diff --git a/docs/control-protocol.md b/docs/control-protocol.md
index 7d9fa04b..e87525bd 100644
--- a/docs/control-protocol.md
+++ b/docs/control-protocol.md
@@ -30,7 +30,7 @@ mode ([mcp.md](mcp.md)) is a thin layer over this channel. Implementation:
 | `engine.commands` | — | Engine + UI commands: id, label, menu, shortcut, params doc, enabled |
 | `ui.menu.list` | — | Menu entries (flat: id, label, menu path, shortcut, enabled) |
 | `ui.menu.tree` | — | The menu bar as shown (File … Help): items `{id, params?, label, shortcut?, enabled, checked?}`, separators, submenus — the model behind the native macOS menu bar and the in-window menus |
-| `ui.inspect` | — | UI state, window, `loupe` (`source`, and `region: {full, window, pending}` when a zoomed view also renders the window on screen at no more than 100 % — `full` is the frame it is cut from, `window` its `[x, y, w, h]` — and `regionBefore` for the Before side of a Before/After view), canvas/image rects, `scroll: {grid, filmstrip}` (scroll offsets in points, `null` until drawn), active photo, selection, perf (`frameMs` = layout, `logicMs` = per-frame logic before it, `updateMs` = both, `maxUpdateMs`, `fps`, render queue …, `gpu` = adapter in use, `gpuReason` = why renders don't use the GPU, `gpuFallback` = latest render redone on the CPU and why — see `docs/gpu-pipeline.md`), status, memory (bytes per cache, see `library.memory`; plus stage caches and textures), `export: {running: {total, done, current} \| null, last}`, `notices` (warnings waiting to be shown, e.g. a damaged settings file; OK = `button:noticeOk`), `quitPrompt` (why quitting was stopped: unsaved changes; `button:quitRetry` / `button:quitAnyway` / `button:quitCancel`), `import: {done, total, imported, cancelled} \| null` (an import runs on a worker thread), `tasks` (other background work: `Find Missing Photos`, `Auto Import`) |
+| `ui.inspect` | — | UI state, window, `loupe` (`source`, and `region: {full, window, pending}` when a zoomed view also renders the window on screen at no more than 100 % — `full` is the frame it is cut from, `window` its `[x, y, w, h]` — and `regionBefore` for the Before side of a Before/After view), canvas/image rects, `scroll: {grid, filmstrip}` (scroll offsets in points, `null` until drawn), active photo, selection, perf (`frameMs` = layout, `logicMs` = per-frame logic before it, `updateMs` = both, `maxUpdateMs`, `fps`, render queue …, `gpu` = adapter in use, `gpuReason` = why renders don't use the GPU, `gpuFallback` = latest render redone on the CPU and why — see `docs/gpu-pipeline.md`), status, memory (bytes per cache, see `library.memory`; plus stage caches — `budgetBytes`, `trimmed`, `sharedSourceBytes` (the one device copy of a big original) — and textures), `export: {running: {total, done, current} \| null, last}`, `notices` (warnings waiting to be shown, e.g. a damaged settings file; OK = `button:noticeOk`), `quitPrompt` (why quitting was stopped: unsaved changes; `button:quitRetry` / `button:quitAnyway` / `button:quitCancel`), `import: {done, total, imported, cancelled} \| null` (an import runs on a worker thread), `tasks` (other background work: `Find Missing Photos`, `Auto Import`) |
 | `ui.widgets` | `{filter?}` | On-screen widgets `{id, rect: [x, y, w, h]}` (screen points) |
 | `ui.clickWidget` / `ui.dragWidget` / `ui.hoverWidget` | `{id, count?, fx?, fy?}` / `{id, toX?, toY?, dx?, dy?, steps?}` / `{id, fx?, fy?}` | Real egui input on a widget (hover: the pointer rests on it, e.g. for preset/profile previews) |
 | `ui.move` / `ui.click` / `ui.drag` | `{x, y, count?, button?}` / `{x, y, toX, toY, steps?}` | Raw pointer input, screen points |
diff --git a/docs/gpu-pipeline.md b/docs/gpu-pipeline.md
index c0a96957..ea7c8c64 100644
--- a/docs/gpu-pipeline.md
+++ b/docs/gpu-pipeline.md
@@ -212,7 +212,7 @@ the device on their first GPU render. `lightcraft_gpu::ready()` asks without blo
   developed image in ~50 ms.
