test(ui): refresh two stale baselines and settle whether they gate

`make ui-visual` had been red on main since #27, and nothing runs it,
so four references had drifted across three unrelated merges. Two were
refreshed with #186; these are the other two.

Each recorded two changes, not one. `app-sidebar` lost Jobs (#27,
shipped) and moved its highlight from Home to Tracks; `now-playing`
gained the source line (shipped) and was playing from "a dynamic mix".
Both are singleton stores read by a case that sets nothing, so the shot
photographs whatever the case above it left behind — blessing that
would have pinned the file's own ordering into a PNG. Both cases state
their world now, and only then are the references re-recorded.

The second half of the issue asks whether this tier should gate, and
the answer is measured rather than preferred: replayed in a bare
ubuntu:24.04 container — CI's `check` image — three of the ten
baselines fail on rendering alone (`track-info` and one `page-header`
shot at ratio 0.03 against a 0.02 allowance, `seek-bar` one pixel
shorter), and the two stale ones disagree about their new height
between the machines. So CI cannot run this suite without a second,
container-recorded baseline set that every local run would then fail
against, and a pre-push hook is the same fault with the machines
swapped. It stays local and opt-in; what replaces the gate is the rule
that a change moving a component's geometry refreshes that component's
baseline in the same commit, having read the image, and never one it
did not cause. Written where a person meets it: the skill's tier doc
has the table, SKILL.md has the obligation, CLAUDE.md has the
constraint.

Deleting the baselines was the third option and is declined: this tier
has caught one thing no other could, the `<span>` that lost the UA
stylesheet's `box-sizing` and grew a badge 36→38px.

Closes #196

.pi/skills/yellowjacket-dev/SKILL.md

@@ -158,7 +158,7 @@ only climb when it cannot.
 | You changed | Run | Cost |
 |---|---|---|
 | A Lit component, a store, the shortcut service | `make ui-test` | ~2 s, no app |
-| …and it renders differently | `make ui-visual` | + 6 baselines, opt-in |
+| …and it renders differently | `make ui-visual` | + 10 baselines, opt-in, never gates |
 | Any Go code | `make test` | 3 passes, ~2 min |
 | A service that emits events | `make test` — assert on the payload, see `backend/queue/emit_test.go` | in-process, no app |
 | A bound method or a bound struct field | `make bindings` then `make ui-test` | ~1.5 s + 2 s |
@@ -180,6 +180,13 @@ less than it looks.)
 
 Two rules about climbing:
 
+- **If you moved a component's geometry, run `make ui-visual` and
+  refresh that component's baseline in the same commit.** Nothing else
+  will: it is the one tier in this repo no hook and no CI job runs, and
+  it cannot be one — its references are machine-specific, measured in
+  [references/ui-tier.md](references/ui-tier.md). Four of them drifted
+  across three merges before anyone noticed (#196). Read the image;
+  never bless a reference you did not cause.
 - **A component test passing is not the app rendering.** If you touched
   anything in `frontend/src`, verify it in the real app too — start it
   headless, `screenshot --filename=/tmp/shot.png`, and *read the PNG*.

.pi/skills/yellowjacket-dev/references/ui-tier.md

@@ -78,9 +78,58 @@ synchronously.
   Microtasks and not a timer, deliberately: a timer hangs forever under
   the suites that install fake ones.
 
-Visual baselines are font-hinting and compositing sensitive, which is
-why they are opt-in: they only mean anything on the machine that
-recorded them.
+## The visual tier does not gate, and that is measured (#196)
+
+`make ui-visual` is the same suite with nine `toMatchScreenshot`
+baselines switched on. **Nothing runs it but a person**, deliberately,
+and the reason is a number rather than a preference: the committed
+baselines were recorded on Arch, and replayed in a bare `ubuntu:24.04`
+container — CI's `check` image — three of them fail for reasons that
+have nothing to do with any component.
+
+| baseline | Arch | ubuntu:24.04 |
+|---|---|---|
+| `page-header` filtered-by-search | passes | ratio 0.03 differ, against a 0.02 allowance |
+| `track-info` | passes | ratio 0.03 differ |
+| `seek-bar` | 1152×18 | 1152×17 |
+
+The two references that were genuinely stale did not even agree about
+their *new* size — `now-playing` renders 1152×65 on Arch and 1152×64 in
+the container. So moving CI's `check` job from `make ui-test` to
+`make ui-visual` is not a one-line change: it needs a second,
+container-recorded baseline set, which every local run would then fail
+against. That is the same trap the other way round, and a pre-push hook
+is the same fault again — one machine's baselines against everybody
+else's renderer.
+
+So the tier stays local and opt-in, and the rule that replaces the gate
+is:
+
+- **A change that moves a component's geometry refreshes that
+  component's reference in the same commit, having read the image.**
+  Look at the PNG; the dimensions in the failure message are the cheap
+  half of the answer.
+- **Never refresh a reference you did not cause.** #196 exists because
+  four of them drifted across three unrelated merges, and every red run
+  made the next person likelier to stop running the tier than to read
+  it.
+- **State the world the shot is taken in.** The stores are singletons,
+  so a visual case that sets nothing photographs whatever the previous
+  case left behind — which is how the sidebar's baseline came to have
+  Tracks lit and `now-playing`'s to be playing from a dynamic mix.
+- **`make ui-visual-update UI_ARGS=<path>` does not filter** and
+  re-records *every* baseline, blessing any stale one in silence:
+  vitest's `--update` takes the following positional as its value. Until
+  #204 lands, record one file with
+  `cd frontend && YJ_VISUAL=1 npx vitest run --update=true <path>`, and
+  check `git status` before committing either way.
+
+What the tier is worth, for the record: it is a *layout* check, blind to
+colour (the component tier has no `:root`, so it renders the fallbacks —
+`make ui-visual` passed unchanged through a whole palette rewrite,
+twice), and it has caught one thing nothing else could — swapping
+`library-status-indicator`'s `<button>` for a `<span>` lost the UA
+stylesheet's `box-sizing` and grew the badge 36→38px.
 
