docs: record the first Phase 5 pass and what it corrected

Two reproductions in this pass were read before Lit had rendered, so
both reported the same answer on the broken build and the fixed one -
the third costume of this plan's most-repeated trap, and the first time
it has appeared in a reproduction rather than a measurement. Also
records that the audit's symptom for H-11 outlives its mechanism, that
fixing H-7's arithmetic does not remove every clipped Duration, and
that two e2e specs spend backend state they never give back.

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

@@ -18,7 +18,7 @@ here has disappeared.
 
 ## Read this part before you fail
 
-Seven things cost a cycle each the first time. They are here, not in a
+Nine things cost a cycle each the first time. They are here, not in a
 reference, because you need them *before* the failure, not after.
 
 - **Time out every binding call.** A bound Go method called with wrong
@@ -46,6 +46,16 @@ reference, because you need them *before* the failure, not after.
   Run against the `bulk` seed a measurement session left behind and 13
   of 36 fail, in a list that reads exactly like a regression in
   whatever you are holding. `make dev-headless SEED=default` first.
+- **…and the suite spends state it cannot always give back.**
+  `view-lifecycle.spec.ts` **skips an autotag album** on every run, out
+  of the eleven the seed has, and does not put it back — so around the
+  eleventh consecutive run against one app it starts failing on an
+  empty queue. Restart between runs (`make dev-stop && make
+  dev-headless SEED=default`) when a spec starts failing that you have
+  not touched, and *before* believing a failure at all. Backend state
+  outlives the page: shuffle used to be left on the same way, which
+  failed `playback.spec` on the next run — that one is fixed, the
+  autotag one is inherent.
 - **A frontend edit is not live until you restart the app.** Vite
   updates the module, but an already-registered custom element class
   cannot be re-registered, so a running page keeps the old one and your

.planning/NOTES.md

@@ -1259,3 +1259,73 @@ And three on measuring, all of which produced a wrong number first:
   sidebar click: one highlighted card, `aria-selected="true"` on the
   right one, in both card grids. Check `viewActive` before believing a
   view did not repaint.
+
+## A reproduction read too early is a fix applied to nothing
+
+Plan 007 phase 5, first pass: the track list's arithmetic, a window
+minimum the layout can actually hold, and one page header for nine
+views. Three items, one inherited one-liner, and the pass's own
+contribution to this plan's longest-running theme — *a number that
+cannot move is not evidence* — which appeared twice more here, both
+times in a **reproduction** rather than in a measurement.
+
+`config-page`'s rename bug is real: the library name's click bubbles to
+the document handler that closes the rename editor, so it opens and
+closes it in the same click. But the probe that "reproduced" it read
+`.edit-input` synchronously after a synthetic `.click()`, and Lit
+renders on a microtask — so it reported "not editing" on the broken
+build *and* on the fixed one. The fix looked like it had done nothing,
+which nearly bought a second, unnecessary fix; the real check needed one
+`await`. The same shape then failed an e2e spec of mine, which captured
+a header count immediately after a navigation and got `null`, making
+every assertion after it vacuous.
+
+Seven things worth keeping:
+
+- **A screenshot disagreeing with a number is the useful signal, not a
+  puzzle to explain away.** A viewport ladder reported the top bar
+  overflowing by 0 px at every width while the PNG showed the job
+  indicator cut off at the edge. The badge was `display: none` — no job
+  was running by then — so nothing was overflowing and the "bug" was a
+  rendering of the app working. Two numbers that must agree are worth
+  more than one number you have to be sceptical about, and a picture
+  counts as one of the two.
+- **An audit's symptom can outlive its mechanism.** `H-11` says the app
+  title "wraps into the nav" below 700 px. It does — but the title
+  block is 80 px tall inside a 64 px bar at *every* width, including
+  1440; what changes at 780 px is the *subtitle* taking a second line.
+  Fixing the visible half is a breakpoint on the subtitle. The
+  permanent 16 px was never the finding and is still there.
+- **Fixing the stated cause does not always remove the stated
+  symptom.** With the 40 px arithmetic fixed, Duration still reads
+  "Durat…" at 800 px — because the column is at its 50 px floor and the
+  *label* no longer fits, while the values do. Same screenshot,
+  different mechanism. Worth writing down, or the next reader
+  reasonably concludes the arithmetic fix did not land.
+- **Count the copies before calling something a duplicate pair.** The
+  audit names Albums and Tracks as the two views with a sort toolbar;
+  `playlist-view` had a third copy of the same twenty lines. The fix
+  was worth 1.5× what the finding implied.
+- **What a model carries decides what a control can offer.**
+  "Artists and Genres have no sort control" is one finding and two
+  different fixes: genres have a track count to sort by, and
+  `library.Artist` carries nothing countable at all. A select with one
+  option is a control that does nothing, so that view gets a label and
+  a direction button.
+- **Backend state outlives the page, and a spec that spends it fails
+  the *next* run.** `view-lifecycle.spec.ts` toggled shuffle and never
+  toggled it back, so a second `make e2e` against the same app failed
+  `playback.spec`'s shuffle assertion — in a list that reads exactly
+  like a regression in the change you are holding, which is what I
+  assumed for half an hour. Stashing the phase's source changes and
+  re-running the same specs is what proved it pre-existing; the same
+  file also skips an autotag album per run, out of eleven, which is
+  inherent and now in the skill. **Restart the app before believing an
+  e2e failure you did not cause.**
+- **A header that appears with the data is the layout problem it was
+  meant to fix.** The first version of `<page-header>` rendered only
+  once a view had loaded, and showed "0 playlists" while loading. Both
+  were caught by reading a screenshot rather than by any assertion. The
+  header now renders during load and omits the count until there is an
+  answer — `null` meaning "no answer yet", which is a different thing
+  from zero and has to be a different value.

