# RFC: cce-ui core rebuild
 
 **Status:** Draft / proposal
 **Scope:** The core of `cce-ui` — tree ownership, layout, paint, clipping, and animation.
 **Appetite:** Breaking changes are acceptable. Migration is incremental, behind a stable
 `Application` trait, one widget/app at a time. Every crate must continue to build standalone.
 
 ---
 
 ## 1. Why
 
 `cce-ui` works, but it is not one system — it is three half-systems overlaid, with nothing
 arbitrating between them. Concretely, from an audit of the current code:
 
 - **Tree ownership is tripled.** A child lives in a container's own
   `Vec<*mut dyn Element>`, in `ctx.layout_tree.children`, *and* in `ctx.widget_registry`,
   kept in sync by hand in every `add_child`/`clear_children`. All three are raw
   `*mut dyn Element`. `Drop` (`core.rs` `clear_widget_references`) clears focus/context-menu
   refs but **not** the registry or parent/child maps, so stale pointers can linger.
 - **Rendering has three paths with no single owner:** the app's top-level `widgets: Vec<Box<dyn Element>>`
   iteration, parent→child `all_rounded_quads` recursion, and immediate-mode
   `render_widget`/`SectionContext`. Nothing enforces that a widget is drawn by exactly one.
   *(This is the direct cause of the breadcrumb "black rectangle": the breadcrumb was sized
   full-width by its container and drawn a second time under the dropdown.)*
 - **Layout is smeared across five mechanisms:** `LayoutStrategy::allocate` (a child-driven
   bump cursor), `Container::layout`, immediate-mode builders that literally *render twice to
   measure*, `Backplate` (which clips children but does not lay them out), and hand-written
   `set_rect` with absolute screen coordinates in page code. There is no measure→arrange pass
   and no owner of any given rect.
 - **No clip/transform abstraction.** Rect clipping is hand-copied rect-intersection in every
   container's `all_quads`/`all_rounded_quads`/`text_*`. The GPU has no `set_scissor_rect`;
   the only GPU clip is a per-fragment circular test in `shader.wgsl`.
 - **Animation barely exists.** One real helper (`hover_animation`) — copy-pasted into a
   second, *dead* implementation in `UiContext` (`tick_hover`/`get_hover_quad`, zero callers).
   Everything else (button hover/press, `network_opacity`) is an **instant boolean flip or a
   static multiplier, not interpolated**. No `Animated<T>`, tween, spring, or easing library.
   "Keep animating" is a bool hand-propagated up the `tick` chain — miss one link and the
   animation silently freezes.
 
 Every fragility we have hit is a symptom of this. The `Element` trait has grown to ~90
 methods spanning layout, paint, hit-testing, clipboard, tree expand/collapse, and ~12
 `as_*_controller` downcast escape hatches — a god-trait that makes each of the above worse.
 
 ### What is already good (keep it)
 
 - **The frame loop is sound.** Demand-driven redraw with a single `redraw` dirty bool, gated
   by a Wayland frame-callback vsync (`frame_callback_pending`). It idles correctly when
   nothing changes. `tick(dt)` plumbing (clamped `dt`, ~60 Hz dispatch) already exists.
 - **Immediate-per-frame tessellation from a retained tree** is a reasonable bones: widgets are
   long-lived, geometry is re-emitted each frame into one shared vertex buffer and one shader
   pipeline. We are not throwing this out.
 - **The primitive tessellators** (rounded rects with per-corner radii, vectors with caps,
   arcs, circles, relief primitives) are solid and reusable as-is.
 
 ---
 
 ## 2. Goals / non-goals
 
 **Goals**
 1. **Solid** — one source of truth for the widget tree; no raw pointers; no manual multi-store
    sync; no dangling-pointer class.
 2. **Flexible** — one real two-phase layout pass (measure → arrange) *separate from paint*,
    with genuine flex/grow, so layout can be recomputed without re-running paint.
 3. **Dynamic** — a first-class animation primitive; hover/press/opacity/slide/scale/collapse
    as interpolated values; the loop automatically keeps frames coming while anything is live.
 4. **Fast** — GPU scissor clipping, no per-frame debug I/O, and a path to per-subtree geometry
    caching later.
 
 **Non-goals (for this RFC)**
 - Changing the Wayland/wgpu/glyphon backend, the `calloop` loop, or the frame-callback vsync.
 - Changing the KDL config system or IPC.
 - A big-bang rewrite. This lands incrementally behind the existing `Application` trait.
 - Per-subtree tessellation caching — designed-for, but deferred (see §9).
 
 ---
 
 ## 3. Target architecture
 
 A retained scene graph with a clean separation of concerns, borrowing the proven
 Flutter/GPUI/Taffy split of *tree · layout · paint*:
 
 ```
             ┌─────────────────────────────────────────────┐
    update → │  Arena (owns all nodes, keyed by NodeId)     │
             │    Node { parent, children, widget, style,   │
             │           layout_out, anim_state, dirty }    │
             └───────────────┬─────────────────────────────┘
                             │
       ┌─────────────────────┼──────────────────────┬───────────────┐
       ▼                     ▼                      ▼               ▼
   Layout pass          Paint pass             Input pass       Anim tick
  (measure→arrange)   (emit DisplayList)     (hit-test by      (advance
   → LayoutOut rects   under clip/xform       layout rect+z)    Animated<T>)
       │                     │
       │                     ▼
       │            DisplayList → existing tessellators → one vertex buffer
       └── taffy (recommended) or hand-rolled solver
 ```
 
 ### 3.1 Node arena — one source of truth
 
 Replace `Vec<*mut dyn Element>` + `layout_tree` + `widget_registry` with a single arena
 (a `slotmap`/generational-index store). No raw pointers cross frames; code passes
 `&Arena` / `&mut Arena` + `NodeId`.
 
 ```rust
 pub struct NodeId(/* slotmap key: generational index */);
 
 pub struct Node {
     pub parent: Option<NodeId>,
     pub children: Vec<NodeId>,
     pub widget: Box<dyn Widget>,   // the payload (see §3.5)
     pub style: Style,              // layout inputs (flex/size/padding/…)
     pub layout_out: LayoutOut,     // computed rect+transform (written by layout pass)
     pub anim: AnimSet,             // this node's live Animated<T> values
     pub dirty: Dirty,              // LAYOUT | PAINT bitflags
 }
 
 pub struct Arena { nodes: SlotMap<NodeId, Node>, root: NodeId, /* free lists, dirty set */ }
 ```
 
 Why an arena and not `Rc<RefCell<>>` or keeping raw pointers:
 - Generational keys make use-after-free a `None` lookup, not UB. The entire dangling-pointer
   class disappears.
 - One store means no hand-sync of three collections; `add_child`/`remove` touch one place.
 - It sidesteps the borrow-checker tree problem: recursion passes `NodeId` and re-borrows the
   arena, and layout operates on `style`/`layout_out` (not the `widget` payload) so it never
   needs `&mut` to two nodes at once. `SlotMap::get_disjoint_mut` covers the rare cases that do.
 
 Widget **identity** becomes `NodeId` uniformly. Today identity is split: parent/child links
 key on `WidgetId` while `hit_test`/focus/`highlight_color` key on the raw `self` address cast
 to `usize`. Unify on `NodeId`.
 
 ### 3.2 Layout pass — measure then arrange, separate from paint
 
 Introduce a real two-phase pass that runs *before* paint and writes `LayoutOut` per node:
 
 - **measure(constraints) → Size** bubbles intrinsic sizes child→parent. Leaves (text, icons)
   measure their content — text via a glyphon measurement hook so wrapping is correct.
 - **arrange(final_rect)** flows final positions parent→child, writing absolute (or
   transform-local, see §3.4) rects into `layout_out`.
 
 **Decision: hand-roll the solver** (taffy was considered and declined). We own a compact
 measure/arrange engine inside `cce-ui`: `measure(constraints) -> Size` bubbles intrinsic sizes
 child→parent (leaves measure content — text via a glyphon hook), `arrange(final_rect)` flows
 final positions parent→child. Start with the layout primitives cce-ui actually needs —
 row/column with gap + padding, flex grow/shrink, main/cross alignment, and fixed/intrinsic
 sizing — rather than a full CSS flexbox/grid clone. This deletes `LayoutStrategy`, the
 `allocate` bump-cursor, and the render-twice-to-measure `SectionContext` pattern outright.
 
 Rationale for hand-rolling over a dependency: full control over the exact box model (no
 impedance-matching a general CSS engine to our primitives), no external version churn, and a
 solver scoped to what the DE uses. The cost is that we implement and test grow/shrink/wrap
 ourselves — acceptable given the constrained widget set.
 
 The key property either way: **layout is computed independently of paint.** That is what makes
 the breadcrumb bug structurally impossible (one owner writes each rect) and what makes animated
 layout cheap (re-arrange without re-emitting paint).
 
 ### 3.3 Paint pass — one path, a display list
 
 Collapse the three render paths into one. Paint walks the arena in z-order and each node emits
 primitives into a single `DisplayList`, given its computed `layout_out` and a `PaintCtx`:
 
 ```rust
 pub enum Prim {
     RoundedRect { rect: Rect, radii: CornerRadii, color: Rgba, corners: Corners },
     Vector { a: Vec2, b: Vec2, thickness: f32, color: Rgba, cap: LineCap },
     Circle { center: Vec2, r: f32, color: Rgba },
     Arc { .. }, Text { .. },
 }
 pub struct DisplayList { prims: Vec<(ZIndex, Prim)>, /* under active clip/xform */ }
 ```
 
 The `DisplayList` is then fed to the **existing tessellators** (`push_rounded_rect_vertices_corners`,
 `vector_vertices`, `circle_vertices`, glyphon text) → the existing single vertex buffer + shader.
 The backend `render()` stays; only its *input* changes from "call `view`/`view_rounded_quads`/
 `overlay_quads` on the app" to "walk the arena into one display list." The top-level `widgets`
 Vec iteration and the `render_widget`/`SectionContext` immediate path both go away.
 
 ### 3.4 Clip + transform stack
 
 `PaintCtx` carries a clip stack and a transform (translate+scale is enough for v1; a full 2×3
 affine is a small extension):
 
 ```rust
 impl PaintCtx {
     fn with_clip(&mut self, rect: Rect, f: impl FnOnce(&mut PaintCtx));
     fn with_transform(&mut self, xform: Affine2, f: impl FnOnce(&mut PaintCtx));
 }
 ```
 
 - Rect clips lower to GPU **`set_scissor_rect`** (a real render-pass feature we currently do
   not use), deleting the hand-copied rect-intersection in every container.
 - Keep the shader's circular clip for the cases that need it.
 - The transform gives **slide / scale / collapse transitions for free** and lets children use
   local coordinates instead of everyone storing absolute screen coords.
 
 ### 3.5 Trait split — kill the god-trait
 
 Replace the ~90-method `Element` with narrow traits, each a single concern:
 
 ```rust
 /// The payload. Most widgets implement only this + Paint.
 pub trait Widget: 'static {
     fn style(&self) -> Style { Style::default() }        // layout inputs
     fn measure(&self, c: Constraints, ctx: &MeasureCtx) -> Option<Size> { None } // leaves only
 }
 pub trait Paint  { fn paint(&self, layout: &LayoutOut, ctx: &mut PaintCtx); }
 pub trait Input  {
     fn hit(&self, layout: &LayoutOut, p: Vec2) -> bool { layout.rect.contains(p) } // default!
     fn on_event(&mut self, ev: &Event, ctx: &mut EventCtx) -> EventStatus { EventStatus::Ignored }
 }
 ```
 
 - Hit-testing gets a correct default from `layout_out.rect` + z-order, so the manual
   `hit_test`-by-`self`-address code across widgets disappears.
 - The 12 `as_*_controller` downcasts are replaced by typed messages / commands through
   `EventCtx` (an app-defined message channel), not runtime `Any` casts.
 - `Container` stops being special: a container is just a `Widget` with children and a flex
   `Style`; it no longer hand-clips or hand-lays-out.
 
 ### 3.6 Animation — first-class
 
 ```rust
 pub struct Animated<T> { current: T, target: T, motion: Motion } // Motion: Tween(easing,dur) | Spring(k,damp)
 impl<T: Lerp> Animated<T> {
     fn set_target(&mut self, t: T);
     fn tick(&mut self, dt: f32) -> bool;  // returns true while still moving
     fn value(&self) -> T;
 }
 ```
 
 - Hover/press/focus become `Animated<f32>` 0→1 factors, not bools. Buttons interpolate color
   by `lerp(idle, hover, factor)` instead of `if hovered { a } else { b }`.
 - `network_opacity` becomes an `Animated<f32>` that actually tweens.
 - **The loop keeps frames coming automatically:** the arena tracks "any node has a live
   animation." `tick` OR-reduces over the arena's animated nodes and sets `redraw`, so the
   hand-propagated `tick`-bool chain (and its silent-freeze failure mode) is gone.
 - Fold `hover_animation` (and delete its dead `UiContext` twin) into this; the global
   thread-local singleton — which today allows only one animated highlight at a time — becomes
   per-node state, so multiple highlights animate independently.
 
 ---
 
 ## 4. How this kills the classes of bug we have
 
 | Bug class | Fixed by |
 |---|---|
 | Dangling / stale `*mut dyn Element` | §3.1 arena, generational keys, single store |
 | Tree desync (Vec vs layout_tree vs registry) | §3.1 one store |
 | Same widget drawn twice at different rects (breadcrumb) | §3.2 single layout owner + §3.3 single paint path |
 | Overflow bleeding past clip regions | §3.4 GPU scissor + clip stack |
 | Animation silently freezes (missed tick-bool link) | §3.6 arena-driven auto frame requests |
 | One-highlight-at-a-time hover | §3.6 per-node animation state |
 | Per-frame debug I/O in hot path | §7 quick win |
 
 ---
 
 ## 5. Impact on the `Application` trait & clients
 
 The client-facing `Application` trait shape stays as stable as possible so apps migrate one at
 a time. Two viable migration shapes:
 
 - **Adapter (recommended):** the old `Element` widgets keep working via a compatibility shim
   that wraps each in a `Node` and forwards `all_rounded_quads`/`text_items` into a `DisplayList`.
   New/migrated widgets implement the narrow traits directly. Both coexist until the last old
   widget is gone.
 - The `view`/`view_rounded_quads`/`overlay_quads` methods become thin shims over the arena walk
   during the transition, then are removed.
 
 Constraint respected: **each crate still builds standalone** — the new core is entirely inside
 `cce-ui`; clients depend on it by path exactly as today. No `[workspace.dependencies]`.
 
 ---
 
 ## 6. Migration plan (staged; every stage leaves the tree building)
 
 - **Phase 0 — Quick wins (independent of the rebuild).** Fix breadcrumb double-render; remove
   the per-frame `eprintln!` text reconstruction in `render()`; add `set_scissor_rect` for the
   existing clip call sites; delete the dead `tick_hover`/`get_hover_quad` twin. *Ships value now.*
 - **Phase 1 — Arena + identity.** Introduce `Arena`, `NodeId`, `Node`; port `add_child`/tree
   ops onto it behind the adapter; unify identity on `NodeId`. No visual change.
   - **1a — Arena data structure: DONE.** `cce-ui/src/scene/arena.rs` — a generational forest
     arena, generic over payload, added additively (`pub mod scene;`) with nothing wired into the
     live path yet. `NodeId` carries a `NonZeroU32` generation so a handle to a removed node reads
     back as `None` (use-after-free → missed lookup, not UB). Tree ops (`insert`, `append_child`,
     `detach`, `remove_subtree`, `get_pair_mut`, `subtree`/`ancestors` iterators, cycle rejection)
     with 14 headless unit tests, including one that owns and trees real `dyn Element` payloads.
     Full `cce-ui` suite: 83 passing.
   - **1b — Adapter: DONE.** `cce-ui/src/scene/tree.rs` — `WidgetTree`, the arena-backed
     replacement for `UiContext`'s two stores (`widget_registry` + `layout_tree`), keyed by a
     `WidgetId → NodeId` index so the public `WidgetId` API is preserved. Consolidates both maps
     into one generational store; link-only-before-register is modeled as `Entry.ptr: None`. 10
     headless tests against real `dyn Element` payloads (register/overwrite, symmetric+deduped
     link, reparent, link-before-register, symmetric detach, clear_children, clear_all, removal
     staleness, registered-iteration). Not yet wired into `UiContext`.
   - **1b — Live swap: DONE (compile + tests; runtime verification pending).**
     `UiContext.{widget_registry, layout_tree}` are replaced by a single `tree: WidgetTree`. All
     access routed through it: the 5 public methods, the internal direct field reads in
     `context.rs` (event routing, `clear_dirty`, `rebuild_spatial_grid`, `is_widget_visible`,
     `tick`, `is_coordinate_covered`, `is_movable_backplate_at`, `is_widget_at`), the widget-layer
     defaults in `widget/mod.rs`, and the direct pokes in `keybinds_control.rs`/`multi_control.rs`
     (now `link_ids`) and `plate.rs`/`parameters_bg.rs` (now `tree.set_parent(id, None)`). `cce-ui`
     builds clean; **93 tests pass**; all 7 app crates that use the API build unchanged. Still to
     do: **`make run`** cce-files + cce-designer to confirm the symmetric-tree change (below) is
     behavior-safe in paint/event handling.
     - Original swap notes below retained for reference.
 