 
 ## Memory (M5.6)
-A view's stages keep an uploaded source only up to 96 MB (a Preview-level source); a bigger one (the original a zoom window is cut from) is kept once, outside the views' stages, for all views of the photo (the Before and After windows and every pan share it) and given back when the app idles (issue #323: kept per view it counted against the stage budget on every tick of a drag, and not kept every new window uploaded 288 MB again).
+A view's stages keep an uploaded source only up to 96 MB (a Preview-level source); a bigger one (the original a zoom window is cut from) is kept once, outside the views' stages, for all views of the photo (the Before and After windows and every pan share it; the Before window of a Before/After waits at first open until the After's job has handed the original over, or each would decode and upload its own copy) and given back when the app idles (issue #323: kept per view it counted against the stage budget on every tick of a drag, and not kept every new window uploaded 288 MB again).
 - `library.memory` reports what the engine's caches hold (decoded thumbnail / preview / full-size
   sources, rendered previews) and the GPU renderer's device buffers (allocated, of which pooled
   and retired); `ui.inspect` → `memory` adds the loupe's stage caches (CPU images, GPU buffers)
diff --git a/docs/parity.md b/docs/parity.md
index 826b8b88..7104b4ad 100644
--- a/docs/parity.md
+++ b/docs/parity.md
@@ -185,7 +185,7 @@ under the UI command that wraps it (a no-op error); the UI command now wins.
 | LR-VIEW-EDIT | Edit view | P0 | ✅ | `cmd:panel.edit` | |
 | LR-VIEW-FULLSCREEN | Full-screen preview | P1 | ✅ | `cmd:view.fullScreenPreview`, `cmd:view.enterFullScreen` | photo on black, arrows step, Esc exits; ⇧⌘F window full screen |
 | LR-VIEW-FILMSTRIP | Filmstrip | P0 | ✅ | `cmd:view.filmstrip`, `crates/ui-egui/src/panels/detail.rs` | the photo context menu on right-click; the mouse wheel scrolls it sideways; it follows the active photo only when that changes (centred if off screen), so a scrolled strip stays put; every selected photo is highlighted, the active one framed in white, as in the grid |
-| LR-VIEW-ZOOM | Zoom & pan | P0 | 🟡 | `cmd:view.zoomFit`, `cmd:view.zoom100`, `cmd:view.zoomIn`, `cmd:view.zoomOut`, `cmd:view.zoomToggle`, `cmd:view.clickZoom`, `cmd:view.navigate`, `crates/ui-egui/src/panels/detail.rs`, `crates/ui-egui/src/headless.rs` | continuous native pinch / modifier-wheel zoom (Fit–800%, anchored under the pointer), two-finger scroll / touch pan with edge limits; Detail (including tools, Before/After and full-screen preview), Compare and Reference; fractional percentages preserve small gestures; image, picker and targeted-adjustment cursors stay steady between gesture events; keyboard steps 25–800 % (not 6–1600 %); Fill in the bottom bar; click / Z / Space eases to the chosen ratio (1:1 default, 2:1, 3:1, 4:1, 8:1). The loupe's cost follows the canvas, not the zoom (issue #323): the whole frame is rendered at about canvas size (at most the 2560 px preview source level unless a larger Preview size is chosen, and never above the GPU's texture side), and once the view is zoomed well past that (more than 25 %) the window on screen is rendered too, cut from a frame of at most the photo's own pixels, so beyond 100 % the GPU magnifies it (`region::plan`, `RenderRequest::window`, `Session::region_job`, `crates/ui-egui/src/region.rs`); 1:1 and 2:1–8:1 are sharp at any photo size. A slider tick takes 7–12 ms on a 1400 × 900 canvas and 10–20 ms on a 2800 × 1700 one at 100 % and 400 % of a 24 MP photo (M1 Max, GPU, `profile_slider_drag`); 20 ms at 100 % on the big canvas is over a 16 ms frame. A continuous pinch holds the render sizes (and the windows there are) until it is quiet. A fit view on a canvas up to about 3200 px (a Retina Mac) never decodes the original; larger displays get a window at fit, which does. Modes that draw another picture over the loupe (mask overlays, soft proofing) and photos whose spots or Auto Mask strokes read too far for one window render the whole frame at the size it is drawn, as