.planning/plans/active/007-ui-reconciliation.md

@@ -1557,6 +1557,134 @@ views; an e2e spec asserting no horizontal overflow on any row at
 1440×900, 1024×768 and the new minimum; and a manual pass with a
 screen reader on the four surfaces that gained live regions.
 
+### Phase 5 — the first pass: the arithmetic, the minimum, and the header
+
+Items 1 and 2 of the proposed four, plus one of the seven inherited
+items. Each landed with its reproduction watched failing first, in the
+running app rather than in a test.
+
+- **`H-7` — the track list fits its container.** Reproduced exactly as
+  the audit says: `scrollWidth 1280` against `clientWidth 1240` on the
+  header row and all 31 track rows, and the 40 px is precisely
+  `24 + 2×8`. `computeDefaultWidths`, `normalizeWidths` and
+  `onColResizeMove` now share one `availableColumnWidth`, and the two
+  constants behind it are read by `colBoundaryPositions` and the grid
+  template too — they were written out separately in four places,
+  which is how they came to disagree. **0 of 31 rows overflow after,
+  at every viewport tested.**
+- **`H-11` — the minimum is a size the layout supports.** Reproduced:
+  at 700×480 the sidebar's eleven items need 406 px of a 352 px pane,
+  and Settings rendered at y=420–454 against a pane ending at y=416 —
+  outside its own box, clipped, unreachable. The pane scrolls now, the
+  sidebar collapses to icons below 900 px (its `.collapsed` mode
+  existed and only a manual drag had ever reached it), the subtitle
+  hides at the same breakpoint, and `MinWidth`/`MinHeight` are
+  **800×600**, chosen by walking a ladder of nine viewports and
+  reading what broke where rather than by picking a round number.
+- **`H-19` and `H-10` — one page header, nine views.**
+  `<page-header>` is title, count, sort and actions; Artists and
+  Genres gain the sort they never had, four views gain a heading, and
+  five gain a count. The header search box keeps its slot everywhere,
+  names its scope in the placeholder and in the header's own line, and
+  is disabled with a reason where it cannot serve.
+- **The Settings rename one-liner** (found in Phase 3, routed here):
+  the library name's click bubbled to `config-page`'s document
+  handler, which exists to close the rename editor, so it opened and
+  closed it in the same click.
+
+#### Where the plan was wrong — the first pass
+
+Seven things, and two of them are about how the finding was checked
+rather than about the finding:
+
+- **The two reproductions I wrote first were both invalid, in the same
+  way, and one of them nearly shipped a fix for nothing.** Reading the
+  DOM synchronously after a synthetic `.click()` reports the state
+  *before* Lit renders — so the Settings rename probe returned "not
+  editing" both before and after the fix. The bug is real (verified
+  properly: `false` before, `true` after, with an await), but for
+  twenty minutes the evidence for it was a number that could not move.
+  Same trap as this plan's `0 ms` view-open and its 150 ms debounce,
+  in a third costume, and now also in an e2e spec that read a count
+  before the view had one.
+- **`job-indicator` is `display: none` when no job is running**, which
+  looks exactly like a control squeezed out of an overflowing header.
+  A ladder of viewport measurements said the top bar overflowed by
+  0 px at every size while a screenshot plainly showed the badge cut
+  off at the right edge; the badge was simply *absent* by then,
+  because the index build had finished. The contradiction between the
+  number and the picture was the useful signal, and chasing it saved
+  fixing a layout that was not broken.
+- **`H-11`'s "the app title wraps into the nav" is a subtitle
+  problem, and it is not new at 700 px.** The title block is 80 px