     Replace `UiContext.{widget_registry, layout_tree}`
     with a `WidgetTree`, and route through it: the 5 public methods apps depend on
     (`register_widget`, `link_ids`, `clear_hierarchy`, plus internally `clear_children_ids`,
     `unlink_child`), the internal direct field reads in `context.rs` (event routing, `mark_dirty`
     walk, `is_widget_visible`, `clear_dirty`, `rebuild_spatial_grid`, coverage/hit tests), and the
     widget-layer defaults in `widget/mod.rs` (`parent`/`set_parent`/`children`/`add_child`/
     `mark_dirty`) plus the direct pokes in `keybinds_control.rs`, `multi_control.rs`, `plate.rs`,
     `parameters_bg.rs`.
     - **One deliberate behavior change to verify at runtime:** the legacy maps are left
       *asymmetric* in a few spots (`set_parent(Some)` writes only `parents`; `plate`/`parameters_bg`
       detach via `parents.remove` only). `WidgetTree` keeps parent/child links symmetric, so
       `children()` — read by paint recursion and event propagation — becomes self-consistent. This
       is almost certainly a latent-bug fix, but it must be confirmed against the running apps
       (cce-files, cce-designer, cce-graph, cce-system-settings) before landing.
     - App compatibility: only `register_widget`/`link_ids`/`clear_hierarchy` have app callers;
       `get_widget`/`get_widget_mut`/`unlink_child` have **zero callers** workspace-wide and can be
       dropped or kept as thin shims.
 - **Phase 2 — Layout pass.** Add `Style` + measure/arrange (taffy). Migrate containers to
   emit `Style` instead of `LayoutStrategy`; delete `allocate` and render-twice measurement as
   containers move over.
   - **`LayoutStrategy`, `allocate` and render-twice: DELETED 2026-10-08.** The last users were
     cce-system-interface's pages (`PageLayoutBuilder` over `AdaptiveGrid`), cce-files' browse
     page (`ColumnLayout`) and the gallery's container-layout exhibit. A section's slot never
     depended on its own height, so `layout::section::PageFlow` places it first and the
     section draws once, in place (`each_section_is_drawn_once`); the browse page is a
     `scene::layout` column; `ContainerLayout` keeps only `layout` / `measure`. Section
     CONTENT is still immediate-mode `SectionContext` — porting each page's content to box
     trees is not done. Every settings page, cce-files (relief on and off, save mode) and the
     gallery drew identically to the pixel before and after, live readings aside.
 - **Phase 3 — Paint unification.** Introduce `DisplayList` + `PaintCtx` (clip/transform);
   route the backend `render()` through the arena walk; retire the top-level `widgets` Vec and
   `render_widget`/`SectionContext`.
 - **Phase 4 — Animation.** Land `Animated<T>` + arena-driven frame requests; convert
   hover/press/`network_opacity`; consolidate `hover_animation`.
 - **Phase 5 — Trait split & cleanup.** Split `Element` into `Widget`/`Paint`/`Input`; remove
   `as_*_controller` downcasts; migrate remaining widgets; delete the compatibility shim.
   - **Approach correction (resolved by experiment).** A *non-breaking supertrait carve-out* of
     `Element` (`trait Element: Paint + …`) turns out to be impossible in Rust here. The structural
     methods the passes need (`rect`, `children`, `set_rect`) are overridden in dozens of widgets
     across cce-ui **and** the app crates (`rect` 31+8, `children` 23+2, `set_rect` 42+4): moving
     them off `Element` breaks every override, and merely *declaring* them on a supertrait makes
     every `elem.children()`/`elem.rect()` call site ambiguous (a supertrait method is always in
     scope on the subtrait). Nor does a blanket "view" `impl<T: Element> Paint for T` let
     `&dyn Element` coerce to `&dyn Paint` — that coercion exists only for real supertraits. So the
     split follows the **adapter** path from §5, not a supertrait split: narrow traits independent
     of `Element`, with `Adapted<W>` bridging a narrow-trait widget into the `*mut dyn Element`
     tree. The compatibility-shim bullet is thus *this* adapter (there was never a discrete legacy
     shim to delete — the earlier migration hung hooks directly on `Element`).
   - **5a — Layout + Paint concerns + adapter: DONE (compile + tests; no runtime surface yet).**
     `cce-ui/src/widget/model.rs` — the independent `Layout` (`layout_style` / `intrinsic_size` /
     `layout_children`) and `Paint` (`color` / `paint` / `clips_children`, where `paint` takes the
     laid-out rect rather than reading a stored one) traits, plus `Adapted<W>`: a wrapper that
     carries the `Widget` base and forwards the `Element` layout/paint methods to `W`'s narrow
     traits. A headless test builds a pure narrow-trait tree (a `Col` container + two `Dot` leaves,
     none of which implement `Element`), wraps each in `Adapted`, and drives it through the
     *existing* `scene::bridge` layout pass and `scene::painter` paint pass — asserting both the
     computed rects and the painted quads. Purely additive: no existing widget or app changes, all
     137 cce-ui tests pass. Runtime verification is N/A until a real widget is migrated onto the
     adapter (nothing in a running app uses it yet).
   - **5b — Input concern: DONE (compile + tests; no runtime surface yet).** `widget/model.rs` —
     the `Input` trait (`hit` / `on_event`, both against the laid-out rect) plus adapter
     forwarding with the RFC's centralizations: `Adapted::handle_event` hit-gates pointer-
     positioned events (`MouseButton`/`MouseWheel`) once, so narrow widgets never carry the
     per-widget "check hit_test first" boilerplate every legacy `mouse_input` override does;
     unconsumed `PointerMove` falls back to the legacy hover bookkeeping, so `base.hovered` and
     the synthesized `MouseEnter`/`MouseLeave` (which re-enter `handle_event` and reach
     `on_event`) keep working; `hit_test` keeps the occlusion (`is_coordinate_covered`) check
     while delegating the geometric test to `Input::hit`. A headless test drives a narrow
     `Clicker` through the *real* `UiContext::propagate_event` router: in-rect click consumed +
     counted, out-of-rect click gated out, hover enter/leave transitions observed on both the
     narrow widget and the base flag. 138 cce-ui tests pass.
   - **5c — First real widget migrated: `ProgressBar`. DONE (runtime-verified, pixel-identical).**
     `widget/display/progress_bar.rs` now implements only `Layout` + `Paint` + `Input`;
     `ProgressBar::new` returns `Adapted<ProgressBar>`, so both construction sites
     (`cce-ui` demo, `cce-test-interface`, incl. `.with_label`) compile unchanged. The migration
     forced the adapter to absorb the legacy surface external render loops actually read, all
     added to `Adapted` in this step: an `all_rounded_quads` **reverse bridge** (the widget's
     `Paint::paint` output converted back to legacy tuples — cce-test-interface renders via this),
     `Paint::corner_style` → `corner_radius`/`rounded_corners` (for style-property painters like
     the demo's `widget_vertices`; transitional, dies with those paths), the detached-label
     convention (`set_rect` inflation + `with_label` builder + content-rect inset),
     `preferred_height` ← `intrinsic_size`, `highlight_quad → None` (narrow widgets own their
     pixels), and `type_name` reporting the *inner* type (layout.rs string-matches
     `"ProgressBar"` for span-full sizing). **Runtime verification:** ran cce-test-interface and
     the demo on the live compositor (via `ccectl center-window` + `grim`); an A/B pixel diff of
     the demo against the pre-migration build showed the two frames identical except a 19×20
     compositor corner artifact — zero differing pixels at any widget. 141 tests pass.
   - **5d — Leaf sweep: `Separator`, `StatusDot`, `UsageBar`. DONE (runtime-verified via the
     settings app's render stream).** New adapter machinery this round: `Deref`/`DerefMut` to the
     wrapped widget (call sites keep `dot.set_status(..)` / `bar.value`); per-prim reverse
     bridges (`Prim::Quad` → `extra_quads`, `RoundedRect` → `all_rounded_quads`, `Circle`/`Arc` →
     `extra_circles`/`extra_arcs`) so apps reading BOTH `all_quads` and `all_rounded_quads` draw
     each prim exactly once; `Input::blocks_backplate_drag`; and the mirrored-by-value-builder
     pattern (`Adapted<UsageBar>::with_colors`) since builders can't flow through `Deref`.
     `Separator` had no `Widget` base (public x/y/w/h fields) — its rect now lives on the adapter
     base, and cce-status-interface's rotation loop was updated to transpose via `rect`/`set_rect`.
     **One deliberate behavior fix:** legacy `StatusDot` emitted **zero** geometry on every render
     path (probe-confirmed — `render_widget` reads only `all_quads`/`all_rounded_quads`, both
     empty for it), so the Processes-page dots were invisible; the narrow `Paint` default emits
     the color quad, and the dots now render (verified in the live app's render dump: 10×10 rects
     in exact status colors). UsageBar verified byte-identical in the same dump (bg+fill rects at
     the exact `with_colors` colors). 147 tests pass; status-interface, system-settings, and
     test-interface all build.
   - **5e — First interactive widgets: `Checkbox` + `Toggle`. DONE (verified end-to-end with an
     injected live click).** New machinery: the legacy polling/value surface on `Input`
     (`take_click`/`take_change`/`value_string`/`set_value_string`/`value` — kept there to avoid
     a fourth bound; dies with RFC §3.5 typed messages); `Input::opens_context_menu` (the adapter
     routes a hit right-press to `UiContext::handle_right_click`, which ctx-less `on_event`
     can't); `Layout::inline_label` (Checkbox/Toggle draw the label inside their rect — no
     `set_rect` inflation/content inset, matching the legacy `label_offset` type-name special
     cases); `Paint::{solid_border, widget_font, sync_label}` (transitional forwards);
     prim-derived `text_labels` for inline-label widgets (one paint source feeds every text
     path) with a base-label fallback replica for detached ones; `as_any` now exposes the *inner*
     widget so legacy `downcast_mut::<Checkbox>()` sites keep working; `Drop` on `Adapted`
     clears the global focus/context-menu refs (bounds moved onto the struct for this);
     `Debug`/`Clone` derives; and an inherent `Adapted::set_label` that shadows
     `Control::set_label` (which writes only the base and left self-painted labels stale —
     caught by a test; `Control` impls override to route here). Both widgets track
     `hovered`/`focused` from the forwarded `MouseEnter`/`MouseLeave`/`FocusIn`/`FocusOut`
     events — the state that becomes `Animated<f32>` in §3.6. In-crate consumers updated
     (`json_layout` direct Element calls, `multi_control` enum variant, `parameters_bg` field);
     app repos updated (system-settings network+notifications, data-editor, layout-interface
     field types — construction sites unchanged). **Verification:** cce-test-interface pixel-
     diffed 0 against the pre-migration baseline, and a `wlrctl`-injected click on the live
     compositor flipped the Toggle's bordered half on-screen — the full input path through the
     adapter exercised for real. 152 tests pass.
   - **5f — `Button` (widest-radius widget: 19 app files + 8 in-crate). DONE.** The press/release
     contract forced an adapter refinement: **presses stay hit-gated, releases now flow ungated**
     — a press-tracking widget must see the release wherever the cursor ended up to commit
     (in-rect → `take_click` + `on_click_cb`) or cancel, exactly the legacy `mouse_input`
     contract (pinned by a router-level test incl. out-of-rect cancel). Also added:
     `Layout::layout_ignore` and `Input::set_selected` forwards. The model ports the per-kind
     color matrix (Primary/Reset/ListRow/CopyIcon × pressed/hovered/selected + bg overrides),
     SVG icon quads, per-kind label justification/fonts, and the Phase 2b `intrinsic_size`; the
     9 by-value builders are mirrored on `Adapted<Button>` (`with_label` comes from the generic
     + `sync_label`). In-crate consumers fixed (multi_control, keybinds_control, ramp, treelist,
     parameters_bg fields; json_layout + demo now call `take_click` on the box instead of
     concrete downcasts); ~11 app repos updated (field types + raw-cast→`as_ptr_mut` cleanups).
     **Verification:** full workspace (minus compositor, which doesn't use widgets) builds;
     154 cce-ui tests + cce-cloud's json_layout hover-simulation test pass; test-interface
     pixel-diffs cursor-only vs the 5e baseline; a live hover A/B against the stashed legacy
     build showed the identical fill pixel (the inert hover on that page is pre-existing app
     behavior, not a regression).
   - **5g — `Label`. DONE.** Text lives on the model, emitted as a `Text` prim; the adapter's
     prim bridge serves every legacy text path. Added the generic synced `Element::set_text`
     override on `Adapted` (same trap as `set_label`: the trait method wrote only the base and
     left the painted text stale — live-updating labels like system-info's CPU readouts hit it
     constantly). Builders mirrored; three app repos' field types updated. Workspace builds;
     155 tests pass; test-interface diff vs the post-Button baseline has a 0x0 bbox at 2%
     threshold (sub-perceptual blend noise only).
   - **5h — `Slider` + `RangeSlider`, and the event-capability layer. DONE.** Introduced the
     RFC §3.5 **`EventCtx`** (`on_event(&mut self, event, &mut EventCtx)`): content rect, widget
     id, `request_focus()` (readout edit mode), and a transitional `ui: Option<&mut UiContext>`
     for the legacy scroll-gesture gating. Added `Input` drag hooks
     (`draggable`/`is_dragging`/`drag_begin`/`drag_update`/`drag_end`, rect-carrying — hosts
     drive drags by direct call) and — critically — **direct-dispatch overrides**: hosts call
     `mouse_input`/`mouse_wheel`/`keyboard_input`/`focus`/`unfocus` directly on widgets, and
     without adapter overrides those hit the inert Element defaults (a latent 5e/5f regression:
     treelist's add-key button and parameters_bg checkboxes were deaf on that path — now routed
     into `handle_event`). `Layout::inflates_label_rect` distinguishes ProgressBar-style rect
     inflation from Slider-style label-eats-into-rect. `text_labels` is now prim-derived PLUS
     base fallback (sliders paint readout text AND have a detached label). **Found the hard
     way:** hosts under-size labeled sliders, so legacy content height went NEGATIVE and the
     flipped quads still rasterized — `content_rect` must not clamp at zero or tracks vanish
     (pixel-diffed to 0 vs baseline after the fix). Slider geometry consolidated into one
     `geom()` helper (legacy re-derived it in five places). 154 tests pass; workspace builds.
   - **5i — Display leaves + `Spinbox`: InfoBox, FontPreview, Sidebar, Splitter, Panel, Spinbox.
     DONE.** New adapter machinery: `Input::drag_reposition` (self-moving widgets — Panel,
     Splitter — return a new origin; the adapter applies it to the base rect the model can't
     reach), `Input::set_drag_bounds`, and `Layout::detached_label_inset` (the legacy
     `Control::control_label` +4px x-offset that the default `text_labels` path lacked — caught
     as a 4px label shift in the pixel diff, fixed to a 1×1-pixel residual). Spinbox ports the
     sub-zone hover (-/+ buttons) into `PointerMove` handling, display-click edit mode with
     cursor placement + `request_focus`, decimals/unit value formatting, and drops its vestigial
     raw-pointer parent/children fields. Skipped for later: `PreviewState` (needs per-label fonts
     on the `Text` prim), `StatusBar` (bigger custom surface), Float3/LayoutPreview (time).
     156 tests pass; workspace builds; test-interface pixel-diff vs the 5h baseline: 1 pixel.
   - **5j — `InteractiveListItem`. DONE (verified in the live settings render dump: service-row
     titles/subtitles + themed overlays through the prim bridge).** Button-pattern press/release
     with themed selected/hover/press overlays. `StatusBar` was surveyed and DEFERRED: it is
     parent-coupled (reads its parent's backplate state/rect/radius at paint time and owns
     `parent`/`set_parent`) — that belongs with the container/children design, not the leaf
     recipe. Also still pending from the leaf tier: Float3, LayoutPreview (time), PreviewState
     (needs per-label fonts on `Prim::Text`).
   - **5k — Controller capabilities + first controller widgets: `Breadcrumb`, `Node`. DONE.**
     The controller tier was blocked because `Element` is implemented exactly once (for
     `Adapted<W>`), so a migrated widget couldn't re-expose its `as_*_controller` downcasts.
     Resolution: transitional **capability hooks on `Input`** (`menu/graph/spreadsheet/path/
     param/geom _controller[_mut]`, default `None`) that the adapter forwards the `Element`
     downcast pairs to — a controller widget returns `Some(self)`. This is the pragmatic half of
     the §3.5 "typed messages" bullet: the controller *traits* are already the typed surface;
     what dies with `Element` is reaching them through the god-trait (end state: hold the
     concrete `Adapted<W>` or a `&dyn XController` directly). Also added:
     `EventCtx::open_context_menu` (Breadcrumb records the right-clicked segment *before* the
     shared menu opens — `opens_context_menu` can't express work-before-menu), an
     `Input::copy_path` forward (context menu "Copy Path"), and an `Adapted::on_cursor_moved`
     override — another direct-dispatch entry (cce-files drives breadcrumb hover through it)
     routing the raw move to `on_event` with the base hover bookkeeping as fallback.
     `ScrollController` was deleted outright: zero implementors and zero live callers
     workspace-wide (only a dead cce-designer helper, removed there). **Breadcrumb** (first
     controller widget, live in cce-files + cce-designer) verified on the live compositor:
     idle/hover captures pixel-identical to the stashed legacy build at the widget, and a
     breadcrumb-segment click navigates correctly through the direct-dispatch
     `mouse_input → on_event → path_click` chain. **Node** (ParamController + GeomController,
     self-moving drag with grid snap via `drag_reposition`) has no live constructors in any app
     repo — compile + router-level tests only. 160 tests pass; all 19 client crates build.
   - **5l — `Spreadsheet` + the tick/scroll adapter surface. DONE (live-verified in the
     designer: startup and pane-open captures diff 4px/32px vs the stashed legacy build, all in
     a one-pixel bottom-edge blend strip).** New `Input` surface: `tick(dt, rect)` +
     `wants_tick` (inertial scroll — hosts broadcast `Element::tick` per frame; §3.6
     `Animated<T>` eventually replaces this), `scrollable` (→ `is_scrollable`), and
     `draggable` now takes the laid-out rect (scroll widgets are draggable only while content
     overflows). **`Adapted` now owns real visibility**: the `Widget` base carries none and the
     legacy `Element` defaults are no-ops, so each hideable widget stored its own flag; the
     adapter stores it once, gating `hit_test` (hosts broadcast wheel/press dispatch and rely
     on hidden widgets rejecting the hit), direct-dispatch `keyboard_input` (hiding a pane
     doesn't unfocus it), and the `text_labels` bridge. Deliberate fix: `set_visible` on
     migrated widgets now works instead of being silently ignored. Spreadsheet's 7×-duplicated
     scroll/thumb math collapsed into one `geom()` helper; its `paint` deliberately does NOT
     emit the translucent `PARAM_BG` background (the designer draws widget backgrounds itself
     from `color()` + `corner_style` — emitting it again would double-blend).
   - **5m — `Graph` + the legacy dual-geometry escape hatch. DONE (designer startup pixel-diffs
     ZERO; cce-graph diffs to an empty 2%-threshold bbox; cce-files' Graph view verified
     visually + getter-contract unit test — its pixel A/B was blocked by the live session's
     terminal covering the capture region).** Graph's three hosts consume DIFFERENT getters
     (designer: plain `extra_quads` + `extra_circles`; cce-files `render_widget`:
     `all_rounded_quads` with highlight-only `all_quads`; cce-graph: the scene path's
     `paint_self`). The model keeps one geometry generator; `paint` emits the rounded view, and
     transitional `Paint` hooks serve the rest: `serves_legacy_plain_quads`/`legacy_plain_quads`
     (verbatim through `extra_quads`, with the adapter emptying `all_quads` to preserve the
     no-double-draw contract) and `text_bounds` (node names clip to the widget rect — the
     adapter now overrides `text_labels_with_[font_and_]bounds`, replicating the
     scroll-ancestor walk when the hook is `None`). Also: `Input::hit` override for the legacy
     edge-exclusive hit test; ctrl-wheel zoom reads `ctrl_pressed` through `EventCtx.ui`;
     the `Element::paint` register_hovered pre-pass is dropped (shared hover highlight is
     suppressed for all adapted widgets). Discovered en route: the designer's CONTENT pane IS a
     real `Graph` (index 1), and `ContentBg` is a standalone grid background whose
     GraphController impl is mostly stubs — it is NOT a Graph wrapper.
   - **5n — The container concern + first containers: `Switcher`, `ContentBg`. DONE
     (render-dump-verified: cce-system-settings — whose every page lives under the Switcher —
     A/B'd byte-identical across four pages against the stashed legacy build, modulo live
     system data).** The children/tree design, resolved transitional-first: tree links already
     live in `ctx.tree` (Phase 1b), so a container needs only (a) its own child-pointer Vec
     (ctx-less `set_rect` arrangement — the same reason legacy containers kept one) and (b) the
     subtree plumbing every legacy container hand-copied. (a) stays in the model behind new
     `Layout` hooks (`has_container_children`/`container_children`/`child_added`/
     `children_cleared`/`parent_changed`/`adjust_rect`/`arrange_children`/
     `layout_children_ctx`/`child_visible`); (b) moved into the ADAPTER once, filtered by the
     `child_visible` policy: plain-quad aggregation with the shared rounded-bg-skip rule,
     rounded recursion (the Element default this override had been shadowing — a latent
     container blocker), per-kind text aggregation, `get_text_items`/`prepare_text`/`tick`/
     popover recursion, container-style `add_child` (parents the child back), the ctx-carrying
     child layout pass, and `is_child_visible`. `Input` gains `hits_through_children` and
     `gates_presses` (event-proxying containers must see every press — Switcher unfocuses its
     child on an outside click). Event proxying itself stays in the model via `EventCtx::ui`,
     bug-for-bug (including Switcher's double `mouse_input` dispatch while a popover is open).
     `ContentBg` turned out to be a leaf and rode the ordinary recipe. NOT for this path:
     deep-composition containers (Page/Plate embed `Layer`; SectionContainer embeds
     `Container`; Paginator embeds ButtonStrip + `Vec<Page>`) — embedding means migrating the
     base struct inverts the dependency; those dissolve when their hosts move to the scene
     walk (Phase 6), not through `Adapted`.
   - **5o — `MenuBar` migrated; standalone `Menu` DELETED (zero constructors workspace-wide —
     dead code).** MenuBar keeps its legacy `ButtonStrip` EMBEDDED in the model (owned by
     value, driven through `Element` calls; events reach it via `EventCtx::ui`;
     `arrange_children` parents it back to the adapter). New adapter surface: `Paint::popover`
     / `draw_popover` (own dropdowns — the container recursion only covered child popovers),
     `Layout::z_order`, `Layout::tracked_parent` (serves `Element::parent` from the model's
     field — legacy parent-chain styling walks use a DUMMY ctx that tree lookups can't
     answer), `Input::set_modifiers` / `visibility_changed` / `is_focused` (conditional focus:
     the bar holds the global slot only while something is open) + `EventCtx::release_focus`,
     the `as_page_selector` capability pair, and `Paint::corner_style` now takes the laid-out
     rect (corners computed against the parent backplate's edges). Dropped, flagged: the
     never-read glyphon buffer caches and the vertical-mode dynamic `rect()` (`with_vertical`
     has no callers). Verified: 168 tests (full open→click→close roundtrip through real
     adapter dispatch), all four hosts run, live A/B on cce-test-interface pixel-equivalent
     (the strip has zero diffs above the 8% threshold; the File-click-opens-nothing behavior
     there is byte-identical pre-existing app behavior). Designer A/B (was pending on screen
     contention): DONE — startup, params-pane dropdown clicks, circular-pane mode, and a
     View→Circular-Pane `trigger_menu_click` roundtrip all pixel-equivalent vs the f1523ab^
     baseline (all residual diffs are composited-cursor + bottom-status-strip artifacts).
     Driven deterministically via the designer's embedded HTTP API on :3000
     (`{"action":"toggle_circular_pane"}`, `{"action":"menu_click","widget_idx":8,
     "menu_idx":2,"item_idx":2}`); the params-pane Circular-Pane dropdown not opening on
     click is pre-existing app behavior, identical in both builds.
   - **5p — `Dropdown` (first popover widget through the 5o `Paint::popover`/`draw_popover`
     surface). DONE (render-dump + live-A/B verified).** Slider label convention
     (`inflates_label_rect=false`), `Control::control_label`'s +4px via
     `detached_label_inset`, `gates_presses=false` (an open dropdown must see the outside
     press that closes it), `opens_context_menu`, dynamic `z_order` (100 while open), and
     focus parity bug-for-bug: `FocusIn` re-claims the global slot (direct `focus()` callers —
     test-interface), `FocusOut` closes without releasing it. New adapter surface:
     `Layout::intrinsic_measure_width` + an `Element::measure` override on `Adapted`
     (identical to the `Element` default unless a widget opts in — preserves `auto_width`
     measuring, which cce-system-settings sizes its page dropdown through). Parity decisions,
     flagged: the public `parent` field stays direct-write-only (legacy `set_parent` never
     wrote it — Ramp's dummy-ctx `set_parent` calls were silently discarded, so the Ramp
     popover clamp and fade-blend parent color were dormant in production and stay dormant);
     the backplate-concentric corner walk starts from a `parent_changed`-tracked pointer and
     hops field-based legacy `parent(&dummy)` impls (exact for cce-graph's
     Dropdown→Plate→Backplate chain; deep tree-only chains lose the adjustment); the row-rect
     hit expansion is dropped, consistent with every migrated control. Verified: 169 tests,
     full workspace builds, cce-system-settings fonts/notifications render dumps
     content-identical (only the detached label's emission order shifts — adapter appends it
     after the widget's prims), cce-graph startup/open/close live A/B **byte-identical**
     (AE=0 open state on a clean run; run-to-run compositor translucency noise ~15k AE dwarfs
     any residual), cce-fonts + cce-test-interface smoke-run. App sweep: 8 repos (graph,
     text-editor, fonts, layout-interface, system-settings, data-editor, files,
     test-interface) — field types to `Adapted<Dropdown>`, raw casts to `.as_ptr_mut()`,
     two `&mut Dropdown` fn params in cce-files pages.
   - **5q — `TextBox` (widest-surface leaf; the clipboard/selection tier). DONE (render-dump
     + live-A/B verified, data-editor byte-identical).** New adapter surface: the `Input`
     clipboard quintet (`cut_selection`/`copy_selection`/`paste_from_clipboard`/`select_all`/
     `clear_text` — defaults replicate the whole-value `Element` defaults so earlier
     migrations keep their shipped behavior), `Paint::prepare_text` (TextBox's glyph shaping
     is load-bearing: `map_x_to_idx` reads the measured advances), `Layout::hit_row_rect`
     (restores the legacy row-substituted, side-label-inset hit geometry — cce-files'
     save-name box relies on row hits; earlier migrations' drop of it stands, opt-in),
     `Layout::adjust_row_rect` + `Layout::rect_assigned` (the width/max-width clamp on both
     rect paths; ungated scroll re-clamp on every `set_rect`), `Input::tracks_base_focus`
     (legacy TextBox's `focus()` never set the base flag — its detached label must not color
     as focused), and **`Paint::legacy_focus_highlight`**: the shared focus-highlight overlay
     the adapter suppresses for every migrated widget is re-enabled per-widget — legacy
     TextBox kept the `Element` default, and the focused editor's primary-tint wash
     (data-editor's teal editing surface) is real legacy behavior. Found the honest way: the
     first A/B came back 1.4M pixels apart; after restoring the overlay (replicated
     byte-for-byte in `Adapted::highlight_quad` + the `all_quads`/`paint_self` inclusion
     points), data-editor startup AND focused-editor states are **byte-identical (AE=0)**.
     The asymmetric legacy render split is preserved (non-rounded `extra_quads`: full-width
     background + disabled special-case; rounded `all_rounded_quads`: side-label inset, no
     disabled branch). Flagged approximations: releases re-check plain-rect containment
     (legacy hit-gated them through the row-substituted test); wheel is now hit-gated by the
     adapter (legacy hosts dispatched it to the hovered widget themselves). Verified: 169
     tests, workspace builds, settings fonts/processes dumps content-identical (processes
     modulo live PID/CPU data), cce-data-editor live A/B byte-identical. Sweep: 9 app repos
     (authenticator, data-editor, display-manager, email, files, fonts, layout-interface,
     system-settings, text-editor) + 5 in-crate embedders (treelist, scrolling_list,
     keybinds_control, multi_control's `InstancedWidget` variant, parameters_bg).
   - **5r — `Paginator` (value-embedded container: ButtonStrip + Vec<Page>). DONE (live-A/B
     verified, cce-layout-interface byte-identical across four states).** The model owns both
     embedded legacy widgets by value and proxies events to them (strip first, draining its
     click into the selection; then the selected page while not hidden); the container
     concern serves them via `container_children`/`child_visible` (strip always, selected
     page only), and `arrange_children` is the legacy `set_rect` body. New adapter surface:
     **`Layout::register_embedded_children(host_id, ctx)`** — legacy `tick`/`layout`
     re-registered the strip + pages into the ctx registry every frame, and that registration
     is load-bearing (the spatial grid is rebuilt from registered widgets; the registered
     strip is what blocks backplate drags over the sidebar —
     `backplate::tests::test_paginator_blocks_backplate_drag`); the adapter calls it from
     `Element::tick` and `Element::layout`, the legacy cadence. **`Paint::
     aggregates_child_extra_quads`** — legacy container `extra_quads` served the CHILDREN's
     chrome only, while the widget's own background quad lived in `all_quads` alone;
     cce-mail and cce-layout-interface render the tab column through `extra_quads` over
     their own backgrounds (emitting the bg there would double-blend), and cce-test-interface
     renders through `all_quads` (dropping the bg there would blank it). The adapter's
     `all_quads` now draws own prims from a shared `own_plain_quads()` helper instead of
     `extra_quads()` (identical for every prior migration) so the two views never
     double-serve. **`Paint::forwarded_highlight(ctx)`** — legacy `highlight_quad` forwarded
     to the strip's (the hovered-tab tint cce-layout-interface draws by calling
     `highlight_quad` directly); served only through that getter, kept out of
     `all_quads`/`paint_self` (gated on `legacy_focus_highlight` now — where the strip's own
     aggregation already carries it, as legacy container `all_quads` overrides did).
     `PageSelector` + `MenuController` ride the existing `Input` capability hooks
     (cce-test-interface reaches `sidebar_w` through an `as_page_selector()` downcast on
     `dyn Element`). Ported bug-for-bug though unused workspace-wide: the `pages` container
     surface (`add_widget_to_page`/`set_pages`/…) — every app manages page content itself
     keyed on `selected_page()`. Verified: 171 tests, all four consumer apps build,
     cce-layout-interface live A/B **byte-identical (AE=0)** on idle, File-tab hover,
     Page-tab click, and Page-selected hover (exercises extra_quads aggregation, highlight
     forwarding, labels, and the click→selection→page-switch path);
     cce-test-interface A/B equals its launch-to-launch noise exactly (same 7.5k-px bbox —
     animated waveform phase; the `all_quads` gallery path contributes zero residual).
     Sweep: 3 app repos re-typed to `Adapted<Paginator>` (email, files, layout-interface);
     test-interface's `Box<dyn Element>` gallery needed no change.
   - **5s — `ParametersBg` (the designer's parameter panel; the last real container). DONE
     (live-A/B verified, cce-designer pixel-equivalent across four states).** The model caches
     its laid-out rect via `Layout::rect_assigned` (all row geometry derives from it — the
     TextBox pattern; `arrange_children` is the legacy child-stacking tail, visible-gated by
     the adapter exactly as legacy gated it), keeps the value-owned per-row widget vecs
     (mostly `Adapted<W>` already) plus the raw-pointer `children` list on the 5n container
     hooks, and ports the full bespoke event surface into `on_event` arms: the every-press
     dispatch chain (`gates_presses=false` — the panel consumes every left press, scrollbar
     thumb drag math, popover-first ordering, per-row-type dispatch with value commit-back,
     the inline emacs-flavored code editor) and the host-driven drag surface on the `Input`
     drag hooks. New adapter surface: **`Paint::serves_legacy_labels` /
     `legacy_labels_with_font_and_bounds(rect, ctx)`** — the text sibling of the 5m
     dual-geometry hatch: the standard bridge gives every own label ONE font and ONE clip
     rect, but this panel assigns them PER LABEL (viewport clip everywhere, code-box clip +
     monospace inside a code row); served as a full replacement, children included. Also
     **`EventCtx::widget_addr()`** — the wheel arm's occlusion check
     (`is_coordinate_covered`) keys on the adapter's address, which `on_event` couldn't
     reach. Reuses 5m's plain-quad hatch for the designer's raw `extra_quads` render path
     (row chrome, section borders, code cursor, scrollbar — with the panel's translucent
     PARAM_BG plate deliberately NOT emitted: the designer draws it from
     `color()`/`corner_style`, the 5l double-blend trap). `window_runner`'s
     `get_child_widget_for_quad` downcast keeps working unchanged (`as_any` exposes the
     inner type; the 9 pub sub-widget fields stay pub). Flagged approximation: the legacy
     scrollbar-press called `self.focus()` (base flag only — nothing reads it; the highlight
     keys on the ctx focus slot and the designer tracks panes by index); dropped.
     Verified: 175 tests (4 new: controller roundtrip + row layout, checkbox/code-editor
     commit flows, hatch split, overflow scrolling), cce-designer builds with ZERO app
     changes (`Box::new(ParametersBg::new())` coerces), live A/B across startup /
     wheel-scrolled pane / dropdown-row click / second click identical except the bottom
     status strip — calibrated as launch-to-launch live-data noise (same 1447×21 bbox,
     77 px between two launches of the SAME baseline vs 81 px old-vs-new). The wheel state
     changed 107k px within a build and matched across builds, so the event path is
     genuinely exercised. (The params-pane Circular-Pane dropdown not opening on click is
     the pre-existing app behavior recorded in 5o, identical in both builds.)
   - **5t — the deferred leaves: `Float3`, `LayoutPreview`, `PreviewState`, `StatusBar`.
     DONE (live-A/B verified: cce-files byte-identical ×3 states, cce-designer params pane
     byte-identical ×3 states, cce-system-settings byte-identical whole-window).**
     Consumer survey first (it reshaped the work): every external `Float3` grep hit is the
     MATH type (designer `GAttribute::Float3` / wgpu `Float32x3`) — the widget's only
     consumer is ParametersBg, whose pub-field reach (`values`/`mins`/`maxs`/`edit_buffer`/
     `editing_idx`) flows through `Deref` unchanged; LayoutPreview has ZERO consumers
     (definition + re-exports only); PreviewState is cce-files' preview pane; StatusBar has
     six construction sites across five apps plus the demo.
     **Float3**: rect cached via `rect_assigned` (`get_row_rects` is pub API with no rect
     param), readout-edit + track-drag into `on_event`/the drag hooks, the readout click's
     legacy `focus::set_focused(self)` rides `EventCtx::request_focus`, commit-on-unfocus via
     the FocusOut arm. **LayoutPreview**: mechanical (paint = quads + text; the duplicated
     SimNode match collapsed into one helper). **PreviewState**: the widget the 5s labels
     hatch was built for — canvas quads flow from `paint`, canvas labels (per-label
     monospace for content lines) through `serves_legacy_labels`; plain `text_labels` stays
     EMPTY like legacy (emitting the text as prims too would double-render under container
     aggregation — the legacy scene path showed no text either, preserved). The adapter
     gains a blanket `impl Default for Adapted<W: Default>` (cce-files constructs it via
     `Default`); the app's struct-literal update became field mutation (the private cached
     rect can't ride functional-update syntax — and now survives updates instead of zeroing
     until the next layout pass). **StatusBar**: the MenuBar parent-coupling pattern
     (tracked parent, backplate-aware color/text-color/blur, corners-against-parent at the
     parent's radius) plus ONE new hook — **`Paint::text_items()`**: pre-shaped glyphon
     buffers for the legacy `get_text_items` path, which prim-derived text cannot serve (it
     returns borrows of widget-owned buffers); cce-status-interface drives the bar by hand
     (`prepare_text` → `get_text_items` into its own paint) and data-editor/system-settings
     host it as a Backplate child. `inline_label` keeps `Element::set_text`'s base-label
     write from leaking a detached label. Deliberately preserved asymmetry: NO
     `widget_font`, so the container text path keeps rendering the bar's text in the
     default font while the buffer path uses the statusbar font, exactly as legacy.
     Verified: 177 tests (Float3 readout/drag flow; StatusBar manual-host pipeline —
     covering the path of the one app, cce-status-interface, not A/B'd live: it is the
     user's session status bar). Sweep: 4 app repos (files: field + literal→mutation;
     data-editor: field + one raw `*mut StatusBar` cast → `as_ptr_mut()`; system-settings +
     status-interface: field types).
   - **Phase 5 widget migration COMPLETE.** Everything remaining on `impl Element` is
     embedded-base machinery by design: the containers
     (Layer/Container/Page/Plate/Backplate/ScrollBox/List/…) dissolve via Phase 6 scene
     adoption instead. Then delete `Element` + `Adapted` once the last widget is across, at
     which point the `as_*_controller` pairs and the `Input` capability hooks die together
     (callers hold concrete types or `&dyn XController`).
 - **Phase 6 — Per-app migration.** Move each `cce-*` app onto the new core; delete legacy paths
   once the last app is across. **Definition of done per app:** the whole frame — geometry AND
   text — is one `Application::display_list()` (+ `display_list_text()`), layout runs through
   the scene solver where the app has a real tree, events reach widgets through routed dispatch
   rather than hand-rolled per-widget loops, and no embedded-base container
   (Layer/Page/Plate/Backplate/ScrollBox/List) is load-bearing. Seven apps already feed
   geometry through `display_list()` (colors, data-editor, files, fonts, graph, text-editor,
   system-settings) — their remaining gaps are text, layout, and events.
   - **6a — display-list text. DONE (live-A/B verified via cce-notifier).** `Prim::Text` now
     carries `font: Option<String>` + `bounds: Option<[f32;4]>` (`PaintCtx::text_with`; plain
     `text` emits None/None), and the backend renders a list's Text prims through the glyphon
     pass — shaped via the shared `get_text_buffer` cache, clipped to the item clip ∩ the prim
     bounds, held in `EngineState::dl_text_items` so the `TextArea`s can borrow the buffers.
     **Opt-in via `Application::display_list_text()` (default false)**: the seven Phase 3
     adopters' lists already carry Text prims that those apps ALSO push as `TextItem`s —
     rendering both would double-draw; each app flips the flag when it stops pushing its own.
     `Application::view`/`text_items` gained no-op defaults so a fully migrated app implements
     neither. Known limitation (scoped out, not a bug): the legacy `text_areas`
     popover-occlusion clip is not applied to display-list text yet — a popover plate does not
     hide list text beneath it (text draws after all geometry); apps with popovers keep their
     own text path until that lands. The 5s/5t labels hatches become expressible as prims once
     hosts consume lists directly.
   - **6b — `cce-notifier` (first app fully on one path). DONE (live-A/B on a private D-Bus
     session: text/accent pixel-identical; 60-px residual is compositor translucency noise in
     the alpha-0.9 background region).** The whole frame is one display list (accent quad +
     three `text_with` prims in the configured bundled family); deleted: the app-side
     `FontSystem`, the `TextItem` cache, `rebuild_layout`, and the scale/rebuild bookkeeping.
     Non-interactive, so no event surface. This is the reference shape for a minimal Phase 6
     app.
   - **6c — `cce-wallpaper` + `cce-screenaver` across; `display_list` gains `(size, scale)`.
     DONE (wallpaper live-A/B AE=0; screensaver background-fill verified live, sim quads are
     the same mechanical loop).** The Phase 6 frame entry point now receives the frame's
     logical size and HiDPI scale like `view` did (fullscreen apps size geometry from it);
     mechanical sweep across the eight implementors. Both apps' dead `TextItem` caches
     deleted.
   - **6d — the paint walk carries per-widget fonts + clip rects. DONE (179 tests; cce-graph
     live A/B shows zero structural diff — all residual below the 8% translucency-noise
     amplitude).** `Adapted::paint_self` no longer forwards `Paint::paint`'s plain Text prims:
     it re-emits the geometry verbatim (through the ctx so the walk's offset/clip apply
     once) and serves text as `text_with` prims from the SAME views the standard text bridges
     use — `own_labels_with_font_and_bounds` (prim text + detached base label, `widget_font`,
     `text_bounds` or the scroll-ancestor clip) or the 5s per-label hatch verbatim (caveat
     noted in-code: the hatch contract includes raw container children). This makes a
     `paint_tree` display list's text renderable-correct for migrated widgets, which is the
     precondition for the seven adopters flipping `display_list_text`. Found and recorded on
     the way: the LEGACY `Element::paint_self` default drains the child-aggregating
     `text_labels` for legacy containers, so scene-path text double-emits under the walk for
     trees that still contain Layer/Page/etc. — invisible today (text prims unrendered
     without the opt-in), but it means an app can only flip `display_list_text` once its
     tree is embedded-base-free. Consistent with the dissolution plan; revisit per app.
   - **6e — `cce-colors` flips `display_list_text` (first of the seven adopters). DONE
     (live-A/B'd; slider drag re-verified via wlrctl).** The whole frame is one
     `display_list()` — the rebuild check moved off the deleted `view`/`view_rounded_quads`
     overrides, `rebuild_layout` keeps the flat PageContent text tuples and the list emits
     them as `text_with` prims; deleted: the app-side `FontSystem`, the `TextItem`
     assembly, and the `CCE_LEGACY_PAINT` fallback. The A/B exposed a PRE-EXISTING runtime
     bug this fixes: the app shaped its `TextItem`s with its own
     `create_font_system_with_system_fonts()`, whose fontdb face IDs don't resolve in the
     engine's render `FontSystem` — every `font: None` label (slider names, channel
     values, hex readout) was INVISIBLE at runtime in the baseline (only the bundled-font
     button labels survived). Shaping through the engine's FontSystem (the dl-text path)
     is what makes the text render at all. Note for the remaining adopters: an app-side
     `FontSystem` is not just dead weight, it is a live font-resolution hazard — check
     each app's text for the same silent invisibility before trusting its baseline
     capture. cce-colors' safety: no popovers, no paint_tree (flat-list bridge), so
     neither the popover-occlusion gap nor the 6d embedded-base double-emit applies; its
     root `Backplate` remains for legacy layout/events (full definition-of-done still
     pending events + layout).
   - **6f — `cce-files` flips `display_list_text`. DONE (live-A/B pixel-identical, AE=0;
     dropdown popover + breadcrumb context menu re-verified live).** Same mechanical shape
     as 6e: rebuild check into `display_list()`, `rebuild_layout`'s text tuples emitted as
     `text_with` prims, `view*`/`CCE_LEGACY_PAINT`/TextItem assembly deleted. Two deltas
     from colors: the app-side `FontSystem` STAYS (TextBox/List `prepare_text` measurement
     still needs it — it's `create_font_system()`, bundled-only, so no 6e invisibility
     hazard), and popover occlusion needed nothing from the engine — cce-files folds
     popover/context-menu/dialog occlusion into each text's clip bounds app-side
     (`occlude_against`, both axes), and those bounds ride along as prim bounds. That's
     the general pattern for flat-list-bridge apps with popovers: the engine's missing
     dl-text occlusion pass only blocks apps that rely on the DEFAULT `text_areas`
     popover clamp (`ui_context().active_popovers`), e.g. widget-tree apps whose popovers
     register through `register_popover`.
   - **6g — popover occlusion for display-list text lands; `cce-system-settings` flips
     `display_list_text`. DONE (live-A/B pixel-identical, AE=0; page-dropdown popup, page
     switch, service-list scroll, live process refresh exercised).** Engine:
     `popover_occlusion_clamp` extracted from the default `text_areas` mapping and applied
     to `dl_text_items` in `render()`, driven by `ui_context().active_popovers` — the
     display_list_text known limitation is gone; apps whose popovers register through
     `register_popover` (which `render_widget` does for any open `popover_rect`) can flip.
     App: same mechanical shape as 6e/6f, with the app-side `FontSystem` kept for
     button-label width measurement (centering) and the search-match highlight rect, and
     the wheel fast-path mutating tuple y/bounds in place exactly as it did TextItems.
     Popovers/context menu were never main-surface here: they draw on the engine's
     xdg-popup surface via the `render_popovers` collector, which is orthogonal to the
     flip.
   - **6h — `cce-text-editor` flips `display_list_text`; `Paint::text_font` lands. DONE
     (live-verified: frame matches baseline modulo a uniform ~2px baseline shift from
     engine line-height shaping; File-menu popup + occlusion of editor text beneath it;
     click cursor placement identical to baseline).** The FIRST app rendering scene-walk
     text prims — its tree (Adapted Dropdown + Adapted TextBox, no embedded-base
     containers) was exactly the 6d-safe shape, unblocked by the 6g occlusion clamp.
     Engine: `own_labels_with_font_and_bounds` split into a parameterized helper — the
     tuple getters keep `widget_font` for every label (unmigrated apps byte-identical),
     while the walk view (`own_labels_for_walk`) attaches the new `Paint::text_font`
     (default = `widget_font`) to prim-derived labels; the detached base label keeps
     `widget_font`. TextBox overrides `text_font`: a customized `font_family`/`font_size`
     serves the bare family name so the value text draws in the widget's own font at the
     label's size (the control-font string's size suffix would otherwise override it) —
     this is what keeps the editor monospace. App: view()'s side effects (registration,
     initial focus, relayout, popover registration) moved into `display_list`; chrome
     text emitted as prims in the system mono family; `rebuild_text_items` + hand-rolled
     Buffer shaping deleted; the app `FontSystem` stays for `editor.prepare_text` (glyph
     advances — cursor↔pixel mapping).
   - **6i — the 6d trap is FIXED in the walk; `cce-data-editor` flips `display_list_text`.
     DONE (live-verified: full-frame parity modulo the uniform ~2px engine line-height
     shift; row selection → inline value editor + statusbar update; choice-dropdown
     popover renders and occludes rows beneath).** Engine, two changes that make a
     `paint_tree` list's text emit exactly once — the embedded-base-dissolution
     precondition is GONE for the flip step: (1) the legacy `Element::paint_self` default
     emits text only for LEAVES — the legacy container `text_labels` overrides
     (Backplate/Layer/Page/SplitBox/Plate) aggregate their children's labels, which the
     walk reaches itself; a legacy container with OWN text overrides `paint_self` (Plate
     now serves its label this way; SectionContainer still aggregates from internal
     non-child widgets and needs the same treatment if it is ever walked). (2) the walk's
     `renders_own_subtree` branch (TreeList) emits the subtree's text from the recursive
     `text_labels_with_font_and_bounds` — the walk never descends there, so the aggregate
     is that subtree's text once, with the tuple pipeline's fonts/bounds. App: the old
     `view()` body (registration, focus, relayout, inline-editor placement, popover
     registration) moved into `display_list`; `rebuild_text_items` shrank to a
     widget-state refresh (rebuild_tree, prepare_text, statusbar text); the toolbar file
     label is a prim; `add_element_labels` and the TextItem cache deleted (−240 lines).
   - **6j — `cce-graph` flips `display_list_text`. DONE (live-verified: static A/B
     residual is only the uniform ~2px engine line-height text shift; node selection,
     View-menu popover occluding the node beneath, control-panel toggle with its
     multi-line info text via the 6i container fix).** Same recipe as 6i; the app
     `FontSystem` deleted outright (no prepare_text dependency). Also fixed a latent
     Phase 3 loss found on the way: the loaded-image pixel quads and selection borders
     were pushed into `view()`'s plain quads, which the backend DISCARDS when
     `display_list` returns Some — images had only rendered under `CCE_LEGACY_PAINT`
     since the Phase 3 adoption; they now emit into the display list itself.
   - **6k — `Application::load_system_fonts` lands; cce-fonts opts in. DONE (live-verified:
     baseline previews Berkeley Mono (bundled) but drew Adwaita Mono (system-only) BLANK —
     the app's core purpose was broken for installed fonts; with the opt-in Adwaita renders
     as its own face).** The render-FontSystem design settled as a bool `Application` hook,
     consulted once at GPU init: the engine's `WgpuAdapter` FontSystem loads system fonts
     additively (bundled first, so fontdb face IDs stay aligned with every
     `create_font_system*` database — the alignment that makes app-side-shaped buffers
     rasterizable engine-side). This was the same face-ID-mismatch class as the 6e
     cce-colors bug, and it predates Phase 6 entirely.
   - **6l — `TextAttrs` lands; `cce-fonts` flips `display_list_text`. ALL TEN display_list
     adopters are now fully on the single paint path. DONE (live-verified: style popover
     renders with labels on top and occludes the text beneath via the ui_context-only
     registration; selecting Italic re-renders the alphabet in the italic face).** Engine:
     `Prim::Text` gains `attrs: TextAttrs { italic, weight }` (toolkit-plain — no glyphon
