Close SLICE-014 and SLICE-015, and settle the two selector calls

docs/SLICE-014-analog-controls.md

@@ -1,6 +1,6 @@
 # CUETools Linux Slice Brief: SLICE-014 analog controls
 
-Version: 0.2. Date: 2026-08-21. Status: Approved for build.
+Version: 1.0. Date: 2026-08-23. Status: Done.
 Companion documents: ARCHITECTURE.md, ENGINEERING.md, DECISION-LOG.md
 (D-075/D-076 icon language, D-077 house voice).
 
@@ -104,7 +104,18 @@ No persistent data.
   name (renamed out of reach), and the lamp lens ran thin (widened).
 - [x] Suite green (143) including the accent-key contrast pins (label
   vs both gradient stops, both themes; disabled exempt per WCAG 1.4.3).
-- [ ] Owner's live eyeball on the real app.
+- [x] Owner's live eyeball on the real app (2026-08-23). It found the
+  one thing this slice's own guardrail had told it to protect, and the
+  slice had broken anyway: section 10 says never make a control's state
+  less legible than before, and "the unpowered read must never cost the
+  user the ability to see a control exists". The dead key's legend had
+  been left on the structural `Line` brush, which measured **1.2:1** on
+  the dark face and 1.5:1 on the light one. That is not dim, it is
+  invisible, and nothing caught it because WCAG exempts disabled
+  controls so nothing had ever measured them. Fixed and bounded both
+  ways by test as D-087.
+  The same pass replaced the hover outline with the housing lamp
+  (D-088), which supersedes this slice's non-geometric hover cue.
 
 ## 10. Agent guardrails for this build
 
@@ -119,17 +130,24 @@ No persistent data.
 
 ## 11. Slice definition of done
 
-- [ ] All acceptance criteria pass with linked evidence.
-- [ ] No unlabeled shortcuts inside the boundary.
-- [ ] Documents updated: statuses, ADD section 15, manual touches if
-  any text describes the old controls.
-- [ ] Owner eyeball complete.
+- [x] All acceptance criteria pass with linked evidence.
+- [x] No unlabeled shortcuts inside the boundary.
+- [x] Documents updated: statuses, DECISION-LOG (D-087, D-088). No
+  manual text described the old controls, so none needed changing.
+- [x] Owner eyeball complete (2026-08-23). See section 9: it caught a
+  real breach of this slice's own guardrail, now fixed as D-087.
 
 ## 12. What this unlocks
 
 - The WPF head's analog pass, from a proven Linux look.
 - The full-sweep restyle decision (combos, spinners) with real
   evidence about whether the bench language helps or hurts data entry.
+  **Taken, as SLICE-016.** The answer was that the question had a
+  false premise: eleven of the fifteen selectors were never lists, so
+  they became banks of interlocked keys rather than restyled combos,
+  and only the four real lists kept a dropdown behind a machined
+  window. Each bank key is this slice's machined cap, so the language
+  carried over without a second control family.
 
 ---
 

docs/SLICE-015-soft-body-keys.md

@@ -1,6 +1,9 @@
 # CUETools Linux Slice Brief: SLICE-015 soft-body rubber keys
 
-Version: 0.2. Date: 2026-08-21. Status: Approved for build.
+Version: 1.0. Date: 2026-08-23. Status: Done, with two measurements
+carried (section 9): press-fidelity renders at each scale factor, and
+the render half of the frame budget. Neither blocks use; both are
+named rather than ticked.
 Companion documents: ARCHITECTURE.md, ENGINEERING.md, DECISION-LOG.md
 (D-080 scope, D-081 the glyph recipe, D-082 the open skill question,
 D-083 why that recipe is right and the upgrade path if it is not
@@ -127,14 +130,36 @@ No persistent data.
 