 ## Bindings
 

.planning/NOTES.md

@@ -4912,49 +4912,3 @@ bridge leaves the wizard up with its "Get Started" button correctly
 disabled — it gates on a directory chosen *in the wizard*, and the
 existing-library check runs once, on mount. A reload clears it. Nothing
 is broken; it cost twenty minutes of believing a tap had been swallowed.
-
-## The sheet's scroll fade, and where a scrim may not go (measured 2026-08-23, headless)
-
-#207's answer. The affordance is two background layers on
-`wa-dialog::part(body)` and the conditionality is
-`background-attachment`, not a scroll listener: a cover of the sheet's
-own colour painted at the end of the *content* (`local`) over a shadow
-pinned to the box (`scroll`), so the cover scrolls up and hides the
-shadow exactly when there is nothing more to see.
-
-Measured at 424x360 (which is where a menu overflows on `main`, since
-`main` does not yet carry #67's eighth item — at 424x439 the track
-list's seven items are `scrollHeight` 364 against `clientHeight` 364,
-fitting exactly). Pixel at x=300, dark ramp, `bgElevated` `#343a40`:
-
-| y | before | more below | at the end of the list |
-|---|---|---|---|
-| 330 | 52,58,64 | 50,56,62 | 52,58,64 |
-| 340 | 52,58,64 | 43,48,53 | 52,58,64 |
-| 350 | 52,58,64 | 33,37,40 | 52,58,64 |
-| 359 | 52,58,64 | 22,24,27 | 52,58,64 |
-
-Three things worth keeping.
-
-**A menu that fits draws nothing**, which is the same measurement: at
-424x439 the sheet is flat 52,58,64 to its bottom edge, because with no
-overflow the `local` layer's positioning area *is* the padding box and
-the cover lands on top of the shadow.
-
-**A scrim over a menu row is that row's text surface**, so the 4.5:1
-rule reaches it and this is why the curve is steep rather than linear.
-A row is 48px with its label centred; 32px of scrim already down to a
-quarter strength at 14px puts about 0.06 at the label. Checked on the
-light ramp (`bgElevated` `#e9ecef`, text `#212529`) by overriding the
-two custom properties on `:root`: background at the label 205,207,210,
-which is **9.9:1**. The first draft — a linear 48px at 0.8 — put ~0.375
-on that label, 5.0:1, passing but visibly greyed. The bottom few pixels
-go to ~2.4:1 in either draft and are deliberately below where any
-label of a *fully visible* row sits; a label that lands there belongs
-to the half-cut row, which is the thing being signalled.
-
-**A dark scrim on a dark surface reads far worse in a shrunk screenshot
-than on screen.** The first two probes (24px/0.45, then 32px/0.75) were
-measurably present — 52,58,64 down to 30,33,37 — and invisible in the
-inline preview. Crop the bottom 70px and scale it up before judging;
-the pixel values are the honest answer either way.

CLAUDE.md

@@ -1402,7 +1402,7 @@ descendants. On the reference device the main panel spans 0-318 of a
 items cut off, with no way to reach them. `showModal()` is Chrome 37
 and uses the real top layer, so a dialog is immune by construction.
 
-Seven things about it are load-bearing.
+Six things about it are load-bearing.
 
 **"Dialogs are fine" needed checking, because every other dialog in
 this app is mounted in `index.html`** — outside `.main-panel` — so it
@@ -1431,25 +1431,6 @@ doing nothing, which reads as the gesture breaking. `menu-dismiss` is
 that signal; the three surfaces that do not use `ContextMenuController`
 bind it themselves.
 