     types in the scene layer), emitted by `PaintCtx::text_attrs`, shaped by
     `get_text_buffer_attrs` (cache key includes them). App: same recipe, plus the popover
     drawn INTO the list (replacing `overlay_quads`) with labels bounded to the popover
     rect, and the open popover registered in `ui_context` ONLY — a global registration
     would spawn an empty xdg popup (no `render_popovers` here). Restored two more Phase 3
     view()-quad losses (panel borders, alphabet box) and fixed the alphabet's premature
     wrapping (legacy passed a LOGICAL width to `set_size` on a physical-unit buffer).
   - **6m — first container dissolution: `cce-data-editor`'s root Backplate. DONE
     (live-verified: A/B residual 43px over the 8% threshold — translucency noise;
     selection/editor/statusbar interactions exercised; held window drag not headlessly
     drivable, covered by the new unit test).** The root-Backplate dissolution recipe,
     now established: (1) the plate becomes prims replicating `Backplate::color()`/
     `corner_radius()` (page-low bg at active backplate opacity, config radius); (2)
     top-level widgets register directly in `ui_context`, parentless, and the paint walk
     runs per top-level widget in the old child order (composite widgets keep their own
     children — the splitter still owns its panes); (3) window dragging answers via the
     new `UiContext::drag_allowed_at` — the `is_movable_backplate_at` candidate walk
     minus the registered-movable-Backplate requirement, because the surface itself is
     the movable plate once the Backplate is gone.
   - **6n — `cce-graph`'s root Backplate dissolved; two engine input holes fixed. DONE
     (live-verified: A/B residual 14px over the 8% threshold; View menu → Control Panel
     toggle → panel renders through the walk).** The 6m recipe applied (plate prims,
     parentless top-level registration, walk in old child order, `drag_allowed_at`);
     popovers moved to the 6l ui_context-only registration (the global registration
     spawned a render-only xdg popup double-drawing the menu; `render_popovers` override
     deleted). Found en route, both pre-existing: (1) a press inside an OPEN popover's
     plate could start a window move and swallow the click when the widget beneath does
     not block dragging (Graph's edge-exclusive canvas hit) — both drag questions now
     veto via `point_in_active_popover`; (2) the engine's render-only popups took input
     with their default full input region — they now carry an EMPTY input region.
     Verification lesson recorded: the compositor drops pointer focus after each click
     (Leave with no re-Enter on in-window motion), so headless click sequences MUST
     re-park the pointer (`wlrctl pointer move -10000 -10000`) before every click — a
     skipped re-park looks exactly like an input regression.
   - **6o — graph's two Plates dissolved: `cce-graph` is EMBEDDED-BASE-FREE (the first app
     to get there via dissolution; text-editor was born free). DONE (dropdown row A/B'd to
     the solver's exact rects after switching to the bridge's own sizing entry,
     `Element::intrinsic_size` — a `measure` call returns display-label widths instead;
     control-panel area A/B: AE=0).** The transparent dropdown-row Plate (pure layout
     shim) became direct placement; the draggable control panel became app state + prims
     replicating Plate's exact visual (config plate color / drag tint, plate opacity,
     negative-alpha blur flag, border, radius) with the label walked standalone via
     Plate's centered-first-child rule. Panel drag reimplemented properly app-side —
     NB the legacy `Plate::on_cursor_moved` forwarded drags only to CHILDREN, so the old
     panel's own drag likely never moved it (manual drag check pending).
   - **6p — cce-fonts' root Backplate + all three Plates dissolved. DONE (live-verified:
     family select, style popover + occlusion, Oblique re-render, select-mode bar).** The
     6m/6o recipes at full width, plus the first APP-OWNED EVENT DISPATCH: a
     `dispatch_widgets` list replicating the Plates' forwarding — popover-first press
     pass, unfocus-on-missed-press, `drag_update` forwarding for dragging children
     (scrollbar thumbs), panel-grouped order. The A/B surfaced another legacy
     double-draw: `Plate::paint_self` aggregated child plain-quads while the walk painted
     the child again, double-compositing the ScrollBox background (~22 units darker) —
     the dissolved single-draw is the correct rendering. Remaining in fonts: `ScrollBox`
     (mid-panel scroll state/scrollbar) and `List` (browse-list scroll/frame) — the last
     two embedded-base types in the app.
   - **6q — fonts' ScrollBox + List dissolved: cce-fonts is FULLY EMBEDDED-BASE-FREE.
     DONE (live-verified: wheel scroll, scrollbar track-jump lands proportionally, family
     click after deep scroll + scroll_to_index, alphabet re-render in the new family).**
     Both were pure scroll frames in this app (List with columns=None; the rows are
     standalone Buttons), so they reduce to one app-owned `ScrollRegion` (~150 lines):
     scroll state, wheel, thumb-grab/track-jump/drag, hover-scoped keyboard scrolling,
     item-y math with List's silently adjusted item height (max(24, list font + 14)), and
     bg/track/thumb prims. The bg is a single list_bg layer — the legacy leaf-walk
     stacked rounded + plain copies (the translucent double-compositing class again);
     residual A/B delta is a 2px bottom-edge strip.
   - **6r — colors' and files' root Backplates dissolved, BOTH A/B'd to AE=0.** These are
     the flat-pipeline (render_widget) apps, and their dissolution surfaced the legacy
     aggregate's GLOBAL tuple-order contract: `render_widget(root)` emitted every
     descendant's PLAIN quads first (via `all_quads` aggregation), then the root's
     rounded bg, then every descendant's ROUNDED quads — so the root's translucent plate
     WASHES over the plain content (colors' muted slider gradients depend on it; a
     naive plate-first order renders saturated). Replication: per-child `render_widget`,
     partition the tuples by radius, and interleave [plain…, plate, rounded…]. Wheel in
     colors propagates per-slider; both apps answer dragging via `drag_allowed_at`.
     Files' view-dropdown popover + breadcrumb context menu re-verified live.
   - **6s — settings' root Backplate + StatusBar dissolved; `Dropdown::set_corner_frame`
     lands. DONE (WINDOW_PC tuple stream byte-identical; pixels AE=0; live: dropdown
     popover → page switch to Processes with statusbar text following, wheel scroll,
     service-list render).** Settings needed what colors/files didn't: its plate radius
     is a hardcoded 12, so the legacy aggregate's corner RESOLUTION mattered — a child
     plain quad flush with a window corner picks up the plate radius there (the
     `render_widget` extra-corners logic against the ROOT rect). The hand assembly
     replicates the full aggregate: child plain quads (window-clipped, root-corner-
     resolved), plate, child rounded quads, root-clamped text, in the old child order.
     Two parent couplings surfaced (the widgets read their Backplate ancestor):
     (1) StatusBar — bg falls back from the backplate-statusbar theme color to
     STATUS_BG, bottom corners round at the PARENT's radius, text color/font are
     backplate-specific; it dissolves outright (pure chrome in this app) into tuples +
     a `status_text` String. (2) Dropdown — the backplate-concentric corner cut walks
     for a Backplate ancestor and silently degrades to a plain rounded box when the
     walk finds nothing; new `Dropdown::set_corner_frame((rect, radius, corners))`
     hands it the frame explicitly and takes precedence. Also found (pre-existing,
     reproduced on the pre-6s baseline): the engine xdg-popup positioner anchors at
     the widget's BOTTOM edge regardless of the app's open-upward popover rect, so
     settings' page popover displays below the window while clicks land on the
     app-side (invisible, in-window) popover rect — the engine popup path's last
     consumer; fix when settings' popovers move to the 6l ui_context-only pattern.
     The root's `with_border` was never rendered (a rounded Backplate emits no plain
     bg quad; the border branch fires only on plain bg quads) — dropped, not ported.
   - **6t — settings' popovers + context menu draw INTO the frame;
     `Application::draws_own_popovers` lands. DONE (live-verified: dropdown popover
     in-window with page geometry occluded beneath, item click switches pages both
     ways, spinbox right-click context menu at cursor with dl-text occluded beneath,
     dismissal).** This fixes the 6s finding at the source: the engine's render-only
     xdg popup anchored at the widget's bottom edge regardless of the app's open-upward
     popover rect, so settings' page popover displayed BELOW the window while clicks
     landed on the app-side in-window rect. The app now runs the same
     `render_popovers` collector into its own tuple stream (appended above window/
     page/search content; kept out of the scrollable vecs so the wheel fast-path can't
     shift popover content) and deletes the override. Page-widget popovers
     (notifications/fonts menus) shift by −scroll_y — the subtraction the popup
     positioner used to apply — keeping display aligned with hit-testing under scroll.
     Engine: `draws_own_popovers` (default false) gates BOTH popup spawn triggers (the
     global popover registry and global context-menu visibility), and under the flag
     `render()` adds the visible context menu's rect to the dl-text occlusion overlays
     (the menu is engine-global state, not a `ui_context` popover; its own labels are
     exempt via bounds == rect). Remaining popup-path consumers: cce-data-editor and
     cce-text-editor (`render_popovers` overrides) — the popup surface, `ActivePopup`,
     `PopoverCollector`, and this flag all go away once they draw their own.
   - **6u — settings' Switcher + Page dissolved; the System page comes back from the
     dead. DONE (audio A/B: window tuple stream byte-identical, pixels AE=0;
     live-verified across six pages — spinboxes, context menu, page switching, wheel
     + fallback, System governor dropdown + scroll, notifications dropdown, fonts
     textbox focus).** The top two tree layers reduce to app state: the active page
     was always `app.current_page`, page scroll was already `scroll_y`, so what was
     load-bearing was Page's scrollbar child, its out-of-bounds event gate, keyboard
     scrolling, and being the propagate root. `dispatch_page_event` replicates the
     routing (OOB gate with scrollbar-drag bypass; scrollbar first with the y-unshift,
     then sections in reverse child order; PointerMove visits all, others stop at the
     first handler) against the app-held SectionContainer clones; the scrollbar is an
     app field whose quads collect into the window assembly's plain slot with the
     legacy one-frame-stale content height. Found on the way: the System page's
     widget-tree render path — the only page not on immediate-mode — SEGFAULTED at
     launch on the pre-6u baseline (raw-pointer one-time section/label tree; the
     use-after-free class this rebuild exists to kill). A complete immediate-mode
     view for it existed in the file, never wired to the `AppPage` impl; 6u flips it
     (labels → `sec.text`, InfoBoxes advance the section cursor, menus linked into
     the clone sections like every other page). Pre-existing, deferred to the
     List/ScrollBox dissolution: the processes lists' inner wheel is dead; the page
     scrollbar's right half sits in the compositor's 8px edge-resize zone.
   - **6v — settings' List/ScrollBox dissolved (all five lists). DONE (A/B render
     dumps on processes/packages/radios: rect streams byte-identical minus one
     duplicated pair per list, see below; live-verified — per-list wheel, page-scroll
     fallback, scrollbar track-jump that sticks, focus tint, package row click →
     selection + info fetch, scroll state surviving watcher rebuilds).** Every
     settings list was a pure scroll frame (`List` with `columns: None`; the pages
     draw the rows), so the recipe is fonts' 6q `ScrollRegion` ported app-side
     (`cce-settings/src/scroll_region.rs`) with the List-flavored visuals (1px
     focus/hover-tinted rounded border + inset bg) and the List-mirroring API the
     pages already used. Routing: `AppPage` grows `extra_dispatch_roots` — the
     `InteractiveListItem` rows dispatch directly as propagate roots (`Adapted`'s
     press/wheel hit-gate makes misses fall through, so root order is immaterial) —
     plus `handle_mouse_wheel` (after widget dispatch, before the manual page-scroll
     fallback: the legacy "inner ScrollBoxes take the wheel first" slot) and
     `handle_key_input` (hover/focus-scoped, before the page's scroll-key fallback);
     regions ride the existing pointer down/move/up hooks (audio's slider-drag slots).
     This FIXES the 6u-deferred dead inner wheel, and two latent visuals of the
     columns=None List path: `List::extra_quads`' early return never removed the
     ScrollBox bg quad, so `render_widget` emitted the border+bg pair TWICE (plain
     bg through the solid-border branch + `all_rounded_quads`) — the 6p double-
     composite class — with the scrollbar track/thumb sandwiched UNDER the second
     translucent bg wash. Single-drawn now; the inner scrollbars are visible for the
     first time. Only other A/B delta: item-label clip bounds relax by ScrollBox's
     4px text inset (rows are fully-visible-culled, nothing renders in that band).
     Replication trap for other apps: the region's `focused` is a local bool
     (press-inside sets, press-miss clears) standing in for the global
     `focus::set_focused(scroll_box)` — ctrl-nav can no longer land on a list, and
     the focused border tint shows through the translucent bg as a green wash
     (legacy did this too, darker under its doubled bg). Spot-check PASSED
     (2026-07-13, via ccectl held-drag injection, cce 881c2b1): held thumb drag
     scrolls and tracks mid-drag; arrow scrolls one row; PageUp/PageDown page both
     directions over a hovered list. The green focused wash appeared as documented.
   - **6w — settings' SectionContainer dissolved; cce-system-settings is
     embedded-base-FREE. DONE (A/B render dumps: all nine pages byte-identical
     modulo live data — the sections never painted; live-verified — notifications
     spinbox + menu open/select with in-frame occlusion, audio spinbox round trip
     through pactl and the watcher, processes filter-box click-to-focus, services
     list wheel).** The per-rebuild section clones were pure event/focus plumbing:
     propagate roots whose `container` children were the pages' widgets, plus the
     ctrl-nav focus targets. `AppPage::section_widgets()` (one widget-pointer group
     per section, old count/order) replaces `get_section_containers` +
     `link_children` + `clear_children`; the widgets dispatch directly as propagate
     roots flattened in the legacy order, and section-level keyboard focus is an
     app-side index, single-slot with the global widget focus exactly as when both
     lived in `FOCUSED_WIDGET` (entry → section 0; ctrl+j/k cycle; ctrl+i descends
     to the section's first widget — the legacy walk went through the
     header/container intermediates; ctrl+u ascends from a widget to its section;
     a focus-taking click and page switches drop the highlight). Also killed a
     latent use-after-free of exactly the class this rebuild targets: the focused
     section clone was dropped and reallocated EVERY rebuild while the global
     focus pointer kept aiming at it — it survived only because same-size Vec
     reallocation tends to reuse the freed block. Ctrl-nav spot-check PASSED
     (2026-07-13, nested rig) — and it took two fixes to get there. (1) cce 114ef89:
     injected key-down/key-up now updates an injected xkb mask and pushes a
     modifiers event (OR'd over the device state), so ctrl/shift/alt/super combos
     land like hardware. (2) The check then caught a REAL cce-ui bug (ec511f4):
     handle_key preferred event.utf8, which xkb control-transforms while Ctrl is
     held (ctrl+j arrived as Character("\n")) — settings' ctrl-nav could never
     have fired from real hardware either. With the keysym preferred under Ctrl,
     ctrl+j/k cycle the section highlight and ctrl+i/u descend/ascend
     (audio-page border diffs). ~~NB accounts/storage take `_sec_focused` and
     render no highlight — ctrl-nav is invisible on those two pages.~~ FIXED
     2026-08-13 (cce-system-interface@`573bf43`): both pages now thread
     `sec_focused` into their `view()` and pass it per section instead of a
     hardcoded `false`. Live-verified — storage cycles the highlight
     Local Storage → Memory on successive ctrl+j, accounts highlights its single
     section (42,513 px changed in a rectangle-outline distribution: dense at the
     well's top and bottom edges, constant down the sides). All 14 pages now
     consume `sec_focused`; none takes it as `_sec_focused`.
     Re-verified on the LIVE session (2026-07-13, after the compositor restarted
     onto the 114ef89 binary — the restart was the only reason this had been
     nested-rig-only): ctrl+j → j → k and ctrl+i → u on the audio page, both
     round-trips returning pixel-identical frames (0 AE diff), highlight
     visually confirmed cycling Output→Input. Nothing pending on ctrl-nav.
   - **6x — data-editor + text-editor off the engine popup path; the render-only xdg
     popup machinery is DELETED. DONE (live-verified: text-editor File menu open +
     item click; data-editor recent-files menu → config.kdl load, tree context menu
     with occlusion + Copy Key through the clipboard; settings page dropdown +
     page switch unaffected after losing its gate).** Both apps now collect their
     ui_context-registered popovers via `PopoverCollector` and emit them last in
     the display list (data-editor appends the global context menu too), labels
     bounded to the overlay rect — the 6l/6t recipe; registration is
     ui_context-only. With the last consumers across, the engine sheds the whole
     popup path: `ActivePopup` (wgpu surface + viewport + vertex buffer per
     popover), the xdg positioner/spawn/despawn block in the event loop, the popup
     render pass, the popup-surface pointer-coordinate translation, the
     `PopupHandler` + `delegate_xdg_popup` plumbing, and the
     `Application::render_popovers` + `draws_own_popovers` hooks (settings'
     override removed; the context-menu dl-text occlusion rect is now
     unconditional). Every popover in the workspace is app-drawn, in-frame, where
     it hit-tests — the 6s below-window-popover class of positioner bugs is
     unrepresentable. NOTE: the compositor-side dismissal in `PopupHandler::done`
     (unfocus popovers + hide context menu when the popup was dismissed) went with
     it — in-frame apps already own dismissal (press-outside), same as
     fonts/settings. The global `widget::popovers` registry is now write-only
     (apps still clear/register into it) — delete it with the legacy paths.
     Drive-by: cce-designer had not compiled since 6k (direct `WgpuAdapter::new`
     call missing the new `load_system_fonts` bool) — fixed.
   - **6y — files' SplitBoxes + BrowseContainer/NetworkContainer dissolved. DONE
     (browse page A/B: zero >8%-amplitude pixel diffs; network page's only diff is
     a removed paint bug, see below; live-verified — row select, double-click
     navigation, view-dropdown page switch, divider hover tint via hover-on/off
     crop diff, preview populate).** The split reduces to an app-owned `SplitPane`
     (frac + divider drag/hover + divider quad — the SplitBox two-child horizontal
     math verbatim); the pane containers were pure layout shims whose child copies
     the pages have always re-rendered on top (the Phase 0 double-paint), so the
     window assembly now emits only the divider quad and the preview pane
     (`render_widget` at the right pane rect, text clamped to the pane like the
     legacy SplitBox bounds clamp). Killed on the way: the left pane's under-copy
     double-compositing every translucent quad, including the NetworkContainer's
     full-width breadcrumb-copy strip that visibly leaked behind the graph page's
     top bar — the exact class the Phase 0 stopgap patched for Browse only.
     Verification trap for the log: `wlrctl` pointer warps land as Enter WITHOUT
     Motion — nudge (`move 2 2`) after warping or app hover state never updates
     (cost an hour chasing a "broken" divider tint that was fine). Held divider
     drag spot-check PASSED (2026-07-13, ccectl pointer-press/release): frac
     tracks the held drag, panes re-lay out, window stays put. The Enter-without-
     Motion trap turned out to be a REAL RUNNER GAP, not just an injection quirk:
     the backend's Enter arm set the cursor icon but never fed the enter position
     to `handle_pointer_move`, so a press straight after crossing into the window
     hit the movable-backplate check with stale hover and moved the WINDOW instead
     of grabbing the divider. Fixed in cce-ui 9e23229 (Enter now routes like
     Motion); re-verified no-nudge cross+press drags the divider.
   - **6z — files' List dissolved; cce-files is embedded-base-FREE. DONE
     (live-verified: row click select with preview/details update, double-click
     navigation, breadcrumb navigation, wheel scroll with selection retained,
     hover tint, item count; 3 new RowList unit tests).** The column-mode List
     flavor ports verbatim to the app-owned `RowList`
     (`cce-files/src/row_list.rs`): column_bounds (Flex/Absolute/RightOffset),
     row virtualization + hit math, 400ms double-click, the scrollbar, and the
     cell layout (icon column, primary/secondary tints, char-estimate truncation,
     viewport-inset clip bounds). The in-List search box became a standalone
     BrowseState TextBox; the open/close shortcuts and SearchChanged plumbing
     move app-side (close returns the empty SearchChanged the legacy
     just_changed flag produced). Two fixes: the 6v sandwich again
     (render_widget emitted scrollbar + row overlays UNDER the rounded bg), and
     a NEW DISSOLUTION TRAP for the checklist — a dissolved widget no longer
     blocks window drags via its registered `blocks_backplate_drag`, so
     `is_movable_backplate_at` must veto its rect app-side; without it every row
     press became a compositor window-move grab and the app saw only the release
     (looked exactly like a dead click). Kept legacy: the view's
     scroll-into-view snaps the wheel back while the selected row would leave
     the viewport. Search typing not headlessly drivable — user spot-check
     pending.
   - **6aa — data-editor's SplitBox dissolved. DONE (live-verified: empty-state
     pixels identical modulo the cursor sprite; config load via the File menu,
     tree wheel, tree row select with the inline value editor, divider hover tint
     via crop diff).** The 6y `SplitPane` recipe on the scene-walk app: panes
     positioned directly from the pane rects and walked as separate roots, the
     divider quad emitted in the splitter's old walk slot. Removes the app's last
     raw-pointer child container and retires the Phase 2b
     `scene::bridge::layout_subtree` showcase that drove the split (the layout
     engine's app-facing debut now waits for the routed-events/scene-layout
     phase). TreeList intentionally NOT dissolved: at ~1.8k lines of tree
     expansion/inline-edit/annotation logic it is a self-contained walked widget
     (renders_own_subtree) whose internal ScrollBox never leaks — porting it
     app-side buys no hazard reduction; it converts to narrow traits with the
     `Element` deletion instead. Held divider drag spot-check PASSED (2026-07-13,
     ccectl injection): press at the divider grabbed it (app log: press 482 →
     release 332), split tracked the full 150px, and the grab stole keyboard
     focus from the raw editor exactly as designed.
   - **6ab — cce-text-editor on routed events + scene-solver layout: the FIRST app
     fully on the target architecture, end to end. DONE (live-verified: menu-open
     pixels match the pre-change capture at 0.13% = cursor sprite; menu item
     click through the routed release; editor click focus; the
     focused-border-after-outside-click oddity reproduced byte-identically on
     the stashed pre-change binary — pre-existing).** Layout: the frame is a
     plain `Arena<LayoutBox>` tree solved by `scene::layout::compute_layout` —
     no Element in the loop, the solver used directly by the app (stretched
     column [top bar fixed 42 / content grow padded 10 / status fixed 30], menu
     a fixed leaf, editor growing) — and it reproduces the legacy hand-math
     rects exactly, clamps included. Events: each handler builds one `Event` and
     routes it through `UiContext::propagate_event` per root; the router owns
     press hit-gating, Enter/Leave synthesis, drag-target recording, and
     KeyInput-to-focused delivery, leaving the app take_change plumbing and
     app-level shortcuts only. This is the shape the remaining widget-tree apps
     (data-editor foremost) migrate toward, and the pattern the demo
     (`cce-ui/src/main.rs`) should teach.
   - **6ac — data-editor on routed events + scene-solver layout; the
     routed-events/scene-layout item is COMPLETE. DONE (live-verified:
     empty-state pixels match 6aa at 0.06% (cursor + caret); config load through
     the routed menu; tree row select → inline choice editor + raw-span sync;
     choice popover open/select; tree wheel).** Layout: chrome + panes are one
     solver tree (stretched column [menubar fixed 42 + File-menu leaf / content
     row grow with pad 10, gap = divider width, panes growing by the SplitPane
     fractions / statusbar fixed 30]) reproducing the 6aa hand rects exactly;
     the SplitPane keeps divider input state, its frame derived from the solved
     panes; the inline value editors stay hand-positioned (they float over tree
     rows). Events: all 30 direct dispatch call sites route one `Event` through
     `propagate_event` per root with the plumbing intact. THE ROUTING TRAP worth
     remembering: the router delivers KeyInput to the ctx-focused widget FIRST
     on every propagate call, so a legacy non-short-circuited keyboard chain
     would deliver a typed key to the focused widget once per call site
     (N-time character insertion) — short-circuit the chain on first handled,
     and re-gate any plumbing that keyed off WHICH call returned true onto
     widget state instead (Enter→ApplyValue now checks the value editor was
     editing when the key arrived). Keyboard flows not headlessly drivable —
     user spot-check (typing, Enter-apply, tree search, keybind recording).
   - **6ad — the demo rewritten as `DemoApp`, the reference `Application`. DONE
     (live-verified: button click, toggle with app-state re-assert, slider wheel
     nudge, dropdown popover open/select with the occlusion clamp visibly
     working, all through routed dispatch).** `src/main.rs` had never been the
     "reference Application" the docs claimed — it was a 1925-line fossil
     predating the engine entirely: a raw Wayland client with its own
     CompositorHandler/SeatHandler impls, its own wgpu state, and hand-copied
     tessellators. Replaced by ~450 teaching-commented lines on the full target
     architecture: display-list frame + display_list_text, solver-driven layout
     (with `shrink` demonstrated for min-width rows), routed events with the
     KeyInput short-circuit rule and state-gated `drain_widget_changes`
     plumbing, ui_context-only popover registration with the in-frame draw, the
     dissolved-root window plate, and `drag_allowed_at` window dragging. API
     footgun surfaced for the log: `Slider::set_value` takes the NORMALIZED
     0..1 value (`with_range` only scales `get_scaled_value`) — passing a
     ranged value silently clamps to 1.0.
   - **6ae — legacy deletion, part 1: the global `widget::popovers` registry is
     DELETED. DONE (write-only since 6x; the mod, its `render_widget` write, and
     the four apps' `clear()` calls are gone; settings' popover renders
     byte-identically after).** Part 1 also produced a CORRECTED precondition
     map for the rest of the deletion — the endgame list had been assuming "the
     last app is across," and it is not:
     - The `view*`/`text_items` paths CANNOT be deleted yet: SIX apps still
       implement them — cce-test-interface (2.1k), cce-authenticator (0.9k),
       cce-display-manager (1.5k), cce-mail (2.5k), cce-layout-interface
       (3.8k), cce-status-interface (4.2k). Each needs its own Phase-6-style
       migration (display-list flip at minimum; dissolutions as found).
       Suggested order: smallest/least-critical first (test-interface,
       authenticator — NOTE it may be the lock screen, verify carefully),
       status-interface last (layer-shell, always-running).
     - The per-widget text getters are additionally load-bearing for the walk's
       legacy branches (`renders_own_subtree`, container `text_labels`
       aggregation) and the migrated apps' hand-rolled window aggregates
       (settings' `collect_window_child`, files' assembly) — they go when those
       consumers move to `paint_self`-only trees.
     - `Element` + `Adapted` go last, after both of the above; TreeList
       converts to narrow traits then.
   - **6af — cce-test-interface across (1 of 6). DONE (A/B: zero >8%-amplitude
     pixel diffs; live-verified — full gallery render, page-dropdown popover
     in-frame, page switch updating the MenuBar title and the status prim).**
     The recipe for the remaining five: move the `view()` +
     `view_rounded_quads()` bodies into `display_list()` in the engine wrapper's
     order (ROUNDED first, then plain, then popover rects — the wrapper reversed
     the intuitive order and apps' visuals bake it in), and re-emit the
     `rebuild_text_items` assembly as `Prim::Text` built fresh per frame,
     deleting the cache + its invalidation call sites + the app FontSystem +
     any `text_areas` override (its extra areas become prims).
     `custom_vertices` stays. Remaining queue: authenticator (verify carefully —
     lock screen), display-manager, email, layout-interface, status-interface.
   - **6ag — cce-authenticator across (2 of 6); the flip FIXED runtime-invisible
     text. DONE (live-verified --standalone + CCE_AUTH_SIMULATE: full dialog text
     renders, zero font-ID warnings — was hundreds per frame — fingerprint-scan
     click drives the animated glow + hint).** It is an xdg-toplevel polkit auth
     dialog (NOT a session lock — safe to run; needs `--standalone` +
     `CCE_AUTH_SIMULATE=1` to show a window without a live polkit request, and it
     auto-exits ~3s in simulate mode so capture fast). The single `view()` (both
     geometry and text) → `display_list()`; the `text_items` assembly → prims;
     app FontSystem / make_text_buffer / text_items field+getter deleted. The
     6e face-ID class again, and worse here — the app used
     `create_font_system_with_system_fonts()`, so EVERY label was invisible at
     runtime; the migration is the fix. General lesson reinforced: any
     legacy-path app with its own FontSystem is a latent-invisible-text
     candidate — don't trust its baseline capture.
   - **6ah — cce-display-manager across (3 of 6). DONE (A/B: 131px AE = 0.004%
     caret blink, zero >8%-amplitude diffs; live-verified --greeter renders
     identically).** The greetd login greeter — run the GUI standalone with
     `--greeter` (daemon mode needs root/greetd). Bundled fonts, byte-match flip.
     view() + view_rounded_quads() → display_list() (rounded then plain, both
     skipping the card); rebuild_text_items → prims; FontSystem / make_text_buffer
     / info_buffer / text_items machinery deleted. The card — a soft radial-glow
     blob drawn with the circular clip disabled — STAYS in custom_vertices
     (escape-hatch layer, on top, untouched). REPO HAZARD hit here: these crates
     live under ~/Dropbox, and a Dropbox sync reverted the edited main.rs to disk
     AFTER build+test but BEFORE the commit landed (a harness-interrupted commit,
     exit 144, left the tree clean at the old file) — had to re-apply and commit
     immediately. Verify `git log`/`grep display_list` actually stuck before
     moving on.
   - **6ai — cce-mail across (4 of 6); the flip FIXED invisible list/detail text.
     DONE (live-verified: inbox list of 3 emails, detail placeholder, and on click
     the full detail view — subject/From/To/Date/body + Reply/Delete/Mark-Unread
     toolbar).** view() body (all plain quads) → display_list() via a small
     `__EmailQuadSink` shim mapping the ported quads.push/extend to PaintCtx::quad;
     rebuild_text_items → emit_text_prims(&mut pc). The 6e face-ID class AGAIN
     (create_font_system_with_system_fonts): the list rows, detail metadata/body,
     and placeholder were all invisible — only the paginator tabs showed. Fix:
     switch the app FontSystem to bundled create_font_system() (KEPT for the
     TextBoxes' prepare_text measurement — now matching the engine render FS) and
     render via prims. Note the AE-vs-baseline metric is misleading for these
     invisible-text fixes (tiny % of dark-on-dark pixels change) — judge by
     whether text APPEARS, not by AE.
   - **6aj — cce-layout-interface across (5 of 6); needed + consumes the new
     boxed-text prim; the flip FIXED invisible text. DONE (live-verified: the
     whole properties/geometry/alignment/add-elements UI renders where the
     baseline showed nothing; 104→0 font-ID warnings).** This app forced the
     boxed-text feature (previous commit): its canvas Element::Text boxes need
     word-wrap + h/v alignment, unrepresentable as a plain Text prim. view() +
     view_vectors() → display_list() (a __LayoutQuadSink shim for the
     quads.push/extend body; vectors → PaintCtx::vector); rebuild_text_items →
     text-prim tuples carrying an optional TextLayout, emitted via text_with /
     text_boxed. Same 6e face-ID class (create_font_system_with_system_fonts);
     fixed by bundled FS + prims. VERIFY GAP: the canvas boxed prim itself
     (wrap/align on the page) could not be pinned headlessly — a placed text box
     defaults to page x=40, behind the ~540px properties panel, and there is no
     virtual keyboard to type a clear coordinate; it renders via the identical
     dl_text_items path as the confirmed-visible text. User spot-check: place +
     drag a text box onto open page, confirm wrap + alignment.
   - **6ak — cce-status-interface across (6 of 6, LAST legacy-path app). DONE
     (live-verified: an isolated `--module clock` instance renders "Friday, July
     10, 2026 … PM" with its rounded background pill through the display list).**
     The persistent layer-shell bar. view() + view_rounded_quads() bodies move
     into display_list() (rounded boxes, then status-bar bg / module rects /
     separators as prims, in the wrapper's ROUNDED-then-plain order); module text
     becomes fresh Prim::Text each frame. overlay_quads() stays a separate on-top
     pass (tray hover highlights over text). The status bar's OWN text is never
     set in this app (get_text_items was a no-op), so its text_items()/custom
     text_areas() overrides are deleted; self.font_system is kept only for the
     modules' measure-then-position shaping. Mechanism: modules build a
     StyledLabel to measure width, then emit via the new StyledLabel::into_prim
     (cce-ui 005a53f) through a draw_label helper — the vertical bar's centered
     per-char text rides the boxed-text TextLayout. Bundled create_font_system(),
     so NO 6e invisible-text hazard — a byte-match flip. Full multi-module A/B was
     avoided (the no-arg binary is a launcher daemon that would spawn a bar
     conflicting with the user's live one); a single isolated module segment was
     the test surface.
   - **All six legacy-path apps are now across.** The `view*`/`text_items`
     Application-trait deletion is unblocked.
   - **6al — legacy `view*`/`text_items`/`text_areas` DELETED. DONE (16 client
     apps compile; 180 tests pass; settings live-verified).** With every app on
     `display_list()`, the legacy geometry/text trait surface was dead code:
     removed `view` / `view_rounded_quads` / `view_vectors` / `text_items` and the
     default `text_areas` mapping from the `Application` trait; in `render()`,
     dropped the `quads`/`rounded_quads`/`vectors` collection + the tuple-wrapping
     `None =>` branch (so `dl = display_list().unwrap_or_else(empty)`) and the
     `text_areas()` call (an empty `areas` vec the dl-text loop fills). KEPT:
     `overlay_quads` (status-bar tray hover), `custom_vertices` (display-manager
     card, test-interface gallery), `display_list` / `display_list_text`. Pure
     dead-path removal — every implementor already took the `Some(dl)`/empty-text
     arms. (cce-designer + cce-cloud drive `WgpuAdapter` directly, never implement
     `Application`, so they're untouched.)
   - **6am — dead widgets SplitBox / MultiControl / KeybindsControl DELETED.
     DONE (cce-ui + all 18 apps compile; 175 tests pass; settings A/B AE=0).**
     Phase-6 dissolutions orphaned all three — no client app (nor live cce-ui
     path) constructs them. Removed the files, re-exports, the two
     `get_*_sub_widget_info` helpers + their `render_widget`/`window_runner`
     downcast blocks (fire only when the widget IS that type — none is, so
     behavior-preserving), the dead `name == "MultiControl"/"KeybindsControl"`
     span-full terms, and `render_widget`'s now-redundant `let mut corners = …`
     shadows. Each carried the full legacy text-getter aggregation, so this trims
     a big slice of the getter consumer graph.
   - **Per-widget text getters — GATED, not yet deletable.** Investigation
     (6am) established that every getter still has a LIVE consumer, so none can
     be removed until those move to `paint_self`/prims first:
     - `get_text_items` → **cce-designer** (custom `WgpuAdapter` render loop in
       `render.rs`: text-buffer cache + curved-menu-text special cases). Designer
       + cce-cloud never implement `Application` — they were skipped by all of
       Phase 6 and still drive `WgpuAdapter` directly.
     - ~~`text_labels_with_font_and_bounds` + `widget_font` → `layout::render_widget`~~
       **DONE (6an).** render_widget now sources its text from the scene walk
       (`paint_root_into` → keep only `Prim::Text` → emit onto the RenderTarget),
       keeping its own geometry path; the prim carries the per-widget font+clip so
       the getters are gone from here. Settings A/B AE=0 on Accounts + (stash-based)
       the spinbox-heavy Audio page. The last difference vs the getter is
       widget_font→text_font, which coincides except for a custom-font TextBox.
     - ~~`.text_labels()` / `.text_labels_with_bounds()` → **four hand-aggregate
       apps** (email, authenticator, display-manager, layout-interface) whose
       `display_list()` emits widget text by calling the getter per widget.~~
       **DONE (6ao)** — see below.
     - Note: the "orphaned" containers Layer / Page are NOT deletable — Layer is
       the embedded base of the live Plate/Page; Page is embedded by the live
       Paginator (transitive liveness through inheritance, not direct app use).
   - **6an — render_widget off the getters (see above). DONE (all 18 apps
     compile; 176 tests; settings A/B AE=0 on Accounts + Audio).** One of the
     three getter-consumer classes cleared.
   - **6ao — the four hand-aggregate apps off the getters. DONE (176 tests;
     all four A/B-verified live).** New `scene::painter::append_widget_text(ui,
     &dyn Element, &mut PaintCtx)`: walks the subtree and appends only its
     `Prim::Text` items — per-widget content font, walk clip composed into prim
     bounds (the 6an recipe as a reusable helper). Every per-widget
     `.text_labels()`/`.text_labels_with_bounds()` call in email, authenticator,
     display-manager and layout-interface replaced with it. A/B results:
     email + layout-interface byte-identical; authenticator + display-manager
     identical except widget-owned labels now render in the widgets' configured
     control font (legacy aggregates dropped the font to `None` — the same
     widget_font→text_font delta as 6an, here visible because these apps
     configure a monospace control font).
     - **Display-manager UAF found + fixed:** its `new()` linked the ui_context
       tree and captured `focused_widget` while the State was a stack local, so
       every registered pointer dangled after the move — the walk's child
       descent was the first render-path consumer to dereference them (abort on
       a garbage-length alloc); `propagate_event` and the `all_*` child
       aggregation read the same stale pointers all along. Fix: per-frame
       idempotent `relink_tree()` (register/link are id-keyed) + initial focus
       re-derived from the boxes' own focus flags. Also: its flat `widgets_iter`
       lists the card AND the card's children, so text moved to walking the TRUE
       roots (bg + root_container) — flat would double-emit — and `LoginCard`
       (a container with own, non-aggregating labels) got the Plate-style
       `paint_self` override for its two header labels.
     - Lesson for the getter deletion: an app whose tree is linked from `new()`
       by value is a dangling-registry candidate — audit any remaining
       `link_parent_child` calls made before the owning struct reaches its
       final address.
   - **6ap — cce-designer's render loop off the getters. DONE (default +
     circular-pane A/B pixel-identical; add_node label renders via the new
     path; full workspace builds).** The custom `WgpuAdapter` loop's two text
     sources (`get_text_items` widget-buffer fast path +
     `text_labels_with_font_and_bounds` fallback) became one walk: per
     non-menubar widget, `append_widget_text` → text prims, shaped app-side in
     `text_buffer_cache` with the same size*1.4 metrics the fallback always
     used. Per-widget special cases (plate-ancestor bounds, circular cull,
     network opacity, curved-ring feed) operate on prim fields unchanged. The
     curved-ring branch is unreachable today (menubars are skipped before its
     condition) — preserved verbatim, flagged for a future dead-code decision.
     **ALL app-side getter consumers are now gone.**
   - **6aq — walk getter-use consolidated to ONE fonted default; the three
     missed app consumers cleared. DONE (176 tests; settings audio AE=0;
     test-interface AE=5 cursor-level; data-editor loaded-tree AE=0; cloud
     fuzzel+json standalone AE=0; designer unchanged).** CORRECTION to 6ap's
     "all app-side consumers gone": three call sites had escaped the audit —
     settings' `renderer.rs` `collect_window_child` (outside render_widget),
     test-interface's gallery loop, and cce-cloud's `jl`/`fuzzel` labels (cloud
     drives WgpuAdapter directly, like pre-6ap designer, and json_layout was
     live only through it). All three now use walk-derived text. Engine side:
     the `renders_own_subtree` walk branch is just `paint_self` (TreeList +
     newly-flagged JsonLayout carry subtree-emitting overrides; descending
     JsonLayout would draw inactive pages and miss its checkbox side-labels);
     the default `paint_self` leaf drain moved from `text_labels()` to the
     FONTED getter — same labels every legacy tuple consumer served. Ramp got
     the own-labels `paint_self` (the Plate/LoginCard class: container own
     text vs the walk's aggregate rule). Traps recorded: a widget with a
     ui-tree parent must NOT also be walked as a top-level root (test-
     interface's page selector under the status bar double-drew ~10%
     brighter); cce-cloud launches reach the user's DAEMON via
     /run/user/UID/cce-cloud.socket — hold the socket aside to A/B a local
     standalone build.
   - **6ar — the per-widget text getters are DELETED from `Element`. DONE
     (176 tests; full workspace builds; nine apps A/B-verified — AE=0 or
     cursor/translucency/status-race noise only).** `text_labels` /
     `text_labels_with_bounds` / `text_labels_with_font_and_bounds` /
     `get_text_items` are gone from the trait, with Adapted's impls, the
     `Paint::text_items` hook, and every container aggregate (Backplate,
     Layer, Page, Plate, SectionContainer, ColorRamp, ControlPanel,
     JsonLayout, ButtonStrip). Every widget reaches the frame through
     `paint_self`. The deleted defaults survive as painter helpers with the
     labels passed in (`paint_legacy_leaf`, `fonted_leaf_labels`,
     `scroll_ancestor_text_bounds`, `base_control_label`); legacy leaves keep
     their label logic as inherent `own_labels()`; TreeList reads its concrete
     Adapted children via the now-pub(crate) `own_labels_with_font_and_bounds`;
     ControlPanel/ParametersBg/JsonLayout source dyn-children labels off the
     paint walk (ControlPanel re-applies its scroll shift + viewport clamp;
     List and ControlPanel are `renders_own_subtree` — walking into a legacy
     scroll frame desyncs text from geometry at scroll ≠ 0). The concrete
     `context_menu::text_labels()` global stays (inherent method, not the
     trait). ~125-method god-trait is now 4 methods lighter and text has ONE
     path: prims.
   - **6as (teardown, in progress).** Landed: (1) census of all 117 `Element`
     methods vs workspace-wide call sites — five were call-less and are
     DELETED (`Element::paint` — the hover-registration default nothing
     invoked — `as_geom_controller`/`as_spreadsheet_controller` & variants,
     `color_u8`, `is_layer`); (2) the seven tree context-menu actions
     (`copy_key`/`copy_value`/`delete_key`/`expand_node`/`collapse_node`/
     `expand_all_nodes`/`collapse_all_nodes`) are now transitional `Input`
     capability hooks with `Adapted` forwards (the 5k pattern), so the global
     context menu's `dyn Element` dispatch survives the TreeList conversion.
   - **TreeList → `Adapted<TreeList>` — DONE (the staged plan below executed
     verbatim; 176 tests; data-editor loaded-tree A/B AE=0 byte-identical incl.
     the focus wash; row select / context-menu Copy-Key-to-clipboard / search
     click-to-focus verified live).** Two new transitional hooks landed with
     it: `Input::tick_ctx` (EventCtx-carrying tick — the rename-commit focus
     re-target needs the routing ctx) and `Paint::paints_own_subtree` →
     `Element::renders_own_subtree` (the field widgets stay ctx-linked for
     event propagation, but the walk must not also descend — descending
     double-painted them and drew the CLOSED add-key popover box). The 5q
     `legacy_focus_highlight` trap struck again (the focused tree's teal wash).
     KNOWN-LATENT (pre-existing since 6ac, verified identical in the
     pre-conversion baseline): wheel-over-tree doesn't scroll and the Add-Key
     button doesn't open its popover — children-first propagation
     short-circuits on the hit child before the tree's own toggle/scroll
     logic runs. Fix belongs to the event-routing follow-up, not the widget.
     Original staged plan (executed):
     - `Layout`: `rect_assigned` caches the rect; the `set_rect` body
       (search box / add-key button / popover box / scroll box arrangement +
       `update_bounds`) moves to the assignment hook. No container children —
       the walk treats the adapter as a leaf, so `renders_own_subtree`
       becomes unnecessary.
     - `Paint`: `color`/`rounded_corners`/`corner_radius`/`solid_border`/
       `widget_font` port straight; geometry aggregates become ctx-less
       (TreeList's children are FIELDS — its `children(_ctx)` ignores the
       ctx already); subtree text rides the `serves_legacy_labels` hatch
       (`legacy_labels_with_font_and_bounds(rect, ctx)` = today's
       `subtree_fonted_labels`); `prepare_text`; popover via
       `Paint::popover`/`draw_popover`.
     - `Input`: the mouse/cursor/wheel/keyboard/tick bodies move into
       `on_event` arms with `ectx.ui` (the 5s ParametersBg pattern);
       drag via the Input drag hooks; `blocks_backplate_drag`; the seven
       tree hooks return their inherent bodies; focus semantics —
       `ctx.set_focused(self)` sites become `ectx.request_focus()` (the
       ADAPTER's pointer, not the inner). CAUTION: `mouse_input` registers
       the inline `edit_box` into the ctx TREE (`register_widget` +
       `link_ids(self_id, …)` + `set_parent`) — under the adapter, `self_id`
       must be the adapter's base id (`EventCtx::widget_addr` precedent).
     - Sweep: data-editor field → `Adapted<TreeList>` (Deref covers the
       concrete calls: `scroll_box.scroll_y`, `set_flat_keys`, `take_*`
       drains, `get_row_rect`, `select_and_show_key`, `focus_search`), one
       raw `*mut TreeList` cast → `as_ptr_mut()`. A/B: loaded tree, row
       click + inline rename (double-click), context menu Copy Key via
       wl-paste, search focus, add-key popover, wheel.
   - **Leaf sweep COMPLETE — every widget is on the narrow traits (6as).**
     Trackpad, KeybindRecorder, FontSelector, ColorSelector, Ramp, ColorRamp
     all converted (each A/B'd: gallery/DE/LI diffs = cursors, launch-phase
     animation, or AE=0). Notables: ParametersBg's typed color rows re-typed
     to `Adapted<ColorSelector>`; ColorSelector's in-file keyboard tests pass
     THROUGH the adapter; Ramp/ColorRamp take the TreeList shape
     (paints_own_subtree + tick_ctx + per-tick re-parenting of field widgets
     so their label fade blends against the adapter's color); and
     `Adapted::paint_self` gained a subtree TEXT PASS-THROUGH — a
     paints_own_subtree widget's Text prims forward verbatim with per-child
     fonts/bounds instead of being flattened to widget_font by the
     own-labels re-derivation (composites with mixed child fonts rendered in
     the default serif without it). `impl Element` now remains ONLY on: the
     containers (Layer/Container/Page/Plate/Backplate/ScrollBox/ScrollBar/
     List/ControlPanel/SectionContainer/JsonLayout/Menu/MenuBar-internals),
     app-local widgets, ContextMenu, and `Adapted` itself.
   - **Then: retire the `as_*_controller` pairs — SCOPE CORRECTED.** The
     earlier four-site estimate came from an over-filtered grep; the real
     surface is ~45 sites: cce-designer's HTTP-action/menu plumbing holds
     ~16 (including generic roster queries like "does ANY widget have an
     open menu" via `as_menu_controller()` over `Box<dyn Element>`), and
     `Switcher` implements MenuController by forwarding to its ACTIVE PANE
     through `as_menu_controller_mut()` on `dyn Element` — a live
     capability-dispatch system, not vestigial casts. Retirement needs a
     design decision first: either a standalone capability registry
     (`&dyn XController` handles registered beside the tree) or deferral to
     the Element deletion itself, where the designer's roster becomes
     concretely typed. Do NOT sweep it mechanically.
   - **Controller-capability decision RESOLVED (6aw): option 2 — defer to
     concrete typing; no registry.** A registry would be permanent
     infrastructure preserving the anonymous-widget pattern the rebuild
     exists to kill. Instead the queries die when their callers get
     retyped. First payoff immediately: `Switcher` — the largest holder
     (17 sites) — turned out to have ZERO constructors workspace-wide
     (settings dissolved its switcher in 6u) and is DELETED. Remaining
     map: cce-designer's roster retype (~16 sites, the bulk), the
     test-interface roster (3), one production site in
     `display/serialize.rs` (`serialize_widgets`' menu-state dump), and
     in-file tests that assert capabilities ride the adapter (die with
     `Element`). The designer retype is therefore the finale's next
     structural step, folded into the tree-machinery retype.
   - **Capability system DELETED (6aw, same session).** The designer's
     accessor block reaches each controller trait by `as_any` downcast to
     the roster index's known concrete type (note: `Adapted::as_any`
     exposes the INNER widget, so the downcast targets `MenuBar`/`Graph`/
