Logs tab, Record page rework, and a Spotify match that works

[revtex]

Five things from one batch.

1. Spotify metadata matched nothing SMTC reported

The attached log showed four failed attempts then "Spotify had no metadata for 'ATB - 9Pm (Till I Come)'." — for a track Spotify obviously knows.

The guard compared Spotify's bare track name against the detected title after running only Spotify's side through the window-title splitter. So it compared a parsed string against an unparsed one for every track the media session found, which is the common case since Phase 7 made SMTC primary — SMTC reports titles verbatim, where the window-title parser has already split them.

DetectedTrackMatch reduces both sides to a common form and compares them in both shapes: the title alone, and the artist joined on. A provider's bare name matches a detected Artist - Title because joining the provider's own artist onto its name produces the same thing.

Normalisation deliberately stops well short of fuzzy. The wrong answer at a track boundary is not "no metadata", it is a file tagged as a different song — so it forgives case, accents, spacing, and the punctuation each source wraps a qualifier in, and nothing else. Three guard tests pin the line: "Song" must not match "Song (Live)", "Song (Radio Edit)" must not match "Song (Extended Mix)", and "9Pm (Till I Come)" must not match "9Am (Till I Come)".

19 new tests.

2. The saved count showed zero

[ObservableProperty] on _savedCount refreshed HasSaved and not SavedCountText, which is derived from the same field. The count was correct in the debugger the whole time and stale on screen, and nothing failed. One attribute.

Worth watching for on the rest of this page — a missing [NotifyPropertyChangedFor] has no compile-time or runtime signal at all.

3. Logs is its own tab, beside Advanced

Third home in three attempts: the whole page, then a fixed-height box, then a collapsed expander. A log wants either the whole surface or none of it, and a shared page can offer neither.

LogsPage is a second view onto RecordViewModel rather than a ViewModel of its own — the lines, the filter and both buttons are already there and subscribed to the sink since startup, so a second ViewModel would mean a second subscription and two histories diverging by construction order. Auto-scroll (follow the tail only while already at the tail) moves with it.

4. Display and Currently recording

• Backlit, not reflective. The panel was pale grey with near-black segments — a calculator face, and the palest object in a nearly black window. Inverted it goes a shade under the window with near-white segments, which is the transflective backlight a field recorder actually has, and contrast goes up. Two knock-ons that are easy to miss: the scanline brush flips black→white (a black line over a near-black ground is invisible) and the glass gradient flips too (top-dark on a dark panel reads as a shadow, not a diffuser).
• The meter is untouched. Its own drawing, behaviour, and the whole decibel-placed spectrum are exactly as they were. Only the ground it sits on moved.
• The track name came off the face. It was the one element that is neither a transport state nor a measurement — a proper noun of arbitrary length, ellipsised, where an instrument prints a legend. It now lives in the Currently recording card with the art and album it belongs to.
• The counter gets the right-hand side and an ELAPSED legend. A bare 00:03:42 on a recorder could be elapsed, remaining, or a timestamp; one small word settles it once instead of every glance. Format text moves beside the transport block, which is where it belongs — "recording" and "recording as what" are one question.
• Currently recording is art on the left, three labelled rows on the right: Artist/Song, Album/Year, Save path. Labels share a column so the values start on a common edge — which matters because two of the three rows arrive a second later, when enrichment lands, and a block that re-flows as it fills is one nobody trusts. The two late rows hide rather than blank, since an empty value beside a label reads as "there is no album" when it means "no lookup has returned yet".
• Session totals moved onto the heading they count, right-aligned as Count: n | Time: hh:mm:ss with a rule filling the gap. They were stacked inside a card describing a single track — two timescales in one block — and the duration was prose (4m 12s) sitting inches from a fixed-width clock, so comparing them meant converting between them.

5. Removals

• Readiness chips and ReadinessProbe, plus fourteen resource strings and the DI registration. A missing ffmpeg is already an InfoBar with a sentence in it, and a library folder is already unsaveable without one. Five green pills reporting nothing is wrong is a row that is always there and never read.
• The FlaUI Desktop suite, deleted rather than skipped — the per-control page suites went on 2026-08-12 at your direction, and the thirteen shell/Record tests that survived were pinned to exactly the layout this PR rewrites. FlaUI.UIA3 and CI's --filter "Category!=Desktop" go with them. Automation ids stay on every control — they are a Phase 9 accessibility requirement in their own right and are what any future suite would need.

Verification

dotnet build clean, dotnet format --verify-no-changes clean, 995 tests green (824 Core + 171 UI). No UI testing run, per your standing instruction — worth a look at the Record and Logs tabs, and at whether the inverted display reads right on your monitor.

🤖 Generated with Claude Code