-**A sheet that scrolls says so, and `background-attachment` is what
-asks whether it does** (#207). The sheet is capped at 85vh — a surface
-covering the whole screen is a page, not a sheet — so a long menu's
-body scrolls, and for three phases it scrolled *silently*: measured at
-424x439, eight items ended at y=470 with the fold at 439, and where the
-cut lands on a row boundary the sheet ends in a clean edge that reads
-as the end of the list. The fade is two background layers on
-`wa-dialog::part(body)` — a shadow pinned to the box (`scroll`) under a
-cover of the sheet's own colour painted at the end of the *content*
-(`local`), which scrolls up over the shadow exactly when there is
-nothing more to see. So it is absent on a menu that fits, present the
-moment one does not, and gone again at the end of the list, with no
-scroll listener and nothing reaching into `wa-dialog`'s shadow root for
-the scroller. **The curve is steep because the rows under it stay
-live**: a scrim over a menu item is that item's text surface, and the
-4.5:1 rule applies to it — 32px already down to a quarter strength at
-14px spends its weight below the last legible label, measured at 9.9:1
-on the light ramp, whose `bgElevated` is `#e9ecef`.
-
 **The playlist submenu is a sheet too, and it had to be.** It is a
 `placement="right-start"` flyout, and making the menu full-width moved
 its anchor — measured at x −182 to 0, entirely off-screen, so "Add to
@@ -3443,6 +3424,25 @@ Pre-commit hooks verify generated code is fresh — always run `make generate` a
 Tests use `database.NewTestDB(t)` for in-memory SQLite, built by the same
 `applySchema` production uses so the two cannot diverge. Test audio fixtures live in `test_data/music_library_test/`. Table-driven tests are the norm.
 
+**`make ui-visual` is the one tier nothing but a person runs, and it
+cannot become one.** Its ten `toMatchScreenshot` baselines were recorded
+on a developer's Arch box; replayed in a bare `ubuntu:24.04` container
+— CI's `check` image — three of them fail on font metrics and
+compositing alone (`track-info` and one `page-header` shot at a 0.03
+mismatch ratio against a 0.02 allowance, `seek-bar` one pixel shorter),
+and two components disagree about their own height between the two
+machines. So CI keeps running `make ui-test`, which is the same suite
+with the comparisons off, and a pre-push hook would be the same fault
+with the machines swapped. What replaces the gate is a rule, in
+`.pi/skills/yellowjacket-dev/references/ui-tier.md`: **a change that
+moves a component's geometry refreshes that component's baseline in the
+same commit, having read the image, and never one it did not cause**.
+That is #196, which was four stale references accumulated across three
+unrelated merges — a red tier nobody could read, which is how it stayed
+red. A visual case must also **state the world it photographs**, since
+the stores are singletons and a case that sets nothing records whatever
+the previous one left in them.
+
 ## Git Workflow
 
 Feature branches and PRs are the only way in: **`main` is a protected

frontend/src/components/menu-surface/menu-surface.ts

@@ -155,55 +155,10 @@ export class MenuSurface extends LitElement {
            bottom was at y=452 on a 439px screen -- the one row a
            destructive action is most likely to be. The cap has to stay
            (a sheet covering the whole screen is a page, not a sheet),
-           so the body is what gives.
-
-           **And a body that scrolls says so** (#207). Scrolling was the
-           whole of the fix above, which left the last item reachable
-           and nothing on screen admitting it was there -- measured at
-           424x439, eight items ending at y=470 with the fold at 439,
-           and worse when the cut lands on a row boundary, where the
-           sheet ends in a clean edge that reads as the end of the list.
-
-           Two layers, and the *order* is what asks the question: a
-           shadow pinned to the bottom of the box (attachment scroll),
-           and over it a cover of the sheet's own colour painted at the
-           end of the *content* (attachment local), which therefore
-           scrolls up over the shadow and hides it exactly when there is
-           nothing more to see. So the affordance is absent on a menu
-           that fits, present the moment one does not, and gone again at
-           the end of the list -- with no scroll listener, no
-           measurement, and nothing reaching into wa-dialog's shadow
-           root for the scroller. background-attachment is Chrome 4;
-           the reference device is Chrome 113.
-
-           **The curve is steep because the rows under it stay live.**
-           A scrim over a menu item is that item's text surface, and
-           this app's rule is that text clears 4.5:1 on every surface it
-           can sit on -- which the light ramp, whose bgElevated is
-           #e9ecef, is what makes non-theoretical. A row is 48px with
-           its label centred, so 32px of scrim that is already down to
-           a quarter strength at 14px reaches y-centre at about 0.06 and
-           spends its weight on the strip below the last legible label.
-           Measured on the dark ramp at x=300, flat 52,58,64 throughout
-           before: 50,56,62 at y=330, 33,37,40 at y=350 and 22,24,27 at
-           the bottom edge, and flat again at the end of the list. The
-           light ramp puts 9.9:1 on the last label. */
+           so the body is what gives. */
         wa-dialog::part(body) {
             padding: 0;
             overflow-y: auto;
-            background:
-                linear-gradient(
-                        var(--yj-bg-elevated, #343a40),
-                        var(--yj-bg-elevated, #343a40)
-                    )
-                    bottom / 100% 32px no-repeat local,
-                linear-gradient(
-                        to top,
-                        rgba(0, 0, 0, 0.6) 0%,
-                        rgba(0, 0, 0, 0.25) 45%,
-                        rgba(0, 0, 0, 0) 100%
-                    )
-                    bottom / 100% 32px no-repeat scroll;
         }
 