     `ParametersBg`/`Spreadsheet`/`Breadcrumb` directly — the first build
     against `Adapted<W>` panicked at launch, caught by the live A/B);
     test-interface downcasts its Paginator; `serialize_widgets` tries
     the two MenuController implementors a roster can hold. With no
     callers left, Element's 12 `as_*` hooks, Adapted's forwards, and the
     14 `Input` capability hooks + their per-widget impls are all
     deleted. Element: 92 methods. A/B: designer/TI static diffs are the
     known noise shapes; designer HTTP `menu_click` (the dynamic-index
     path) verified live.
   - **Gallery containers went app-local (6as).** cce-test-interface's
     widget gallery was the last constructor of ControlPanel / Plate /
     SectionContainer / Backplate: it now owns `ti_widgets.rs` — a verbatim
     ControlPanel copy plus passive Plate/SectionContainer/Backplate
     lookalikes (each reproducing EVERY Element getter the render/event
     paths read: plate color/opacity/blur alpha-negation/corner radius/
     border/label_offset bg shift, the childless section-header row, the
     backplate bevel). A/B: launch-phase progress-bar animation only.
   - **CONTAINER TYPES DELETED (6as).** `Backplate`, `Plate`, `List`,
     `ControlPanel`, `SectionContainer` (+`SectionHeader`) removed from
     cce-ui outright — five files, the re-exports, the `List` branches in
     the scroll-ancestor text-bounds walks, and the then-dead
     `own_labels_with_font_and_bounds`. `ColumnWidth`/`ListColumn` moved to
     their only consumer (cce-files `row_list.rs`). TreeList's drag tests
     re-anchored on `drag_allowed_at`. 171 tests (5 died with their files);
     settings/DE A/Bs cursor-only / AE=0.
   - **Finale opened: dead-flag constant-fold sweep (6at).** With the
     plates gone, census round 2 found no zero-CALL methods but seven
     zero-OVERRIDE ones (only the trait default exists ⇒ they are
     constants). Folded and deleted: `capture_event` (the capture-phase
     branch in `propagate_event_impl` was unreachable), `is_active` (folded
     into the `highlight_color` default), `is_plate` (designer render.rs's
     whole `parent_plate_rect` text-clamp machinery was dead),
     `is_backplate` + `is_movable_backplate` (folded the drag walks —
     `UiContext::is_movable_backplate_at` had become constant-false and is
     DELETED; the `Application` default now just returns false — plus the
     backplate-parent theming/corner branches in MenuBar, StatusBar, and
     Dropdown's `backplate_ancestor` walk; MenuBar/StatusBar `corner_style`
     still reports the parent radius for children that read it through the
     parent pointer, but corners never round). `corner_radii`/`mark_dirty`
     also have zero overrides but carry real derived logic — they die with
     the retype, not by folding. Element: 112 → 107 methods. A/B: files
     AE=0, settings audio AE=0, designer diff = terminal behind the
     translucent window.
   - **Layer + Page DELETED (6au).** Per the 5r survey, no app ever put
     content in Paginator's pages — every consumer keys its own content
     on `selected_page()` — so the `Vec<Page>` was empty containers being
     arranged/registered/toggled/event-proxied for nothing. Their one
     visual (the page-area bg quad, page_color × page_opacity) moved into
     `Paint::paint`. With the stack gone Page had no constructor and
     Layer's only constructor was Page's base: both files deleted, plus
     `Element::is_page`, ScrollBar's Page-downcast write-back, and the
     dead 8/11 of `PageSelector` (now just selected_page /
     set_selected_page / sidebar_w; MenuBar's impls + `page_hidden` field
     went with it). A/B: email/LI/files AE=0; live LI tab click switches
     pages correctly. Census round 3 then found the deletion stranded two
     more zero-override methods — `check_out_of_bounds` and
     `transform_event_for_child` (Page was the only override of each) —
     folded and deleted in the follow-up. Element: 104 methods.
   - **ScrollBar DELETED, ScrollBox demoted off `Element` (6av).**
     ScrollBar's only consumer was cce-system-settings' page scrollbar
     (the dissolved Page subtree's survivor, evented through
     `propagate_event` and painted through `collect_window_child`) — the
     file moved there verbatim and the cce-ui type is gone. ScrollBox is
     never ctx-registered by either consumer (TreeList + the
     test-interface panel copy call it concretely), so its `Element`
     impl was dyn-dispatch ballast: now a plain struct whose former
     Element entry points survive as inherent methods with
     default-derived parity (the scrollbar-click focus claim became
     `focus::clear_focus()` — unfocusing the previous holder was its
     only observable effect). The painter/model scroll-ancestor text
     clamps folded to `None` (no tree parent can be a ScrollBox; none
     ever was at runtime). A/B: data-editor AE=0 plus live wheel +
     track-jump-scroll on a 100-key tree; settings diff = process-row
     churn; TI sub-threshold. Census round 4: only `corner_radii` +
     `mark_dirty` remain zero-override (real derived logic — they die
     with the retype, not by folding).
   - **Dead raw-widget sweep (6ax).** The raw-`Element` census after the
     capability deletion found four more zero-constructor widgets:
     Header, VBox, HBox (export-only) and Svg — which also rode Button
     as an `Option<Svg>` payload no caller ever set, so Button's icon
     branches were statically dead and went too. All deleted.
   - **Container DELETED (6ax part 2).** Settings' system-info
     `actions_row` was a dead field; dm's `root_container` was a
     transparent origin-anchored fan-out — dissolved into direct
     dispatch/walk roots (session list first for events, matching the
     reversed child order; card first for text, matching `children()`
     order). dm A/B: background-animation phase only (B-vs-B control
     differs full-frame), click behavior identical to baseline, 3/3
     interaction rounds alive on both builds. One UNREPRODUCED
     bogus-alloc seen once on the new build — dm's known latent
     stale-pointer signature, 0/3 repro on either binary; watch it.
   - **JsonLayoutWidget moved into cce-cloud (6ay).** The KDL/JSON
     launcher-layout host had one consumer; it cannot be demoted off
     `Element` (cloud feeds it to the paint walk as `&dyn Element`), so
     the file moved app-side verbatim — the impl dies with the machinery
     retype. `Justification` stayed in cce-ui (Button/files/settings
     share it), same path. A/B: content-identical renders (raw diff =
     the overlay's run-to-run spawn position + wallpaper bleed through
     the translucent plate); live checkbox click toggles.
   - **Canvas + Viewport3D moved into cce-designer (6ay part 2).** Both
     designer-only (preview.rs's "canvas" is its own internal type).
     A/B: AE=0. cce-ui's raw `impl Element` surface is now exactly ONE
     production type: ButtonStrip, the ctx-registered embed of
     MenuBar/Paginator — load-bearing in the tree, pinned to the
     machinery retype.
   - **App-local impls onto the narrow traits (6az) — COMPLETE:
     `Element` has exactly ONE production implementor (`Adapted<W>`).**
     Each remaining raw `impl Element` converted to
     `Layout`/`Paint`/`Input` + `Adapted<W>` ahead of the machinery
     retype; only test mocks still implement the trait directly.
     Done: cce-colors' ColorSlider (constructor returns the wrapper, so
     construction and direct-dispatch sites are untouched; A/B AE=0,
     click/wheel live-verified); settings' ScrollBar (now plain data —
     its raw-pointer parent/children fields, Drop, and unsafe Send/Sync
     had zero consumers; strip A/B AE=0 and cross-build byte-identical
     after an identical wheel + track-click sequence); cloud's Fuzzel
     (set_rect side effects → rect_assigned; overlay A/B identical, row
     click moves the selection live) and JsonLayoutWidget (the TreeList
     shape — paints_own_subtree + whole-subtree routing in on_event with
     gates_presses off + tick_ctx; the old overrides survive verbatim as
     inherent methods paint composes with a dummy ctx; render sites'
     all_quads/all_rounded_quads calls now resolve to the adapter's
     reverse bridges, same tuples; checkbox toggle live-verified) —
     cce-cloud is raw-impl-free. Designer's four (PassivePlate, Canvas,
     NodePalette, Viewport3D) followed — constructors return the wrapper
     so the roster's Box pushes and as_any downcasts stand; PassivePlate's
     full getter surface rides Paint, Viewport3D's wheel inertia moves to
     on_event/tick; static A/B = the status sliver only, cross-build
     captures after an identical HTTP circular-toggle + viewport wheel
     structurally identical — cce-designer is raw-impl-free. TI's four
     followed (part 6): ControlPanel keeps its aggregate overrides as
     inherent methods composed by a paints_own_subtree paint, its child
     arrangement in Layout::arrange_children (adapter as parent), and its
     routing in on_event; the raw `as *mut ControlPanel` casts became
     as_any downcasts; inline_label keeps the adapter's label machinery
     out of all four. Gallery + child-window A/Bs pixel-parity, cross-
     build sequences identical — cce-test-interface is raw-impl-free.
     The greeter's three followed (part 7: labels fold into paint off the
     laid-out rect; the display list's card-skip moves from base-pointer
     equality to id comparison; cursor-only diffs, 3/3 click rounds alive
     on both builds, after-state byte-identical). ButtonStrip closed the
     sweep (part 8): embedders hold `Adapted<ButtonStrip>`; two wrapper-
     shadowing collisions (`set_selected`, `take_click` — Element's bool
     signatures vs the model's `Option<usize>` ones) route through
     `inner_mut()`; presses stay ungated so a tab press lands under an
     open dropdown popover; LI's Paginator strip and email's MenuBar
     strip both cross-build byte-identical after identical clicks.
   - **Compiler census of the dyn surface (6ba, experiment reverted —
     the finding is the deliverable).** Hypothesis: with one implementor,
     many `Element` methods are only ever called on concrete
     `Adapted<W>` receivers and could move off the trait before the
     retype. Method: delete a method from the trait, keep it inherent on
     the wrapper (same signature — concrete sites resolve unchanged),
     `cargo check --workspace`; every error is a true dynamic-dispatch
     site. Verdict: **~91 of 92 methods fail — the trait IS the
     machinery surface; there is nothing to slim first.** Three consumer
     classes pin it: (1) cce-ui machinery (context.rs routing/focus/
     drag, the paint walk, layout.rs render paths — `T: Element`
     generics count: measure/preferred_height/set_row_rect live there —
     and core.rs context-menu actions on `dyn` targets); (2) container
     child aggregation over raw child pointers (menu/paginator/treelist
     in-tree, ControlPanel/JsonLayout app-side) touching the full
     paint+input getter surface; (3) the designer/test-interface roster
     broadcast loops (`Vec<Box<dyn Element>>`) calling nearly everything.
     NOTE: a `--workspace` check that fails in cce-ui never reaches the
     app crates — the first pass under-reported; the app rosters were
     where 4 of 5 "clean" candidates actually failed.
   - **The designer roster is concretely typed (6bb).** First retype
     slice, in census order: `Vec<Box<dyn Element>>` → one
     `Box<WidgetSlots>` of 17 named concrete fields (boxed whole so
     registered pointers stay stable while `State` moves). Const-indexed
     sites reach fields directly; the genuinely index-driven paths
     (draw order, focus cycling, broadcast dispatch, the `*_IDX`-keyed
     HTTP API) go through `get_dyn`/`get_dyn_mut`; serialize takes a
     per-slot `dyn_refs` view (`serialize_widgets` now takes
     `&[&dyn Element]`). Static A/B byte-identical; cross-build final
     after an identical menu/toggle/wheel/click sequence within 6 px.
     TI's two-mode gallery followed (part 2): 53 named concrete slots
     for the gallery, typed enums for the child window's runtime-variant
     slots, a `Roster` enum carrying whichever mode runs; the numeric
     indexes keep addressing slots through `get_dyn`. **No app stores
     widgets behind `Box<dyn Element>` anywhere — the roster phase is
     complete.** Cross-build A/Bs: gallery empty masks static and after
     an identical interaction sequence; Ramp child at noise level;
     non-Ramp child arms hit the pre-existing unconditional
     Ramp-downcast panic identically on both builds.
     Cloud's `JsonControl` closed the owned-storage class (part 3): the
     JSON-config controls become a typed enum over `Adapted<T>` replacing
     `JsonWidget`'s `Box<dyn Element>`; the label walk dropped its
     `as_ptr` round-trip unsafe for a plain reborrow. **No owned
     type-erased widget storage remains anywhere in the workspace.**
   - **The borrowed-pointer retype design (6bc, decided 2026-07-12).**
     The replacement handle is **`WidgetId`, resolved through the
     generational `WidgetTree` at every use**. Rationale: apps own
     widgets concretely (the 6bb rosters) and re-register pointers
     idempotently per frame from boxed storage, so the registry is the
     one place a raw pointer is refreshed before use; every *other*
     stored `*mut dyn Element` bypasses that guard and is exactly where
     the real UAFs happened (display-manager 6ao — dangling registry
     from a by-value `new()`; settings 6w — `FOCUSED_WIDGET` surviving a
     rebuild via same-size alloc reuse). Under the retype, raw pointers
     remain only (a) as the `WidgetTree` registry payload and (b) as
     transient same-frame values inside resolution helpers; every stored
     reference becomes a `WidgetId`, and a stale id resolves to `None`
     and is skipped — the UAF class becomes unrepresentable outside the
     registry itself. Public signatures taking `&dyn Element` stay
     (deriving the id from `base()` internally) so most call sites
     survive verbatim; direct field readers convert compiler-driven.
     Slices, each independently shippable and A/B-verifiable:
     1. **Focus (DONE 2026-07-12)** — both stores
        (`UiContext.focused_widget` AND the `core.rs` thread-local
        `FOCUSED_WIDGET`) → `Option<WidgetId>`, kept as two stores with
        their existing reader sets (merging them changes observable
        focus behavior — not this phase's job). Thread-local fns that
        must dispatch `unfocus`/`FocusOut` gained a ctx param (every
        dispatching call site had one in reach). `set_focused` /
        `set_focused_ptr` refresh the registry with the pointer they
        are handed, so focus on a not-yet-registered widget keeps
        working; `is_focused_addr` → `is_focused_id` (base-id
        comparison). Verified: 168 tests; identical click sequences on
        text-editor (AE≤6, empty 8% masks) and data-editor (focus
        click AE=0); settings spinbox click-to-focus live.
     2. **Popovers + context-menu target (DONE 2026-07-12)** —
        `active_popovers: Vec<WidgetId>`, `ContextMenuState.target:
        Option<WidgetId>`; every occlusion/render walk resolves through
        the tree; `register_popover` takes `&mut` and self-registers;
        `show_context_menu` derives + registers the target id;
        `context_menu::mouse_input` takes the resolving ctx;
        `is_coordinate_covered` id-keyed with `WidgetId(0)` as the
        no-base sentinel; `EventCtx::widget_addr` deleted (use
        `ectx.id`). Verified: A/B vs slice-1 captures byte-equivalent
        (incl. the TE File-menu popover); DE leaf context menu Copy
        Key → wl-paste; settings page-dropdown popover switches pages.
     3. **Container child storage (DONE 2026-07-12 — by deletion, not
        retype).** The stored-field census found the slated vecs were
        ballast: `ParametersBg.children` was never populated anywhere
        (deleted with its parent-tracking twin, the container hooks,
        eleven dead iteration blocks, `collect_child_quads`, and
        window_runner's quad-attribution loop); `ScrollBox.children`
        was write-only (deleted with TI's one push). What still stores
        raw pointers after this slice: the `WidgetTree` registry (by
        design) and TI's app-side `ControlPanel.children` — live,
        pointing into boxed slots, consumed in ctx-less paint/arrange
        paths, so it retypes when those paths gain ctx (the endgame).
        Paginator's `container_children` is a transient field ref, not
        storage. The `Element::children`/`parent`/`add_child`/
        `set_parent` *signatures* still traffic in pointers, but every
        value is transient and tree-resolved at call time — they die
        with the `Element` endgame rather than warranting a standalone
        signature sweep.
     4. **`propagate_event(event, root: WidgetId)` — DONE (2026-07-13,
        the plumbing retype).** The handle rule it establishes: raw
        `*mut dyn WidgetHost` may appear ONLY as (a) the `WidgetTree`
        registry payload — the one ownership bridge, written at
        registration; (b) a registration argument derived from a live
        `&mut` (`register_widget`, `set_focused_ptr`-class
        self-registration — never stored); (c) machinery-internal
        transients resolved from the registry inside one call.
        Everything else crossing an API boundary carries `WidgetId` and
        resolves through the generational tree at use — a stale id is a
        loud no-op (`eprintln` canary), never a deref. Executed: the
        router resolves the root at the top of `propagate_event`
        (`propagate_event_impl` keeps its private resolved-ptr param);
        ~470 app dispatch sites across 14 apps went `.as_ptr_mut()` →
        `.id()` (field paths regex-converted; `let ptr = …` pairs,
        ptr-Vec collections, and dyn-roster receivers hand-converted);
        settings' `section_widgets`/`extra_dispatch_roots`/
        `page_dispatch_roots` retyped to `Vec<WidgetId>` with the
        keyboard section-focus block on `focus::is_focused_id`/
        `set_focused_id`; dead `WidgetPtr` + caller-less
        `register_popover_ptr` deleted. THE CONTRACT THE RETYPE
        SURFACES: a dispatch root must be REGISTERED. Most apps get
        registration as a `render_widget`/`paint_root_into` side
        effect; the canary caught every gap live: TI (roster never
        registered — per-frame `register_roster()`), settings chrome
        (wiped by `rebuild_layout`'s `clear_hierarchy` — re-registered
        after the view pass) + per-page rows rebuilt on data refresh
        (`AppPage::register_extra_dispatch_roots(ctx)` runs before
        each dispatch — the same liveness cadence the ptr router had)
        + custom-drawn menus (fonts/system pages), email (hand
        aggregate, per-frame block), authenticator (same), dm's bg
        root, LI's word-processor box, cloud (registers at its five
        dispatch sites), designer (frame re-registration skipped
        INVISIBLE slots while the wheel loop dispatches the whole
        roster). BUG FOUND: cce-graph registered its graph under a
        hand-minted `NEXT_WIDGET_ID` instead of the widget's own base
        id — `graph.id()` was unresolvable all along (focus/drag
        lookups on it silently failed); registration now uses the
        real id and the synthetic field is gone. Verified: 165 cce-ui
        tests + full workspace suite; canary-silent pointer/click/
        wheel probes over 13 apps and all 10 settings pages; TI
        dropdown→Grid relayout lands end-to-end through the id
        router; designer /state serves; demo's four event loops shed
        their unsafe self-alias entirely.
        FOLLOW-UP (same day): **`children`/`parent` left the trait
        (~52→50)** — tree structure is read off `ctx.tree`
        (`parent_id`/`parent_ptr`/`child_ids`/`children_ptrs`); no
        trait method returns a raw pointer anymore. `Adapted`'s
        container branch was redundant: Paginator (the one
        `Layout::container_children` implementor) tree-links its strip
        every tick via `register_embedded_children`, so the tree
        serves the walks identically (worst case a first-frame gap
        before the first tick). Machinery consumers (propagate
        descent, `find_hovered_scrollable`, the paint walk,
        `all_quads`/`all_rounded_quads` defaults, designer's render
        walks, TI's flat-walk parent skip) now read the tree directly
        — sanctioned transient-pointer class. DEAD CODE FOUND: both
        `navigate_focus` twins deleted — `UiContext::navigate_focus`
        had zero callers, and `focus::navigate_focus` (settings'
        ctrl-nav preamble) resolved parent/children through a
        freshly-made EMPTY UiContext, so it always returned false
        (parent has no field-derived form; ctrl+i needed a focused
        Paginator, which is never focusable). Settings' real ctrl-nav
        is its own section machinery, unchanged. serialize.rs's
        dummy-ctx child lookup could only ever surface Paginator's
        strip — kept via the 6aw concrete downcast. Verified: 165
        tests + workspace suite; email tab strip paints and a Sent
        click lands through the tree link; TI page-selector crop
        byte-identical (no double-draw); designer /state + full panel
        text intact; settings/files/demo canary-silent.
        FOLLOW-UP: **`as_ptr`/`as_ptr_mut` left the trait (50→48).**
        They live on as inherent `Adapted<W>` methods (the
        registration-bridge class) — every concrete call site resolved
        unchanged; dyn/generic receivers became the plain casts the
        impl always was (`w as *mut (dyn WidgetHost + 'static)`;
        `impl_widget_base!` no longer generates them). Designer's
        `find_widget_index` now honestly takes the thin `*const ()`
        address its body always compared. Verified: 165 tests +
        workspace suite; designer capture pixel-identical to the
        prior slice's (empty 8% mask) + /state live; TI click probe
        canary-silent.
        **THE DESIGNER EVENT REDESIGN — DONE (2026-07-13), closing the
        6bd deferral.** The resolution is a design decision, not a
        router conversion: the designer's event layer IS its own
        z-ordered windowing system (custom hit shapes with
        circular-pane overrides, hardcoded pane z, pane-focus
        derivation, unfocus rituals) and deliberately delivers through
        `handle_event` directly — the UiContext router's
        hit-gating/descent/drag tracking cannot own that policy, and
        mixing the two would double-run drag state machines. What the
        redesign fixes is the CONFLATION the deferral named:
        `drag_widget` no longer doubles as app-mode-drag marker. A new
        `AppDrag` enum (NetworkResize/ParamResize/SpreadsheetResize,
        each variant carrying its whole gesture state — dir, start
        rect/width/height, start mouse) owns the floating-pane edge
        resizes; the three `is_resizing_*` flags and four
        `drag_start_*` scratch fields are deleted; `drag_widget` only
        ever names a widget drag driven through the slot's Input drag
        hooks; exactly one of the two is armed per press. Release
        teardown, cursor hiding, path-change resets, and the
        hover-loop gate all read the split state; the hot-path
        MouseInput debug `println!` died as a rider. Verified: A/B vs
        stashed baseline — canvas-click and menu-open frames
        byte-identical (AE=0), id-stripped /state identical, launch
        frame differs by cursor sprites only (same-binary control
        AE=0). Held-gesture spot-check (2026-07-13, ccectl held-drag
        injection): node drag PASSED (Camera [1,1]→[3,3] via /state),
        NetworkResize right-edge PASSED, ParamResize left-edge PASSED,
        and SpreadsheetResize top-edge PASSED (the earlier "pane opens
        collapsed" report was a misdiagnosis: the HTTP menu_click driving
        the toggle was a SILENT NO-OP — see below — so the pane was never
        open; a real click opens it at its correct 250px default and the
        held top-edge drag grows it). Found instead: the HTTP `menu_click`
        action only reaches the MenuBar widgets' index-matched dispatch in
        window.rs (LEFT_MENUBAR's View menu has items 0-4), while the
        pane-toggle items ("Show Spreadsheet Pane" etc.) live in the
        OTHER menu system — the button-param menu pane drained by
        `sync_parameters_to_project`'s label match; out-of-range item
        indices replied "success" while doing nothing, and the handler's
        synthesized cursor park (-9999) sheds stray hover-diff pixels
        that can masquerade as the click's effect. The menu dispatch
        existing TWICE (index-matched in window.rs vs label-matched in
        app.rs, with diverging item sets) is the trap that produced the
        misdiagnosis. Both API holes are since FIXED (cce-designer
        b4ac763 + afe0f78): `menu_click` validates its indices against
        the target menubar's real dropdowns (a non-menubar widget_idx
        used to PANIC the app) and echoes the clicked label; the label
        match is extracted to `State::execute_menu_action`, and the new
        `{"action":"menu_action","label":"Show Spreadsheet Pane"}` drives
        the menu-pane items directly (verified: pane toggles on at its
        exact rect via curl, toggles off byte-identical). The network pane's breadcrumb-strip "panel
        move" is CONFIRMED INERT: the press arms
        `drag_widget = NETWORK_PANEL_IDX` on `Adapted<PassivePlate>`
        ("no children and no events"), so DragUpdates land on a widget
        with no drag hooks — dead scaffolding, DELETED (cce-designer
        00e148a: both press arms, the always-false hit_menubar block,
        the NETWORK_PANEL_IDX drag-driver arms, and the two is_dragging
        layout read-backs; border/strip presses keep their real focus+
        consume behavior — regression-verified live). And the
        spot-check caught a real press-routing bug the A/B frames could
        not: VIEWPORT_IDX and PARAM_IDX shared the -4 z tier, and the
        stable sort's index tiebreak (3 < 6) sent EVERY press over the
        floating params pane to the viewport — slider/Float3 held drags
        were dead end-to-end (only the scrollbar, armed via its own
        press path, worked). Fixed in cce-designer 4b78aa5 (PARAM_IDX
        gets its own -3 tier above the viewport it floats over);
        live-verified Position X 2.50→10.00 via /state. The
        `draggable`/`is_dragging` trait methods still have this
        cascade + TI's ControlPanel as dyn consumers — they leave the
        trait with the CP endgame.
        **THE CONTROLPANEL ENDGAME — DONE (2026-07-13). The last
        stored child-pointer surface is gone, and `draggable`/
        `is_dragging` left the trait (49→47).** TI's ControlPanel is
        DISSOLVED to scroll chrome (~120 lines: ScrollBox + bg/border
        paint + drag hooks): its `Vec<*mut dyn>` children, label-
        matched arrangement, aggregate views, dummy-ctx event/tick/
        drag forwarding, and scroll-translated coordinates are all
        deleted. The app owns the panel now: `arrange_control_panel`
        lays the child slots at SCREEN coordinates (scroll offset
        applied at layout time, re-run every frame — the wheel moves
        content on the frame it repaints); `display_list` emits child
        geometry/text clamped to the panel viewport with the legacy
        partial-clip radius-zeroing and border-inset rules; children
        dispatch as ordinary routed roots. THE GATE THE DISSOLUTION
        REQUIRES: children at real rects are hit-testable even when
        clipped below the fold — `cp_gate` (panel rect ∪ open child
        popovers, the legacy `ControlPanel::hit`) gates the press
        pre-scan, release broadcast, and wheel; keys stay
        focused-path-only; the panel takes the wheel before its
        children (legacy scroll-frame order). BUG THE GATE FIXED
        LIVE: a fold-hidden StatusBar toggle stole the page-selector
        press. FOUND: the legacy panel double-drew its scrollbar
        (rounded AND plain aggregate views — the 6p/6v class);
        single-drawn now, thumb correctly dimmer. `draggable`/
        `is_dragging` became inherent `Adapted` reads; the two
        index-driven rosters (TI 53+5 slots incl. child-mode enum
        variants, designer 17) route them through generated per-slot
        matches; 14 UFCS test forms became dot calls. Verified live:
        Windows page A/B vs pre-dissolution baseline (static +
        scrolled + popover states — masks empty except the scrollbar
        single-draw strip), dropdown item select updates the Surface
        Info description end-to-end, spinbox +/- increments, wheel
        scrolls with content following, page switching intact both
        ways, Controls page unchanged, child mode alive; 28-target
        suite. Scrollbar thumb drag: spot-check PASSED (2026-07-13,
        ccectl pointer-press/release): the held thumb drag scrolls the
        Windows-page ControlPanel with content tracking the drag —
        Create Window/Width/Height/Window Type scroll off as the
        Window Elements/Border/Bevel tier comes in, thumb following.
     5. **window_runner render plumbing + remaining `as_ptr` sites —
        DONE (2026-07-13, the last slice).** The slice-3 census was
        right: window_runner held no pointer state (its one mention
        was a doc comment). The real residue was the paint walk's
        entry signatures — `paint_tree`/`paint_root_into` now take
        `&dyn WidgetHost` (the walk only reads; descent resolves
        children through the registry and derefs those transients
        internally), and `append_widget_text`'s lifetime-erasing
        transmute died with the ptr param it bridged to. Every caller
        simplified: the unsafe self-alias blocks that existed ONLY to
        mint `*mut` arguments (text-editor, graph, data-editor, the
        demo, fonts' walk calls) are plain shared borrows now;
        `render_widget`'s internal cast is gone. What still carries
        `*mut dyn WidgetHost`, all deliberate and documented: the
        WidgetTree registry payload + registration arguments (the
        ownership bridge), machinery-internal walk transients, the
        `Layout::arrange_children`/`container_children` hook
        signatures (narrow-trait, ctx-less by design), and
        EventCtx's transient host ptr. Verified: 28-target suite;
        text-editor/graph/data-editor/fonts live captures render
        fully, canary-silent. THE POINTER-RETYPE PROGRAM IS
        COMPLETE — no further slices are recorded.
        ~~then the `Element` + `Adapted` endgame (own design pass)~~
        — landed long since as the 6bd flip (`Element` deleted,
        `Adapted` survives as the one host wrapper).
     Stored-pointer state remaining after slices 1–3, all deliberate:
     the `WidgetTree` registry; TI's `ControlPanel.children`; and the
     tick-refreshed parent copies in MenuBar/StatusBar/Dropdown models
     (`parent_changed`/`tracked_parent` — written each tick by the
     re-parenting pattern with the live host pointer, read in ctx-less
     popover-direction/corner-radius math). Endgame option for the
     parent copies: snapshot the *data* read through them (parent rect,
     radius, is-Ramp flag) at re-parent time instead of the pointer —
     same refresh cadence, no deref of potentially-dead memory; watch
     the one-frame rect lag on resize if reads move to snapshots.
   - **The legacy tuple views retired (the same day).** The five apps moved to `paint_self`
     (cce-secrets, the greeter, the settings app's dropdown and search box, the gallery, the
     designer's render loop), then the toolkit's own readers: `ParametersBg`, the ramp and the
     menubar read their children's `painted_prims`; `Adapted::paint_self` computes the
     TextBox's focus highlight itself; the bridge and `corner_radii` read `Paint::corner_style`;
     `append_widget_plate` stopped drawing the widget's arcs; `paint_legacy_leaf` and
     `widget_vertices` / `push_widget_vertices` had no callers left. The views and the hooks
     that served only them (`serves_legacy_plain_quads`, `legacy_plain_quads`,
     `aggregates_child_extra_quads`, `forwarded_highlight`) are deleted, and the tests that
     asserted on them read `painted_prims`. Pixel A/B in a scale-2 shadow: the designer (five
     pane layouts) and the gallery identical before and after the toolkit change; each app's
     own move is described in its commit (cce-secrets and the greeter changed look by choice,
     the settings dropdown gained its face plate, the gallery lost a stray outline).
   - **WidgetHost 38 → 31 (the same day).** The legacy tuple views (`extra_quads`,
     `extra_arcs`, `extra_circles`, `all_quads`, `all_rounded_quads`, `highlight_quad`,
     `corner_style`) moved to `WidgetHostExt` too, computed from the paint model exactly as
     `Adapted` did (its private `content_rect` / `painted_prims` / `visible_children` went
     with them). Five apps still draw through them; moving those to `paint_self` is what
     retires the views themselves, and it changes how some look (cce-secrets draws its
     controls flat through them). Every app that reads them drew identically to the pixel.
   - **WidgetHost 57 → 38 (2026-10-08).** The trait had grown back past the blueprint's ~55
     with the keyboard and accessibility hooks, and most of what `Adapted` implemented was a
     one-line forward to a narrow trait. The narrow traits are dyn-compatible, so the host now
     hands its widget out as them (`layout_model` / `paint_model` / `input_model` /
     `input_model_mut`) and `WidgetHostExt` (blanket over every host, `dyn` included) carries
     the 19 forwards and 3 derivations (`label`, `corner_radii`, `mark_dirty`) as provided
     methods; `plate_bevel`, never overridden and so always `None`, went. What stays on the
     trait adds host state to the model's answer. Apps changed only by importing the
     extension trait; the designer, the demo, the gallery, cce-files, the data editor and
     the settings app drew identically to the pixel before and after (scale-2 shadow).
   - **The `Element` endgame design (6bd, decided 2026-07-12).** The
     endgame is a **trait replacement, not an app rewrite**. Grounding
     facts (consumer survey): direct per-method dispatch
     (`mouse_input`/`cursor_moved`/`keyboard_input`/wheel on concrete
     `Adapted` fields) exists in nearly every app — making routed
     events a prerequisite would gate the endgame on ~12 app
     migrations, so the direct-dispatch surface *stays on the new
     trait* and shrinks later as apps move to routed events at their
     own pace. Designer's index-driven roster broadcast needs only a
     small dyn set (unfocus, drag hooks, set_modifiers, z_index,
     hit_test, visibility, tick, prepare_text, focused, as_any). The
     legacy tuple getters' consumers are the `T: Element` generics in
     layout.rs (render_widget/Column/Section — settings, LI, files,
     colors render paths) and window_runner's tessellators
     (widget_vertices — designer/TI): generic, not dyn, so they can
     re-bound onto a narrower bound without touching the dyn surface.
     Target shape: a **`WidgetHost` trait of ~40 methods** implemented
     once by `Adapted<W>` (blanket over the narrow traits). `Adapted`
     does NOT die — it survives as the single host wrapper owning base
     state; what dies is the 92-method god-trait and its Option-base
     escape hatches (`base()` becomes a guaranteed `id()`/state
     access, killing the `WidgetId(0)` sentinel class). Phases, each
     shippable:
     1. **Capability actions → one enum method (DONE 2026-07-12)**:
        the 13 context-menu action methods (`cut_selection`…
        `copy_path`) became `context_action(ContextAction) -> bool`
        across Element, the Input hooks, and the Adapted forwards; the
        Input default keeps whole-value Cut/Copy/Paste through the
        value-string pair; TreeList's seven action bodies moved to
        inherent methods via an impl split (zero code movement);
        core.rs maps the menu strings (incl. the load-bearing "Cear")
        to enum values. Element is down to 79 methods. Live-verified:
        tree Copy Key/Copy Value/Collapse and TextBox Paste through
        the enum dispatch.
     2. **Tuple getters — DISSOLVED into the flip (measured
        2026-07-12).** The premise was wrong on two counts. (a) The
        dyn consumers are real: designer's render loop reads
        `extra_quads`/`extra_circles`/`extra_arcs`/`color` directly
        off its `&dyn` handle AND feeds it to the tessellators, and
        the paint walk's legacy-leaf branch reads
        `all_quads`/`all_rounded_quads`/`extra_*`/`widget_font` via
        dyn — the visual surface must ride the host trait object.
        (b) A separate `LegacyVisual` trait can't be reached from the
        existing trait object (no cross-trait-object casts; a
        supertrait split trips over `rect`/`color` defaults needing
        `base()`). The genuinely generic consumers (layout.rs
        render_widget/Column/Section, settings'
        `collect_window_child`, dm/cloud's local tessellator copies
        with concrete receivers) simply re-bind at the flip.
     3. **Tree-link methods — landed pre-flip after all (2026-07-12).**
        `children`/`parent` are core machinery walk methods (propagate,
        painter, navigate) — host-trait material, they stay.
        `clear_children` left in batch 2; **`set_parent`/`add_child`
        left in batch 4**: both are inherent `Adapted<W>` methods now
        (files' concrete sites resolve unchanged); the dyn callers
        were only three — `focus::link_parent_child`'s body (rewritten
        as the register + `tree.link` + `tree.set_parent` ops the pair
        always was), TI's page-selector/StatusBar roster pair (now one
        `link_parent_child` call), and TI ControlPanel's per-arrange
        dummy-ctx child re-parent (deleted — every effect was
        discarded with the dummy ctx, the same inert-ritual class as
        Ramp's tick re-parents). A/B: TI gallery (flat-walk `parent()`
        skip intact — no page-selector double-draw) + dm greeter, both
        empty 8% masks. Note Paginator both serves
        `container_children` AND `link_ids`-registers its strip — the
        ctx-less walks (popover_rect/prepare_text/render_popover) are
        why the field-derived form must stay.
     4. **The flip — DONE (2026-07-12): `Element` is deleted; the
        trait is `WidgetHost`.** Landed in two shippable halves:
        **(a) the base() guarantee** — `base`/`base_mut` return
        `&Widget`/`&mut Widget` (no Option), killing the escape hatch
        and the `WidgetId(0)` no-base sentinel class;
        `as_any`/`as_any_mut`/`as_ptr`/`as_ptr_mut` became required
        (their defaults manufactured DummyAny/null-DummyElement
        stand-ins nothing could use); every Option-handling call
        site collapsed 1:1 to direct reads (12 repos); the
        layout/arena/tree test mocks grew a base field via
        `impl_widget_base!`. **(b) the rename** — 634 word-boundary
        occurrences across 18 crates; the workspace compiled on the
        first pass. `ElementState` (input enum) keeps its name;
        cce-layout-interface's local `Element` document enum was
        already alias-insulated (`Element as UiElement`). Since the
        shrink batches had removed every non-blueprint method first,
        the rename IS the retype — the registry/context/painter/
        window_runner signatures all read `dyn WidgetHost` now.
        Verified: 163 tests; settings audio render stream
        byte-identical across BOTH halves; files + data-editor A/B
        AE=0; live settings page-dropdown popover → Fonts page
        switch. The trait sits at ~65 methods; the remaining
        shrink-later blocks (direct-dispatch, value, as_ptr
        transitional, `preferred_height`/`value` dyn consumers)
        thin out per-app as routed events / concrete slots spread.
     5. **The routed-events tail — DONE (2026-07-13).** All 12 apps
        (+ TE/DE/demo from 6ab–6ad) dispatch through
        `propagate_event`; self-routing composites (Paginator, whose
        tree-registered strip would consume its presses under the
        router's children-first descent) go through `handle_event`;
        only designer's press/move cascade stays direct, by recorded
        design (its `drag_widget` doubles as app-mode drag with
        circular-pane hit shapes). En route, the ROUTED-DRAG GAP was
        found and fixed: the router's DragStart/DragUpdate/DragEnd
        fell into `Input::on_event`'s default and every routed drag
        was silently dead — `Adapted::handle_event` now maps them
        onto the Input drag hooks (regression test drives a full drag
        through `propagate_event`).
     6. **Direct-dispatch block census (2026-07-13) — the collapse
        design.** With every app-side dispatch caller gone, the
        block's remaining consumers are: (a) cce-ui widget-INTERNAL
        forwards — composites driving embedded children (ramp,
        parameters_bg, treelist, dropdown, menu, paginator,
        breadcrumb, color_selector, scroll_box, TI's ControlPanel,
        cloud's json_layout); (b) designer's deferred cascade (dyn
        roster calls); (c) ~40 tests (UFCS `WidgetHost::` forms);
        (d) `Adapted`'s own entry-point impls (die with the methods).
        THE COLLAPSE: every remaining caller rewrites as
        `handle_event(&Event::…)` — behavior-identical by
        construction (the entry points literally forward there, and
        the drag fix routes `Event::Drag*` to the hooks) — then
        `mouse_input`, `cursor_moved`, `on_cursor_moved`,
        `mouse_wheel`, `keyboard_input`, `drag_begin`, `drag_update`,
        `drag_end` leave WidgetHost (8 methods, ~67→59). TWO
        CAVEATS: (1) `Adapted::keyboard_input`'s `!visible()` gate
        must MOVE INTO `handle_event`'s KeyInput arm (designer's
        hidden-widget broadcast relies on it; the routed path
        currently lacks it — moving it also fixes that latent
        inconsistency); (2) `Adapted`'s PointerMove arm calls the
        trait `cursor_moved` internally — inline the
        coverage-gate + hover-recompute body as inherent before
        deleting. The QUERY/POLLING surface (`draggable`,
        `is_dragging`, `take_click`, `take_change`, value getters)
        stays — no Event form; dies with typed messages (§3.5) or
        container dissolutions. ~100 call sites, one session.
        **DONE (2026-07-13): WidgetHost 67→59.** The eight left the
        trait; in-crate composite forwards to concrete embedded
        children resolve unchanged through the inherent `Adapted<W>`
        entry points (the batch-1 recipe — far cheaper than the
        feared 100 rewrites); dyn callers (designer's cascade, TI's
        ControlPanel child forwards, cloud's json_layout slider
        drag) build the equivalent `Event` and call `handle_event`.
        Both caveats landed: the keyboard `!visible()` gate lives in
        `handle_event`'s KeyInput arm (closing the routed path's
        missing-gate hole), and the coverage-gated `cursor_moved`
        default became the inherent `Adapted::cursor_moved`. The
        trait's default `handle_event` serves test shims only.
        Verified: 164 tests; settings render stream byte-identical;
        TI interactive four-state A/B empty masks; designer /state
        identical across a canvas click.
        ~~The MenuBar/StatusBar/Dropdown parent-pointer
        snapshot change rides this phase.~~ **Landed early
        (2026-07-12): the census showed all five stored widget-side
        parent pointers production-DEAD** (nothing ever set_parent's
        a MenuBar/StatusBar; ramp's per-tick re-parents fed a
        write-only field through a dummy ctx — legacy behaved the
        same). TextBox.parent deleted outright; Dropdown.parent
        became `parent_snapshot` read-DATA (rect/is_ramp/color, same
        direct-write activation, Ramp-clamp test adapted) and its
        write-only `tracked_parent` died; MenuBar/StatusBar lost the
        fields, their `parent_changed`/`tracked_parent` overrides,
        MenuBar's never-firing `adjust_rect` clamp, and now report
        the 0.0 corner radius production always read; the
        `Layout::parent_changed`/`tracked_parent` hooks are deleted
        (implementor-less), `Adapted::parent` is tree-only, and the
        Ramp/ColorRamp tick_ctx re-parent rituals are gone. Stored
        `*mut dyn Element` survives ONLY in the WidgetTree registry
        payload, EventCtx's transient host ptr, and TI's
        ControlPanel. A/B: text-editor + TI Ramp-child static AND
        preset-popover-open frames all empty 8% masks; 163 tests. **Measured blueprint (~55
        methods, from the machinery's actual call sites):**
        identity/tree — id (guaranteed, no more `Option<&Widget>`),
        type_name, label, as_any/as_any_mut, as_ptr/as_ptr_mut
        (transitional), visible/set_visible, z_index,
        is_child_visible, children(ctx), parent(ctx), add_child
        (transitional, TI);
        layout — rect, set_rect, measure, layout, label_x_offset,
        set_row_rect;
        events — handle_event, hit_test, mark_dirty, tick/wants_tick,
        set_modifiers, focus/unfocus/focused, context_action,
        blocks_backplate_drag, is_scrollable, plus the
        direct-dispatch block (mouse_input, cursor_moved, mouse_wheel,
        keyboard_input, drag_begin/drag_update/drag_end/is_dragging/
        draggable, take_click, take_change) — shrinks per app as they
        move to routed events;
        value — get_value_string/set_value_string, set_text,
        set_selected — shrinks as app loops go concrete-slot;
        paint — paint_self, clips_children, renders_own_subtree,
        prepare_text, popover_rect, render_popover, dirty-flag access;
        visual tuples (walk legacy branch + designer loop) — color,
        all_quads, all_rounded_quads, extra_quads, extra_circles,
        extra_arcs, corner_radii, plate_bevel, solid_border,
        widget_font, highlight_quad.
        Everything else on today's Element (79 methods plus the
        generic-only surface) either moves to inherent `Adapted<W>`
        methods for the generic render machinery or dies.
        **Execution mode: shrink Element IN PLACE toward the blueprint,
        then rename it to `WidgetHost` when it matches** — a parallel
        trait can't be reached from the existing trait object, but
        removing non-blueprint methods one census-driven commit at a
        time keeps every state shippable. **First shrink batch (DONE
        2026-07-12, 79→73):** `highlight_color` folded into the
        `highlight_quad` default (zero overrides); `set_drag_bounds` +
        `intrinsic_size` moved to inherent `Adapted<W>` methods (their
        concrete callers — designer's network panel, fonts'/graph's
        hand-laid sizing — resolve unchanged); `layout_style`/
        `layout_children` deleted with **scene/bridge.rs itself** (its
        last production user was retired in 6aa; the narrow
        `Layout::intrinsic_size` hook stays — `Adapted::measure` reads
        it); `layout_ignore` deleted with its only consumers, the
        uncalled `layout_widgets`/`layout_widget_ptors`. Census
        lesson: grep BOTH `.method(` and UFCS `::method(` forms — the
        fonts/graph `Element::intrinsic_size(&x)` callers only
        surfaced at compile. **Second shrink batch (DONE 2026-07-12,
        73→69):** `hovered`/`set_hovered` deleted — the state is the
        base `Widget::hovered` flag, read/written directly by the
        `cursor_moved`/`on_cursor_moved`/`highlight_quad` defaults and
        `serialize.rs`; Button/Checkbox keep inherent accessors for
        immediate-mode hosts (cloud's json_layout downcasts to
        concrete `Checkbox`, so it already resolved to those).
        `corner_radius` + `rounded_corners` replaced by ONE
        `corner_style() -> (f32, (bool,bool,bool,bool))` mirroring the
        narrow `Paint::corner_style` — NOT folded into `corner_radii`,
        which is lossy: the radius is meaningful with every corner off
        (Menu/StatusBar report their parent's radius to children
        through the flags-off channel; breadcrumb can be
        flags-true/radius-0, whose legacy radius-0 rounded bg quad
        would vanish). `clear_children` moved to an inherent
        `Adapted<W>` method (every caller is a concrete Adapted field
        in cce-files). Consumer commits: settings renderer,
        TI ControlPanel aggregates, cloud json_layout. Verified:
        settings audio render stream byte-identical, files A/B AE=0,
        TI gallery empty 8% amplitude mask, 163 tests.
        Census facts for the leftovers: `preferred_height` has DYN
        consumers (container_layout.rs child-ptr walks + layout.rs
        machinery on `&dyn` children) — blueprint-adjacent, rides the
        flip, not inherent-movable; `value` has a live dyn consumer
        (`serialize.rs` over designer's `dyn_refs()`) — rides the
        flip. Remaining non-blueprint candidates:
        `on_cursor_moved` (belongs in the direct-dispatch block —
        blueprint addition, not a deletion), `set_parent` (flip
        material, with the parent-ptr snapshot change), `value`,
        `preferred_height`.
     7. **The value/polling block — DONE (2026-07-13): WidgetHost
        59→52.** This is the §3.5 "typed messages" resolution, and it
        lands the way 5k's controller half did: no app-defined message
        channel is needed — the polling drains stay concrete (inherent
        `Adapted<W>` forwards to the narrow `Input` hooks), and what
        dies is reaching them through the host trait. Seven methods
        left: `take_click`, `take_change`, `get_value_string`,
        `set_value_string`, `value`, `set_text`, `set_selected`.
        Census: five had ZERO non-test dyn consumers (the old
        designer-side serialize consumer of `value` is gone; the
        in-crate `widget/display/serialize.rs` inspector feed was the
        one live reader — now a concrete downcast chain over the five
        `Input::value` implementors Checkbox/Dropdown/Slider/
        RangeSlider/Spinbox, pinned by a unit test that fails if a new
        implementor is missed). The dyn readers of the rest went
        concrete-slot: TI's index-driven roster reads route through
        app-local `Roster::take_click/value/get_value_string/set_text
        (idx)` matches onto the concrete gallery slots (arms exist per
        drained slot; an unwired slot panics loudly); cloud's
        `JsonControl` grew an inherent variant-matched `take_click`
        (both call sites already gate on the button type); designer's
        pane-focus menubar loop writes `set_selected` on its five
        concrete `Adapted<MenuBar>` fields. The UFCS test forms became
        dot calls resolving to the inherent methods. STILL on the
        trait, each with live dyn consumers: `draggable`/`is_dragging`
        (designer's deferred press/move cascade + TI's ControlPanel
        child pointers), `preferred_height` (layout.rs container
        machinery) — these ride the designer event redesign / CP
        dissolution. Verified: 165 tests (new serialize pin);
        settings audio render stream byte-identical vs the stashed
        baseline; TI live probe — Button/Toggle clicks, Layout
        dropdown popover open, and a "Grid" selection re-laying out
        the gallery through the new roster drains end-to-end.
        RIDER: the `Control` subtrait (set_label + control_label) is
        DELETED — zero dyn consumers, zero `control_label()` callers;
        every impl just routed `set_label` to the inherent shadow, so
        the deletion is call-site-invisible (compile-verified across
        the workspace).
     Former slices 4/5 fold in: the app `as_ptr_mut` dispatch sites
     are rewritten by whichever of routed-events (per app) or the
     phase-4 flip reaches them first; no standalone pointer-to-id
     signature sweep.
 