+  tall inside a 64 px bar at *every* width — it merely stops being
+  visible about there, when the subtitle takes a second line and it
+  becomes 98 px. Hiding the subtitle under the breakpoint fixes the
+  visible half; the 16 px of permanent overflow is cosmetic and
+  untouched.
+- **`H-7` is not the only reason Duration looks clipped.** With the
+  arithmetic fixed, at 800 px the column is at its 50 px floor and the
+  *label* still ellipsises to "Durat…", because the saved widths are
+  scaled proportionally from whatever size they were set at. The
+  values fit; the heading does not. Different mechanism, same
+  screenshot, and worth knowing before someone "fixes" the arithmetic
+  again.
+- **The sort toolbar existed three times, not twice.** The audit names
+  Albums and Tracks as the two views with sort controls; `playlist-view`
+  has a third copy of the same twenty lines. All three are now the
+  header's.
+- **Artists cannot have a sort *select*.** `library.Artist` carries a
+  name, an MBID and three image URLs — nothing countable — so "the two
+  missing sort controls" is really one control and one direction
+  toggle. The header renders a label instead of a select with a single
+  option in it.
+- **Two e2e specs were spending state they never gave back**, which is
+  not in any audit and cost most of an hour to attribute.
+  `view-lifecycle.spec.ts` toggled shuffle and left it on, so the
+  *second* `make e2e` against the same app failed `playback.spec`'s
+  shuffle assertion — a failure that reads exactly like a regression
+  in whatever you are holding, and which I first assumed was mine.
+  Stashing the phase's source changes and re-running proved it
+  pre-existing. Shuffle is restored now; the same file also skips an
+  autotag album per run out of the eleven the seed has, which is
+  inherent and is now in the skill instead.
+
+#### Not done, and still worth doing (after the first pass)
+
+Items 3 and 4 of the four, in that order: the dialogs, the context
+menu's keyboard model and the ARIA tail as **one** pass (they are one
+focus/semantics story, and splitting them is how two focus traps get
+built); then the smaller items — landing on Home, the Home card's
+missing-art placeholder, the album page's primary action and its
+unexplained ✓ badges, the `?` overlay, Settings reordered with a
+Playback section, and an Album column in the track list.
+
+Five of the seven inherited items remain: `cover-grid`'s dead album
+dropdown (`perf.p2` — still a missing feature, not housekeeping), the
+context menu's keyboard model, `tracklist.delete` (which needs a
+"remove from library" that does not exist, and a decision about what
+it removes), keyboard seeking from a focused track row, and the
+header search box on `smart-playlist-details` — which is a *detail*
+view and so was outside this pass's nine primary ones.
+
+One finding of my own, not fixed: **the Home shelves' cards render a
+missing cover as nothing at all** (`H-9`), which is plainly visible in
+any Home screenshot now that the page has a header above it.
+
+And two things about CI, neither mine and both pre-existing on
+`main` at `9e92721`:
+
+- **`player-truth.spec.ts` fails in the CI container**, on the elapsed
+  clock (17 s and 11 s adrift, against a tolerance of 1) and
+  intermittently on the auto-advance skip. It passed 18 h earlier and
+  failed on a **docs-only** commit, so it is the container's audio
+  clock rather than a regression. All 44 specs pass locally, twice in
+  a row against one app.
+- **`CLAUDE.md`'s claim that commitlint enforces the commit format in
+  CI is stale** — there is no config and no workflow running it — as
+  is the implication that semantic-release runs. Left alone pending a
+  decision: wire them up, or stop saying it.
+
 ---
 