 ## 9. Verification evidence required
 
-- [ ] Model tests green in the committed harness (no renderer needed).
-- [ ] Pointer/keyboard input tests green.
-- [ ] Rest-identity and press-fidelity renders archived, both themes,
-  at 1.0x / 1.25x / 1.5x / 2.0x.
-- [ ] The LCD-subpixel check on the REAL Avalonia window surface
-  (D-081's named build gate), not on an SKBitmap-backed canvas.
-- [ ] Software-rendering and mid-rip frame-budget measurements.
-- [ ] Owner walkthrough.
+- [x] Model tests green in the committed harness (no renderer needed).
+  31 tests, including the six added on 2026-08-23 for the off-key drag
+  (D-086).
+- [x] Pointer/keyboard input tests green.
+- [x] Rest-identity and press-fidelity renders archived, both themes:
+  `docs/evidence/2026-08-23-seam-light/`. Rest, hover and a held corner
+  press, plus the dead-key legend and a bank position.
+- [~] Renders at 1.0x / 1.25x / 1.5x / 2.0x. Covered for LAYOUT by
+  ScaleMatrixTests, which asserts no clipping and landmark reachability
+  at all five factors in both themes on every push. NOT covered as
+  archived press-fidelity images at each factor.
+- [-] The LCD-subpixel check on the REAL Avalonia window surface.
+  **Retired by D-085.** This gate belonged to D-081's texture recipe,
+  which existed to feed the mesh renderer. The owner's gate verdict
+  kept the projective path, which rasterises no texture and samples
+  nothing, so there is no resampled glyph to check. Nothing was
+  skipped here; the thing being measured stopped existing.
+- [~] Software-rendering and mid-rip frame-budget measurements.
+  The per-frame ARITHMETIC is measured: a homography solve is
+  **12.897 us** and one field sample **0.802 us**, so a single
+  deforming key costs **0.155 percent** of a 120 Hz frame. For scale,
+  the mesh path this replaced measured 2999 us per key per frame, or
+  36 percent of the same budget - about 230x more. What is NOT measured
+  is Avalonia's own render and composite cost for the extra layers, or
+  either number under software rendering or during a live rip.
+- [x] Owner walkthrough. 2026-08-23, on the flag-gated build. Verdict
+  recorded as D-085: the shaded dip reads as below the surface, so the
+  rim stays pinned and the mesh renderer is unfunded. The same
+  walkthrough produced D-086 (the off-key drag defect), D-087 (the
+  unreadable dead legend) and D-088 (light instead of an outline).
 
 ## 10. Agent guardrails for this build
 
@@ -161,13 +186,24 @@ No persistent data.
 
 ## 11. Slice definition of done
 
-- [ ] All acceptance criteria pass with linked evidence.
-- [ ] No unlabeled shortcuts inside the boundary.
-- [ ] Documents updated: statuses, ADD section 15, the manual if any
-  user-visible text describes the old press.
-- [ ] The soft-body skill written, and D-082 resolved so it is
-  actually discoverable from both heads.
-- [ ] Owner walkthrough complete.
+- [x] All acceptance criteria pass with linked evidence, except the two
+  measurements named in section 9 and marked there rather than here.
+- [x] No unlabeled shortcuts inside the boundary. Two are labeled: the
+  rail keys do not deform (light only, see D-088 and section 4), and
+  the frame-budget measurement is arithmetic-only.
+- [x] Documents updated: statuses, DECISION-LOG (D-085 through D-088).
+  No manual text described the old press, so none needed changing.
+- [x] The soft-body skill written, and D-082 resolved so it is actually
+  discoverable from both heads. Verified live: the repo-root symlink
+  surfaces it here with no per-skill wiring.
+- [x] Owner walkthrough complete (2026-08-23, D-085).
+
+**Kill criterion fired, in the good direction.** Section 12's first
+entry says: if the owner sees the shaded dip and cannot tell the
+outline is not bending, ship the cheap transform path and drop the
+mesh. That is what happened. The dip read as below the surface, so the
+cheap path is not a fallback here - it is the answer, and it measures
+about 230x cheaper than the alternative it displaced.
 
 ## 12. Kill criteria
 

docs/SLICE-016-selectors.md

@@ -71,14 +71,27 @@ in `RipView.axaml`, and the Advanced page tooltips.
 | Encoder, mode, enum | Encoder settings | up to 9 codecs, per-codec modes | window |
 | Action | Queue | short fixed set | window |
 
-CTDB sharing is the interesting one. It has only three options, which
-says bank, but its labels run about 330 pixels laid out as one, which
-says window. Long labels win. Shortening them to Ask, Always, Never
-would make it a clean bank, but that is a wording change to a consent
-control and it is the owner's call.
-
-Queue action is the other judgement call. It sits inside a table row,
-where a bank fights the column sizing.
+CTDB sharing is the interesting one, and it stays a window. Measured
+2026-08-23: the Settings row is 455 DIP wide and its control gets 296.
+A three-key bank with these labels needs about 352 for the keys, and
+the row label "Share verified rips with CTDB" takes another 190, so it
+does not fit beside its label at any padding worth having. It WOULD
+fit with the label stacked above, but that is the only row on the page
+that would break the label-left rhythm.
+The alternative is shortening the options to Ask, Always, Never, which
+reads fine under that row label and would make a clean bank. Not
+taken. It is a wording change to a consent control, `settings.md` line
+34 quotes the current strings, and the long labels are the ones that
+still say what they mean when a screen reader reads the option without
+the row label. If the owner wants the bank, the change is those three
+strings, the manual line, and an accessible name on the bank; it is
+about five minutes and it is reversible.
+
+Queue action BECAME a bank. The earlier reasoning was wrong on a fact:
+it is not in a table row. It sits in a `LastChildFill="False"` toolbar
+where a wider control just pushes the next one along, and its options
+are `Verify` and `Convert` - two short labels, which is the rocker
+case. Both now show without a click.
 