 - **Phase 7 — Plate unification: backplate becomes a ROLE of Plate (7a/7b/7c stages DONE 2026-08-25; open questions below).**
   Finish what 6as/6at began. The `Backplate` and `Plate` container widgets are deleted and the
   `is_backplate`/`is_movable_backplate`/`is_plate` flags are folded, but "backplate" survives as
   a second vocabulary for what is now one concept — a lit base surface (`Prim::Plate`). What
   remains under the old name: the `style.surface.backplate.*` config namespace and its getters
   (`backplate_{padding,gap,color,blur,corner_radius}`, the menubar/statusbar sub-styles); a
   partial merge already in the tree (`layout::plate_corner_radius()` falls back to
   `backplate_corner_radius`); and — the real content — a ROLE: "the plate that meets the window
   edge" (window-background drag via `blocks_backplate_drag`/`drag_allowed_at`, MenuBar/StatusBar
   carving into it, the compositor clipping every window at the span-widened backplate radius).
   Each app also hand-rolls its root-surface painting from the backplate getters (`DemoApp`
   in `src/main.rs` is the reference copy; every client repeats a variant).
 
   **Motivation.** One surface concept instead of two makes plates fully compositional: a plate
   can be the base surface of a window OR a child of another surface, with nothing but role data
   distinguishing them. The concrete driver is detachable plates — cce-designer's detached panes
   already behave exactly like this (a pane plate becomes a new window's root plate; its interior
   corners become window corners), but the geometry lives app-side in
   `cce-designer/src/render.rs::pane_plate_radii` and the detach machinery is designer-only.
 