 ## Phase 6 — Explore starts the conversation

CLAUDE.md

@@ -405,6 +405,54 @@ result would silently reorder a queue — and because `cover-grid`'s drag
 cache stores them per album. A `libraryID` of 0 means "every library",
 matching an unset library filter.
 
+**A view says what it is, in one component.** `<page-header>`
+(`components/page-header/`) is title, count, sort and actions, and the
+nine primary views use it rather than each writing its own — which is
+how four came to have a heading and four not, and how the same sort
+toolbar came to exist three times with different bugs. Three rules in
+it are load-bearing. **An empty `heading` is a mode, not a missing
+value**: `cover-grid` and `track-list` are also embedded in the artist
+and genre pages, which have a heading already, so there they keep the
+count and the sort and drop the title rather than growing a second
+arrangement. **The header renders while the view loads** — a heading
+that arrives with the data is the shifting layout it exists to stop —
+and the *count* is `null` until there is an answer, because "0 albums"
+that corrects itself a moment later is worse than saying nothing. And
+**the header asks for a sort, it does not perform one**: the host owns
+the field, the direction and their persistence, so the control cannot
+disagree with the list.
+
+**The header search box is view-scoped, and now says so.** It sits in
+the app header and reads as global; typing `tide` on Playlists answered
+"No playlists match your search" with three *Tideline* tracks in the
+library. The scope was always real — the fix is that the placeholder
+names it ("Search albums"), the page header repeats it ("Showing
+artists matching ‘tide’"), and `search-store` holds the one map of
+what each view searches. It also **keeps its slot everywhere and is
+disabled** where it cannot serve, rather than being hidden: its
+appearing and disappearing is what moved the library filter and the job
+indicator on every navigation. On Explore, which has its own catalog
+search, the disabled box points at it.
+
+**The window's minimum is measured, not aspirational.** `MinWidth`/
+`MinHeight` are 800×600 because that is where the shell was checked to
+still work: below ~780 the header subtitle wraps and pushes the title
+out of the 4em bar, and below ~600 tall the eleven sidebar items stop
+fitting at once. The old 512×384 promised a size at which Settings and
+Jobs were unreachable — the sidebar clipped them with `overflow:
+hidden` and nothing scrolled. The sidebar scrolls now, and collapses to
+its (long-existing, previously drag-only) icon mode below 900px, which
+is the same breakpoint that hides the subtitle.
+
+**A row's columns must fit the row.** `track-list`'s
+`computeDefaultWidths` shared out the host's whole `clientWidth` while
+every row spends 24px on the favourite column and 2×8px on its own
+padding first, so the grid was always exactly 40px too wide and the
+last column was clipped at every size. Both numbers are constants read
+by the three places that need them (the default widths, the
+normaliser, and the resize handles' positions), because they were
+written out separately and that is how they came to disagree.
+
 **Event-driven communication**: Backend emits events via Wails runtime; frontend stores subscribe to them. Event names are constants in `backend/events/`.
 
 `frontend/src/events.ts` is **generated** from `backend/events/events.go`