-session-sync-proof exercises the production runtime's sleep and seat handlers with simulated notifications. A supervised real ControlClient reconnects across pause/resume, checks its mapped window/workspace/overview state, then checks that suspend locks and redacts snapshots and that waking without a shell cannot unlock. This is nested runtime/control propagation evidence, not physical VT/S3, libseat fd revocation, KMS scanout or libinput recovery qualification; those still require a native seat/DRM proof. No row status is upgraded.

Tests column​

Comma-separated references, checked by scripts/tuna-ledger in CI (the parity-ledger job):

  • cargo:<name> names a workspace test, by full path or by its last segment (cargo test -- --list). The check job runs every test, so a listed test is a passing test.
  • journey:<ID> names a state assertion recorded as <ID> pass in a journey's assertions.txt (see scripts/lib/tuna-introspect.sh).
  • proof:<ID> names a stage recorded as <ID> pass by the GTK shell proof (scripts/tuna-gtk-shell-proof), the screen magnifier proof (scripts/tuna-magnifier-proof), the on-screen keyboard proof (scripts/tuna-osk-proof) or the DRM smoke test (scripts/tuna-drm-smoke).

A row whose status starts with pass must cite at least one test, and every cited test must exist and pass. CI fails otherwise.

How to update​

Add a row when a GNOME 51 behavior enters scope. Move a row to pass only with a cited test or a recorded review entry (docs/reviews/). A row whose cited test disappears or fails breaks CI.

v1 qualification check​

Use scripts/tuna-ledger check --require-complete with the same --tests and --assertions inputs as the CI ledger check. This stricter mode rejects every partial, missing, deviation, untested and pass (nested) row; a nested-only proof does not establish the shipped Marlin desktop. It also rejects empty ledgers and invalid statuses. The normal check still permits incremental work.

Passing this check establishes only the ledger contract. Published image, daily-driver, gap audit and paired performance evidence remain separate v1 gates.