   **Design.** No widget returns (6as stays won). The unification lives in the paint/geometry
   layer: a `PlateSpec` — rect, per-corner radii, color, blur, plus role flags:
   `window_corners: (bool, bool, bool, bool)` (which corners lie on the window silhouette) and
   `drag_background: bool` (whether uncovered area is a window-drag region). The toolkit computes
   per-corner radii from the flags (a window corner wears
   `window_corner_radius() * corner_span_factor()`, an interior corner wears
   `plate_corner_radius()` — the `pane_plate_radii` math, moved in from the designer), and the
   engine paints any plate root-or-nested through the one path, absorbing the per-app hand-rolled
   root painting. A window's base surface is just a plate whose four corners are all window
   corners.
 
   **Invariants.**
   - *The window silhouette stays a shared cross-process contract.* The compositor clips windows
     from the SHARED corner value; per-plate radius freedom must never leak into a
     `window_corners=true` corner (the designer's config.kdl radius override already documents
     this trap at its `pane_plate_radii` call site). The role flags are where the constraint
     lives: flagged corners read the shared value, period.
   - *Blur regime follows the role.* A root plate frosts against the compositor's blur-behind
     (the negative-alpha marker convention); a nested plate blurs against app content. Detaching
     moves a plate between regimes; the marker choice keys off the role flags, and this is the
     subtlest part of the phase — it gets its own design note before code.
   - *No root container widget.* `PlateSpec` is data consumed by the paint path, not a node that
     owns the window.
   - *The droplet family rides the plate PUSH-CONSTANT block, not the plate concept.*
     `Prim::Droplet` (shader2d `MODE_DROPLET`) reuses the plate block's fields by mode-10
     reinterpretation only — it deliberately does not consume `PlateSpec` or the backplate
     getters. 7b may reshape how plates are DESCRIBED, but the shader-side field packing is
     shared: changing the plate block means re-checking the droplet arm. If droplets are ever
     folded into a generalized plate role, two external contracts must survive: the
     `module { droplet "k=v" }` spec-string idiom, and the compositor's scenefx droplet node,
     which parses the same `DropletSpec`.
 