 ## 4. Width, and where v0.1 got it wrong
 
@@ -145,9 +158,9 @@ this landed as one pass.
   role are untouched. `Classes="window counted"` adds the position
   readout, which stays opt-in per the v0.1 ruling.
 
-Sixteen call sites converted: six to banks (quality, layout on Rip and
-on Settings, metadata search, album art search, network) and ten to
-windows.
+Sixteen call sites converted: SEVEN to banks (quality, layout on Rip
+and on Settings, metadata search, album art search, network, and the
+Queue action added 2026-08-23) and nine to windows.
 
 ## 8. Verification
 

src/CUETools.Linux.App/Views/QueueView.axaml

@@ -15,9 +15,9 @@
         <TextBlock DockPanel.Dock="Left" Text="Action"
                    Foreground="{DynamicResource Muted}" FontSize="11"
                    VerticalAlignment="Center" Margin="0,0,8,0" />
-        <ComboBox Classes="window" DockPanel.Dock="Left" Width="100" VerticalAlignment="Center"
-                  ItemsSource="{Binding Actions}"
-                  SelectedItem="{Binding SelectedAction, Mode=TwoWay}" />
+        <ListBox Classes="bank" DockPanel.Dock="Left" VerticalAlignment="Center"
+                 ItemsSource="{Binding Actions}"
+                 SelectedItem="{Binding SelectedAction, Mode=TwoWay}" />
         <Button DockPanel.Dock="Left" Width="220" MinWidth="140"
                 VerticalAlignment="Center" Margin="8,0,0,0"
                 Click="OnCodecPickerClick"

tests/CUETools.Linux.Tests/SelectorLayoutTests.cs

@@ -53,16 +53,28 @@ public void NoSelectorBankEverHasAPositionCutOff()
         var window = new MainWindow(new ThemeState(), graph.Verify, graph.Convert, graph);
         window.Show();
 
+        // every page that carries a bank, not just Rip. The clipping this test
+        // exists to catch is invisible at page level, so a page left unvisited
+        // is a page left unchecked.
+        var pages = new (string Name, Action Show)[]
+        {
+            ("rip", window.ShowRipPage),
+            ("settings", window.ShowSettingsPage),
+            ("queue", window.ShowQueuePage),
+            ("convert", window.ShowConvertPage),
+        };
+
         var checkedAny = false;
         foreach (ThemeVariant variant in new[] { ThemeVariant.Dark, ThemeVariant.Light })
         {
             Avalonia.Application.Current!.RequestedThemeVariant = variant;
             foreach (var (w, h, scale, name) in Widths)
+            foreach (var (pageName, show) in pages)
             {
                 window.SetRenderScaling(scale);
                 window.Width = w;
                 window.Height = h;
-                window.ShowRipPage();
+                show();
                 Avalonia.Threading.Dispatcher.UIThread.RunJobs();
 
                 foreach (ListBox bank in window.GetVisualDescendants()
@@ -80,10 +92,10 @@ public void NoSelectorBankEverHasAPositionCutOff()
                         double right = topLeft!.Value.X + item.Bounds.Width;
 
                         Assert.True(right <= bank.Bounds.Width + 0.5,
-                            $"{name} {variant}: bank position \"{item.Content}\" ends at {right:0.0} " +
+                            $"{pageName} {name} {variant}: bank position \"{item.Content}\" ends at {right:0.0} " +
                             $"inside a bank only {bank.Bounds.Width:0.0} wide, so it is cut off");
                         Assert.True(item.Bounds.Width > 1,
-                            $"{name} {variant}: bank position \"{item.Content}\" collapsed to " +
+                            $"{pageName} {name} {variant}: bank position \"{item.Content}\" collapsed to " +
                             $"{item.Bounds.Width:0.0} wide");
                     }
                 }