         /* A sheet is dragged at with a thumb, so it says where its top

frontend/test/components/chrome.test.ts

@@ -137,6 +137,11 @@ describe('<app-sidebar>', () => {
   });
 
   it('looks the way it did last time', async () => {
+    // Stated rather than inherited: `activeViewStore` is a singleton, so
+    // without this the shot records whichever view the *previous* case
+    // left in it and the reference moves when the file is reordered.
+    activeViewStore.setView('home', true);
+
     const el = await fixture('app-sidebar');
 
     await visual(el, 'app-sidebar');

frontend/test/components/menu-surface.test.ts

@@ -206,56 +206,6 @@ describe('menu-surface', () => {
       expect(dismissed, 'no menu-dismiss reached the document').toBe(1);
     });
 
-    /**
-     * The scroll affordance (#207), and this is the mechanism again
-     * rather than the symptom.
-     *
-     * The sheet's body has scrolled since #60 and said nothing about
-     * it: measured at 424x439, eight items ended at y=470 with the
-     * fold at 439, and where the cut lands on a row boundary the sheet
-     * ends in a clean edge that reads as the end of the list.
-     *
-     * What makes the fade *conditional* — absent on a menu that fits,
-     * present the moment one does not, gone again at the end of the
-     * list — is `background-attachment`, not a scroll listener: a cover
-     * of the sheet's own colour is painted at the end of the content
-     * and attached `local`, over a shadow pinned to the box and
-     * attached `scroll`. So the pair of attachments *is* the feature,
-     * and it is what this asserts. The rendered result was measured in
-     * the harness (dark ramp 52,58,64 flat before; 52,57,63 at the last
-     * label and 22,24,27 at the bottom edge with more below; flat again
-     * at the end of the list) and is on the PR.
-     */
-    it('paints the fade only while there is more below', async () => {
-      const el = await surfaceWithPanel();
-
-      const wrapper = el.shadowRoot?.querySelector('wa-dialog');
-
-      await (wrapper as HTMLElement & { updateComplete: Promise<unknown> })
-        .updateComplete;
-
-      const body = wrapper?.shadowRoot?.querySelector('[part~="body"]');
-
-      expect(body, 'no body part to scroll').not.toBeNull();
-
-      const style = getComputedStyle(body as Element);
-
-      expect(style.overflowY, 'the body is what gives, not the cap').toBe(
-        'auto',
-      );
-
-      // The cover scrolls with the content; the shadow does not. Either
-      // one alone is a fade that is always there or never there.
-      expect(
-        style.backgroundAttachment,
-        'the cover must be local and the shadow must not',
-      ).toBe('local, scroll');
-
-      // Both sit at the bottom, or the cover hides nothing.
-      expect(style.backgroundPosition).toBe('50% 100%, 50% 100%');
-      expect(style.backgroundSize).toBe('100% 32px, 100% 32px');
-    });
-
     /**
      * A dialog with no accessible name is what `utils/name-dialog.ts`
      * exists for; here the name is already written on the panel, so no

frontend/test/components/now-playing.test.ts

@@ -366,6 +366,16 @@ describe('<now-playing>', () => {
     const el = await fixture('now-playing');
 
     emit(Events.TrackChanged, { ...TRACK, trackChangeId: 6 });
+    // Stated rather than inherited: the queue store is a singleton, so
+    // without this the shot records whichever source the *previous*
+    // case left in it and the reference moves when the file is
+    // reordered. Three lines is what the bar renders while playing
+    // from somewhere, which is the arrangement worth recording.
+    setQueue([queueTrack(1, 'Ashes to Ashes')], 0, {
+      type: 'album',
+      id: 7,
+      label: 'Scary Monsters',
+    });
     await flush();
     await el.updateComplete;