   **Stages.**
   - **7a — Vocabulary. REVISED at implementation, 7a-1 DONE (2026-08-25).** The original
     text said "`style.surface.plate.*` becomes canonical" — implementation surveying found
     `style.surface.plate.*` ALREADY EXISTS as the NESTED-plate style namespace (padding,
     color, border_color, border_thickness, blur) carrying deliberately different values
     from `backplate.*`; a flat alias would have merged root styling into pane styling.
     The canonical namespace is therefore role-scoped: **`style.surface.plate.root.*`**
     (with `plate.root.menubar.*` for the bar sub-style) — truer to the phase's thesis
     anyway: backplate = plate in the root role. Getter names follow as `root_plate_*`.
     - **7a-1 DONE.** Both config paths aliased: the layout style-registry table maps
       `plate.root.*` rows onto the same slots as `backplate.*` (slot names keep the
       historical prefix — invisible), and color.rs's JSON-pointer loads are canonical-first
       chains (`/style/surface/plate/root/…` `.or_else(` `/backplate/…)`), so the new
       spelling WINS when both are present; in the registry table both spellings write one
       slot and document order decides (single-spelling configs — all real ones — are exact).
       Canonical getters (`root_plate_{padding,gap,opacity,corner_radius}`,
       `root_plate_{menubar,statusbar}_{color,text_color,blur}`) with the old `backplate_*`
       names as plain delegating wrappers — NOT `#[deprecated]` yet: 15 crates + the
       compositor still call them (16-crate caller census in the 7a-1 commit). "root" joined
       `PROP_NODES` for the config-editor path helpers. cce-ui's own callers (config.rs
       tests, DemoApp, cce-relief, cce-ramp) migrated. Tests: the legacy styling test now
       reads through canonical getters (legacy-config → canonical-getter equivalence), plus
       a canonical-spelling test proving parse, precedence over legacy, and legacy-only
       fallback. 245 lib tests green; designer A/B AE=0.
     - **7a-2 — DONE (2026-08-25).** All caller crates migrated per-repo and the
       `backplate_*` wrappers flipped to `#[deprecated]`. Census correction: the compositor's
       9 census hits were all its OWN vocabulary (serde fields + a local default fn) — zero
       cce-ui getter calls; what it actually needed was the CANONICAL KDL alias, since it
       parses the silhouette block from the shared config.kdl itself (`plate { root ... }`
       accepted canonical-first, legacy `backplate` unchanged, tested both ways —
       cce-compositor@0ef2901). Implementation trap for the record: exact-match renaming of
       `backplate_corner_radius()` also matched the compositor's
       `default_backplate_corner_radius()` calls while its serde `default = "..."` string
       attribute did not — audit renames for substring collisions against local wrappers.
       The original census, by call sites:
       cce-compositor `server/config.rs` (9 — reads the SHARED silhouette values; the
       migration must not change which slot it reads), cce-files (10 across main/
       preview_pane/pages), cce-test-interface (11), cce-data-editor (5), cce-terminal (3),
       cce-graph (2), cce-fonts (2), cce-color-editor (2), cce-cloud (2), cce-text-editor
       (1), cce-system-interface `main.rs` (1), cce-authenticator (1). DONE: cce-designer
       (vk-smoke, the exemplar). CONFIRMED CLEAN, nothing to migrate: cce-status-interface
       (its bar styling reads `module { }` keys + `/style/status/*`, not the root-plate
       getters — verified by its owning session 2026-08-25). The
       `Application::is_movable_backplate_at` trait-method NAME is 7b vocabulary
       (behavioral role naming), not 7a's. With 7a-2 done, 7a is COMPLETE: new code uses
       `root_plate_*` / `plate.root.*`; the deprecated wrappers and the legacy config
       spelling were kept for out-of-tree configs.
     - **7a-3 — DONE (2026-09-06).** Every live config (shared, cce-graph, cce-designer,
       cce-notifier's per-app `plate { }`) was rewritten to the canonical spelling, then the
       aliases came out: the `backplate.*` pointer/registry read-aliases in cce-ui, the
       compositor's `backplate` node fallback, cce-grid's and cce-notifier's fallbacks, the
       `#[deprecated] backplate_*` getters, and `backplate` in `PROP_NODES`. A `backplate`
       block in a config is now silently ignored.
     - **7a-4 — DONE (2026-09-06).** The internal vocabulary followed: registry slots
       (`root_plate_*`), the `ROOT_PLATE_*` statics, `set_root_plate_*` setters,
       `read_root_plate_opacity_if_configured`, the trait methods
       `Application::is_movable_root_plate_at` and `Widget::blocks_root_plate_drag`, the
       compositor's `root_plate_{color,blur,corner_radius}` config fields (and the flat-form
       keys of the same name), cce-gallery's `RootPlate` exhibit and `--root-plate` flags,
       and the comment prose across every crate. "backplate" survives only in this RFC's
       history and in test fixtures that prove the legacy spelling is ignored.
   - **7b — `PlateSpec` + window-corner math toolkit-side.** Introduce the spec, port
     `pane_plate_radii` in, and give the engine a root-plate paint path fed by a spec instead of
     each app's hand-rolled quads (DemoApp first, then the clients). The designer's per-pane
     plates convert to specs with computed role flags. A/B: AE=0 per app.
 