before windows. Before/After (side by side, stacked, `\\` and both wipes) gets the same windows on both sides, clipped to their pane or side of the line. Windows render on the GPU like whole frames (equivalence-tested against the CPU), in the browser build too (windows no wider than its GPU's texture side, at most 3072 px), and are not made while hovering a look, soft-proofing or showing an overlay, nor over a picture that is not the frame (an unsupported raw's embedded JPEG with another crop). No texture is larger than the GPU allows. Spots and Auto Mask brushes work while what a window reads fits one render (8192 px along an axis, 4096 in the browser): the window grows to hold the spot and its source patch, or the dabs near it, and if that would take more (a clone from far away at 8:1) the loupe keeps the whole-frame render instead, and spots with an automatic source pick it from a fixed-size render of the whole frame (so a window's choice can differ from a whole-frame render's, which picks at its own resolution). The wide stages (dehaze, clarity, highlights/shadows base) and Refine Edges on masks read a window margin capped at 768 px; measured on hard-edged blocks that costs at most a few levels at 8:1. the histogram and Navigator come from the whole-frame render |
+| LR-VIEW-ZOOM | Zoom & pan | P0 | 🟡 | `cmd:view.zoomFit`, `cmd:view.zoom100`, `cmd:view.zoomIn`, `cmd:view.zoomOut`, `cmd:view.zoomToggle`, `cmd:view.clickZoom`, `cmd:view.navigate`, `crates/ui-egui/src/panels/detail.rs`, `crates/ui-egui/src/headless.rs` | continuous native pinch / modifier-wheel zoom (Fit–800%, anchored under the pointer), two-finger scroll / touch pan with edge limits; Detail (including tools, Before/After and full-screen preview), Compare and Reference; fractional percentages preserve small gestures; image, picker and targeted-adjustment cursors stay steady between gesture events; keyboard steps 25–800 % (not 6–1600 %); Fill in the bottom bar; click / Z / Space eases to the chosen ratio (1:1 default, 2:1, 3:1, 4:1, 8:1). The loupe's cost follows the canvas, not the zoom (issue #323): the whole frame is rendered at about canvas size (at most the 2560 px preview source level unless a larger Preview size is chosen, and never above the GPU's texture side), and once the view is zoomed well past that (more than 25 %) the window on screen is rendered too, cut from a frame of at most the photo's own pixels, so beyond 100 % the GPU magnifies it (`region::plan`, `RenderRequest::window`, `Session::region_job`, `crates/ui-egui/src/region.rs`); 1:1 and 2:1–8:1 are sharp at any photo size. A slider tick takes 7–12 ms on a 1400 × 900 canvas and 11–25 ms on a 2800 × 1700 one at 100 % and 400 % of a 24 MP photo (M1 Max, GPU, `profile_slider_drag`); about 20–25 ms at 100 % on the big canvas is over a 16 ms frame. A continuous pinch holds the render sizes (and the windows there are) until it is quiet. A fit view on a canvas up to about 3200 px (a Retina Mac) never decodes the original; larger displays get a window at fit, which does. Modes that draw another picture over the loupe (mask overlays, soft proofing) and photos whose spots or Auto Mask strokes read too far for one window (only photos longer than 8192 px) render the whole frame at the size it is drawn, as before windows; while a preset or profile is hovered, or over a picture that is not the frame (an unsupported raw's embedded JPEG), the whole frame stays canvas-sized. Before/After (side by side, stacked, `\\` and both wipes) gets the same windows on both sides, clipped to their pane or side of the line. Windows render on the GPU like whole frames (equivalence-tested against the CPU), in the browser build too (windows no wider than its GPU's texture side, at most 3072 px), and are not made while hovering a look, soft-proofing or showing an overlay, nor over a picture that is not the frame (an unsupported raw's embedded JPEG with another crop). No texture is larger than the GPU allows. Spots and Auto Mask brushes work while what a window reads fits one render (8192 px along an axis, 4096 in the browser): the window grows to hold the spot and its source patch, or the dabs near it, and if that would take more (a clone from far away at 8:1) the loupe keeps the whole-frame render instead, and spots with an automatic source pick it from a fixed-size render of the whole frame (so a window's choice can differ from a whole-frame render's, which picks at its own resolution). The wide stages (dehaze, clarity, highlights/shadows base) and Refine Edges on masks read a window margin capped at 768 px; measured on hard-edged blocks that costs at most a few levels at 8:1. the histogram and Navigator come from the whole-frame render |