     **Blur-regime design note (required before code; written 2026-08-25).** Two frost
     regimes; the role selects between them:
     - *Root plate* (all four corners on the window silhouette): the fill stays
       POSITIVE-alpha translucent; the COMPOSITOR frosts what lies behind the window
       (`plate.root.blur` in the shared config drives cce-fx's blur-behind). The app
       draws no frost of its own.
     - *Nested plate* (any interior corner): frost is the NEGATIVE-ALPHA sentinel on
       the fill — the in-app vk frost pass blurs app content drawn BEFORE the plate
       (draw-order-dependent by design; see `param_plate_fill`). A pane touching some
       window edges is still nested for blur purposes: it frosts app content.
     `PlateSpec` therefore stores `color` with positive alpha plus `blur: bool`, and
     `fill()` applies the role-correct encoding: root → alpha forced non-negative,
     nested+blur → alpha negated. DETACH is exactly a role flip: interior corners become
     window corners, `fill()` flips regimes, and the formerly-frosted app content beneath
     simply does not exist in the new window. No other app-side blur change is needed.
 
     **Radii rule.** A window-flagged corner wears
     `window_corner_radius() * corner_span_factor()` (the SHARED silhouette curve — the
     invariant); an interior corner wears `plate_corner_radius()` (the app-overridable
     pane value). This is `pane_plate_radii` verbatim, moved in.
 
     - **7b-1 DONE (2026-08-25).** `PlateSpec` in `scene/paint.rs` (rect, positive-alpha
       color, `blur`, per-corner `window_corners`, perimeter `depth`) with
       `window_corner_flags(rect, win_w, win_h)`, `radii()`/`radii_for()`, role-aware
       `fill()`, and `PaintCtx::plate_spec`. DemoApp's hand-rolled root plate migrated —
       DELIBERATE visual correction: its radius was the un-spanned
       `root_plate_corner_radius`, so under squircle `corner_shape` its perimeter shading
       detached from the compositor's span-widened clip; the spec snaps it to the
       silhouette (demo AE≠0 expected and intended). The designer's `pane_plate_radii`
       delegates to the toolkit (A/B AE=0). Two 7a stragglers the getter census could not
       see (raw JSON-pointer reads, not getter calls) gained the canonical-first chain:
       `layout::window_corner_radius`'s shared-config read, and cce-grid's silhouette
       read.
     - **7b-2 — DONE (2026-08-25).** Nine clients migrated: files, terminal,
       system-interface, authenticator, data-editor (Prim::Plate hand-rolls →
       `plate_spec`, plus data-editor's concentric `corner_frame` now follows the
       silhouette), and graph, fonts, color-editor, text-editor (non-Plate root emissions —
       rounded_rect/border/legacy tuples — take their values from the spec via the new
       `layout::window_silhouette_radius()` scalar, which `radii_for` also uses). A/B
       revision: the original "AE=0 each" predates 7b-1's discovery that migrating IS a
       correction — every app moved off the un-spanned radius, so corners change by
       design. Verified: seven apps diff ONLY within 120px corner squares; the eighth
       (system-interface, hardcoded r=12 → silhouette, the largest jump) also shifts the
       perimeter roll's edge gradient, eyeball-confirmed as the arc correction.
 
       **CORRECTION (2026-08-25, post-7c): the Plate-group half of that story was
       inverted — a double-span, since fixed.** `plate_spec` fed the spec's FINAL
       radii into `Prim::Plate`, whose contract is NOMINAL radii spanned downstream
       by `plate_push_raised(scale_corners = true)` — so the five Plate-group
       clients drew window corners at span² (12 → ~100 logical at n=4.5). Those
       apps had been CORRECT all along (nominal in, spanned once by the push); the
       "arc correction" the A/B eyeball accepted was the regression itself, caught
       when the user reported corners rounder than the desktop grid. Measured on a
       live corner diagonal: clip/grid arc at the expected 0.202·span depth, plate
       arc ~2× deeper. The four scalar-group clients (graph, fonts, color-editor,
       text-editor) tessellate without a downstream span, so for them the spanned
       scalar was and remains the genuine correction. Fix: `plate_spec` pre-divides
       by `corner_span_factor()` so the push's multiply reconstructs the spec's
       exact values; unit test `plate_spec_emission_round_trips_the_span` guards
       it. The same double-span reached DemoApp via 7b-1 — its "AE≠0 expected"
       diff bundled the genuine correction WITH the overshoot.
       Authenticator is values-only verification (never launch it in a shadow — it claims
       the PolicyKit D-Bus name). SKIPPED deliberately: cce-cloud (overlay popup windows —
       whether they share the decorated-window silhouette was an open question, since
       CLOSED: **yes** — resolved 2026-08-25 post-7c, cce-cloud@4829451. The compositor
       never clips layer surfaces, so the app's drawing IS the overlay's silhouette, and a
       launcher-sized panel at the nominal radius read nearly square beside real windows;
       its root emission is now a PlateSpec with all four corners window-flagged, verified
       pixel-identical corner depth to a real window's plate) and
       cce-test-interface's `Backplate` gallery shim (a legacy-lookalike test fixture;
       migrating it would defeat its purpose — still skipped, still deliberate). Designer pane EMISSION: radii and the
       nested-blur sentinel already flow from spec-derived values; full spec-OBJECT
       emission is deferred into 7c, because pane plates carry focus tint and
       widget-driven bevel styling `PlateSpec` does not yet model — detach will dictate
       whether the spec grows those fields or the widget hooks stay authoritative.
       7c builds on the role flip.
   - **7c — Detach/dock generalization.** Lift the designer's plate-corner control, collapse,
     and dock-drag onto `PlateSpec` so any app can offer them. The detached-window PROCESS model
     and sync channel (`default_project.json` polling) remain app policy — the toolkit provides
     the plate-role flip (interior→window corners, blur regime swap, CSD hookup via the existing
     `standard_csd`/`take_window_action` hooks), not the process management.
     - **7c-1 DONE (2026-08-25).** `widget::plate_dock`: the app-agnostic PROTOCOL —
       constants (control radius/inset, min plate span, stub height, drag threshold),
       `PlateDockState { collapsed, detached }` with `stubbed()`, `corner_center(rect,
       stubbed)` (rect placement incl. the stub exemption; the designer's circular-pane
       arc placement stays app policy), `corner_hit`, `press_becomes_drag`, and
       `standard_menu(state, can_detach)` → (label, `PlateDockAction`) rows the host
       appends its own items after (the designer's spreadsheet span modes). Deliberately
       LEAN: no container type, no widget — a second consumer decides those.
       `PlateSpec::detached()` is the role flip (all corners→window, radii snap to the
       silhouette, `fill()` swaps frost regimes), unit-tested. The designer delegates:
       constants re-exported, rect placement/hit/threshold/standard-menu all
       toolkit-calls; PLATE_SLOTS membership, visibility, dock regions, layout
       application, and the detach process spawn stay designer policy. Verified live:
       corner menus (Collapse/Detach; Reattach-only stubs), the collapse→stub→expand
       cycle, and dock-drag arming+drop all behave identically through the delegation.
     - **7c-2 — DONE (2026-08-25): cce-files is the second consumer.** Its preview pane
       carries the corner control; Collapse narrows every page's pane column to a Preview
       title stub (the list keeps the freed width — the feature's actual value), the
       stub's control restores the prior split fracs, and the divider/wheel are inert
       while collapsed. The adoption answered the open API questions:
       - `draw_corner_dot(pc, center, emphasized)` EARNED — both hosts drew the identical
         dot; the designer now uses it too.
       - A `PlateDock` CONTAINER did NOT earn its place: one `PlateDockState` field
         sufficed. Revisit only if a host manages many dockable plates outside its own
         state arrays.
       - Press arming was designer-specific after all: it exists to disambiguate click
         from dock-DRAG, and a single pane has nowhere to dock — files opens the menu
         directly. Moreover, files keys the whole interaction on mouse RELEASES: its
         routed-widget path consumes left PRESSES before the `Application` hook (only
         releases reliably arrive there), a per-app dispatch reality any adopter must
         check first. The designer's release-opened menu means the two feel identical.
       - The detached-window CSD packaging question is ANSWERED (2026-08-25,
         cce-files@805be06 — a working, end-to-end-verified preview-pane detach;
         RETIRED the same day at the user's direction, cce-files@f2a89d5:
         detach is designer-only for now. The answers below were derived from
         that implementation and stand; the code is one revert away at 805be06
         for whichever app adopts detach next): **no new toolkit packaging was
         needed.** A detached window is an ordinary
         `Application` whose root plate carries the detached role (all window
         corners); the compositor's decoration IS the CSD in this DE (border =
         grab surface, no titlebars), and plate_dock's existing pieces —
         `corner_center`/`corner_hit`/`draw_corner_dot`/`standard_menu` — cover
         the control. What the second implementation DID establish as the
         convention worth naming: **the child's Reattach is process exit** — the
         parent `try_wait`s and reclaims the pane, so every way a detached
         window can die reattaches it; and the child exits itself when its sync
         file or parent pid disappears, so orphans (crash, stale session
         restore) self-collect. The sync-file idiom generalizes: parent pid
         then payload, written before the spawn, unlinked on reattach.
         One packaging gap surfaced and stands as app policy for now: the
         toolkit context menu dispatches through a widget tree a minimal
         detached window does not have, so the child draws its one-row menu
         itself — a third consumer hitting this earns a widget-tree-free menu
         helper.
       - The designer pane spec-OBJECT emission question got its answer from the
         same implementation: detach dictated NO new spec fields — the files
         pane detached with `PlateSpec` as it stands, so focus tint and
         widget-driven bevel styling stay widget-hook territory and the spec
         stays lean. The designer's emission migration remains optional and
         unblocked.
 
   All numbered stages DONE (2026-08-25), and all three open questions CLOSED the same
   day: cce-cloud overlays share the silhouette (7b-2); the first non-designer detach
   (cce-files' preview pane — built, verified, then retired by product choice the same
   day, see 7c-2) answered CSD packaging — nothing new was needed, the conventions are
   recorded there — and established that detach dictates no new `PlateSpec` fields,
   leaving the designer's spec-OBJECT emission optional and unblocked. The designer
   remains the only detach host. Phase 7 is COMPLETE.
 
   **Design note (recorded 2026-08-30) — why carve GROUPING exists, and why its
   rarity is correct.** The mechanism predates this phase: it landed with the SDF-lit
   plate system (cce-ui@3877dd6), and its rationale lived only in that commit message
   and the `tessellate_display_list` doc comment until now. The tessellator promotes a
   `Recess` emitted while a `Plate`/`Bevel` is still "open" into a **CSG feature of
   that plate's single draw** (the per-frame feature UBO); every other carve renders
   through the standalone overlay branch. Two things justify the dual path:
 
   - *Junction correctness at the perimeter roll.* A grouped plate is ONE composite
     height field — the rolled edge minus its carves — lit once per pixel from summed
     analytic slope vectors, so a carve wall meeting the plate's perimeter roll is an
     arithmetic junction. The overlay branch approximates that meeting with the
     host-box fade.
   - *Features never blend in color space.* An overlay is shading drawn over
     already-lit pixels, so stacked shading double-counts — the same reason
     `Prim::Ridge` exists rather than a boss+recess pair (double-counted specular at
     the crest).
 
   The corollary that makes the design coherent: **the two paths differ visibly only
   near the host's rolled perimeter.** An interior carve (a TreeList or TextBox well
   in the middle of a window plate) never touches the roll, so the fallback is
   effectively exact there. Two scope corrections recorded 2026-08-30 (the first
   version of this note got them wrong): what groups is **full-ring untinted carves**
   (button grooves, slider wells) — the flush menubar/status bands never group, by
   design since cce-ui@80d50de: an edge-suppressed carve's wall rect extends past the
   boundary, relying on the overlay cover quad to clip it, a clip the grouped
   whole-plate draw does not have (grouped, the extended walls smeared across the
   plate). So the junction where a flush band meets the plate's roll is ALWAYS the
   host-box fade; grouping's value is the single-evaluation lighting of full-ring
   carves, exact wherever one sits near the roll. And per the audit that shipped
   `CCE_PLATE_DEBUG` (cce-ui@949e35e, three apps): **no misgrouping — every fallback
   is a documented rule firing correctly**; grouping is rare (demo 2 of 10, cce-files
   0 of 7) because apps constantly interleave flat fills with reliefs, and each one
   correctly closes the grouping window (the carve's shading is baked into the
   plate's earlier draw). Neither path is retirable: grouping-always is impossible
   for exactly that reason, and fallback-only would forfeit the exact junctions
   full-ring carves get when they do group.
 
   The standing hazard is the *silent flip*: three of the six grouping conditions are
   dynamic (draw order, sibling plates, whether another plate claimed the host's
   feature run), so the same widget can render through either path depending on its
   surroundings. That shipped as a bug once — a hovered button's opaque fill severed
   every later button from the backplate they carve into — fixed by making the
   carve-host tracker a stack (`plate_stack`, cce-ui@9cfadad). `CCE_PLATE_DEBUG=1`
   reports each carve's verdict, the fallback reason, and which prim closed a
   grouping window; it is the first tool for any "same widget, different look"
   report. Hardening (landed 2026-08-30): debug builds warn loudly — once per
   geometry, no env var — when a groupable full-ring carve is enclosed by a
   still-open plate, its shaded region reaches that plate's roll band, and a dynamic
   rule (occlusion, feature-run contiguity, budget) rejected it: the one class where
   the flip is visually significant (`near_roll_fallback_reason` in
   `backend/window_runner.rs`, unit-tested). Deliberately a warning, NOT an assert:
   the audit established every rejection is conservative-correct — the render is
   right, it is the frame-to-frame look that flips — so a panic would crash debug
   builds on correct behavior. The ubiquitous accepted case (ordinary geometry
   already closed every grouping window) stays quiet by construction: no open
   enclosing plate remains for the check to run against.
 
 - **Design note (2026-09-01) — the scroll-virtualization contract, and the ScrollRegion
   de-duplication sweep.** A class fix, recorded because the class outlived every
   individual sighting of it. `ScrollBox::get_item_draw_y` returned a row's position
   only when the row was FULLY inside the viewport, so callers drew nothing for a row
   straddling the edge — cards/rows visibly vanished mid-scroll. The helper predates
   the §3.4 clip stack (when there was no way to draw a row "cut", culling whole was
   the only option), and the contract then traveled: the struct around it was copied
   into cce-system-interface, and from there into cce-fonts, cce-mail, cce-cloud, and
   cce-layout-interface as each dissolved its List/ScrollBox embedded base (Phase 6q),
   plus reimplemented in cce-files' RowList — the multi-repo copy-drift failure mode,
   in widget form. The bug was then rediscovered and fixed **per app**
   (cce-system-interface first; cce-cloud@1063422 fixed only the CLICK half), which is
   exactly the cost the sweep exists to stop paying.
 
   The sweep (one commit per repo, 2026-09-01): the fixed system-interface copy was
   lifted verbatim into **`widget::ScrollRegion`** (cce-ui@764d9e7) — the union of all
   the copies' APIs — so the contract lives in one place: **`get_item_draw_y`/
   `get_draw_y` return every row that INTERSECTS the viewport; callers draw those rows
   under a clip (the §3.4 stack, or exact per-quad clamping for flat pipelines), and
   hit-test the SAME partial rows the draw shows** — visible ⇒ clickable, culled ⇒
   not; a fix to only one half just mirrors the bug (1063422's blank-band click gate
   became the sliver-selects-correctly gate with no shape change once the draw side
   caught up). `ScrollBox::get_item_draw_y` itself moved to the intersection contract.
   Deliberately NOT migrated: TreeList's `get_row_rect` keeps full containment — it
   places a floating overlay that draws over the well unclipped, and an editor hanging
   half off the list edge is worse than one that waits — and TreeList's row-bottom
   separator gate, which is spatial (the separator would land outside the well), not a
   cull.
 
   Two pipeline lessons from the app migrations, for anyone adding a scrolled list:
   the clip must survive to EVERY stage that renders row content, or partial rows
   bleed instead of cut. cce-mail's hand-emitted row labels ride a boundless labels
   drain — they drew whole into the menubar until given viewport bounds; cce-cloud's
   span assembly dropped the paint walk's merged clip on the floor (`bounds: None`)
   until it was threaded through to the glyph pass. And where widgets keep their full
   rect while drawing cut (cce-layout-interface's row buttons above other controls),
   the stored hit rect is re-clamped to the visible sliver after the visuals are
   recorded, so the hidden part cannot shadow what's beneath it. New scrolled lists
   build on `widget::ScrollRegion`; hand-copying it back into an app is how this
   class got six lives.
 
 Order rationale: each phase is independently valuable and reversible, and no phase requires the
 next to compile. Phase 0 can land immediately regardless of the rest.
 
 ---
 
 ## 7. Quick wins to land first (Phase 0 detail)
 
 1. **Breadcrumb black rectangle — DONE.** `cce-files` `BrowseContainer::set_rect` now positions
    the breadcrumb to exactly match `browse::view`'s layout (inset by the page margin, reserving
    the dropdown width) so the container's duplicate paint sits fully behind the page copy
    instead of leaking a dark strip. This is a stop-gap; the real fix is the single paint path in
    Phase 3 (the breadcrumb is still painted twice — the copies now just coincide).
 2. **Hot-path debug I/O — DONE.** Removed the per-frame `eprintln!` in `render()` that
    reconstructed every text area's string via `layout_runs()`.
 3. **Dead code — DONE.** Removed `UiContext::tick_hover`/`get_hover_quad` (zero callers
    workspace-wide), the dead duplicate of `hover_animation`.
 4. **GPU scissor — DEFERRED to Phase 3.** Threading clip rects to `render_pass.set_scissor_rect`
    is not actually a "quick win": clip rects are computed CPU-side and folded into geometry
    today, with nothing carried to the render pass. Doing it properly needs the clip stack from
    §3.4, so it lands with the paint-pass rework rather than as a risky standalone change.
 
 ---
 
 ## 8. Risks & mitigations
 
 - **Migration surface across ~19 apps.** Mitigation: adapter shim + per-app Phase 6; the core
   lands and is validated before any app is forced across.
 - **Borrow-checker friction with an arena tree.** Mitigation: layout operates on
   `style`/`layout_out`, not the `widget` payload; `get_disjoint_mut` for the rare dual-borrow.
 - **Hand-rolled layout correctness.** Grow/shrink/wrap/alignment are subtle. Mitigation: keep
   the box model small (row/column + flex + align + gap/padding only), and unit-test the solver
   in isolation — it operates on `Style`/`Size`, independent of paint, so it is directly testable.
 - **Effort.** This is multi-week. Phasing keeps every intermediate state shippable so it can be
   paused/resumed without a broken tree.
 
 ---
 
 ## 9. Deferred (designed-for, not built now)
 
 - **Per-subtree geometry caching:** cache tessellated vertices per node, re-tessellate only
   dirty subtrees instead of the whole scene each frame. The arena `Dirty` flags are the hook;
   worth it only once scenes are large.
 - **Full affine transforms / rotation** beyond translate+scale.
 - **Damage-rect partial redraw** at the GPU level (currently full-surface clear each dirty frame).
 
 ---
 
 ## 10. Decisions (resolved 2026-07-07)
 
 1. **Layout solver: hand-rolled** (not taffy). Compact measure/arrange engine owned in
    `cce-ui`, scoped to the DE's box model. See §3.2.
 2. **A new shared dependency in `cce-ui` is acceptable** when needed (one shared path dep does
    not break standalone builds). Note the layout decision means no layout dep is required.
 3. **Phase 0 quick wins land now**, as separate commits ahead of the rebuild. See §7.