 | LR-VIEW-NAVIGATOR | Navigator mini map | P1 | ✅ | `cmd:view.navigator`, `crates/ui-egui/src/panels/detail.rs` | shown while zoomed (bottom right); click/drag pans |
 | LR-VIEW-BEFOREAFTER | Before / after | P0 | ✅ | `cmd:view.showOriginal`, `cmd:view.beforeAfter`, `cmd:view.beforeAfterSplit`, `cmd:view.beforeAfterTopBottom`, `cmd:view.beforeAfterSplitTopBottom`, `cmd:beforeAfter.setBefore`, `cmd:beforeAfter.copyAfterToBefore`, `cmd:beforeAfter.copyBeforeToAfter`, `cmd:beforeAfter.swap`, `cmd:beforeAfter.resetBefore` | all four layouts; before = the import state (defaults + import preset) or a chosen history step / version / the current settings; copy and swap (View → Before/After Settings, History row menu). The chosen before lasts for the session |
 | LR-VIEW-COMPARE | Compare two photos | P1 | ✅ | `cmd:view.compare`, `cmd:compare.swap`, `cmd:compare.makeSelect`, `crates/ui-egui/src/panels/compare.rs` | select / candidate, synced zoom + pan, arrows move the candidate; no zoom-link toggle |
@@ -480,7 +480,7 @@ under the UI command that wraps it (a no-op error); the UI command now wins.
 | LR-PREF-LOCALSTORAGE | Storage & cache | P1 | ✅ | `cmd:library.preferences`, `cmd:library.clearPreviews`, `cmd:library.compact` | thumbnail cache size + clear in Settings → Performance; library location + Open Library… in General |
 | LR-PREF-ACCOUNT | Account | OOS | 🚫 | | |
 | LR-PREF-INTERFACE | Interface options | P1 | ✅ | `cmd:app.settings` | filmstrip names/badges, grid badges (auto/always/never), square-grid names, navigator, info overlay |
-| LR-PREF-PERFORMANCE | GPU / performance | P1 | ✅ | `cmd:app.gpu`, `cmd:app.memoryBudget`, `cmd:app.settings` | GPU on/off, preview size (Automatic = the whole-frame loupe render at canvas size, up to 2560 px; or a 1600–5120 px size for it; zoomed views add a window at up to 100 %, issue #323), memory budget (a quarter of it is the idle per-view stage caches' share; the open photo's views are counted and reported, not trimmed), thumbnail cache size in Settings |
+| LR-PREF-PERFORMANCE | GPU / performance | P1 | ✅ | `cmd:app.gpu`, `cmd:app.memoryBudget`, `cmd:app.settings` | GPU on/off, preview size (Automatic = the whole-frame loupe render at canvas size, up to 2560 px; or a 1600–5120 px size for it; zoomed views add a window at up to 100 %, issue #323), memory budget (a quarter of it is the idle per-view stage caches' share; the open photo's views are counted and reported, and cut only past four times that, the before side's and the zoom window's first — a Before/After at 1:1 on a 2800 px canvas holds about 1 GB of GPU stages, plus the one 288 MB copy of the original, which `ui.inspect` reports as `sharedSourceBytes`), thumbnail cache size in Settings |
 | LR-PREF-PEOPLE | Face recognition | P2 | ⬜ | | |
 | LR-PREF-WATERMARK | Watermark settings | P1 | ✅ | `cmd:export.savePreset`, `crates/ui-egui/src/panels/dialogs.rs` | set in the Export dialog; kept with Export with Previous and in saved export presets |
 | LR-PREF-SHORTCUTS | Shortcut customisation | — | ✅ | `cmd:app.setShortcut`, `cmd:app.resetShortcuts`, `crates/ui-egui/src/panels/keymap.rs` | an extra (the reference app has no keymap editor): Help ▸ Keyboard Shortcuts (⌘/) lists every command; click a shortcut and press the new keys (Esc cancels), × removes, ↺ restores, Reset All; a key another command had moves over; saved in `ui.json`; native menu accelerators follow. Rating/label digits and secondary keys are listed read-only and yield to a key given to a command |