# Phase 8 — CMS Track Browser

Status: spec / one VM, three views. Review decisions folded in 2026-06-11 (waveform-in-grid,
DI-scoped VM, BatchEdit extraction confirmed). **A data-model normalization (§0) is now a hard
pre-requisite — it must land before any Phase 8 UI work.** Author: product-designer.
Date: 2026-06-11.
**Plan only — no code edits made by this doc.**

Cross-references: `PLAN.md §8` (the concise phase entry, with §8.0 the normalization gate),
`PLAN.md §6.2` (the deferred "card-contextual filtering" item this phase supersedes), memory
*One source, multiple views*.

---

## 0. Pre-requisite: Phase 8.0 — `TrackEntity` Normalization

**This must land before any of §8.1–§8.5 UI work.** It is a breaking data-model change, not a UI
change, and every UI section below assumes the normalized schema. Daniel required this gate during
review on 2026-06-11.

### 0.1 What

Split the current flat `TrackEntity` into two normalized tables.

**`ReleaseEntity` (new)** — release-cardinal data (the "header" shared across all tracks in a
release):

| Field | Type | Notes |
|---|---|---|
| `Id` | `long` | PK |
| `Title` | `string` | the album/release name (currently `TrackEntity.Album`) |
| `Artist` | `string` | |
| `Genre?` | `string` | |
| `ReleaseDate?` | `DateOnly` | |
| `ImagePath?` | `string` | cover art |
| `ReleaseType` | `ReleaseType` enum | |
| `CreatedByUserId?` | `long` | |

**`TrackEntity` (updated)** — track-cardinal data only:

| Field | Type | Notes |
|---|---|---|
| `Id` | `long` | PK |
| `ReleaseId` | `long` (FK → `ReleaseEntity.Id`) | nullable — see open question (1) |
| `Release` | navigation property | |
| `EntryKey` | `string` | FileDatabase link (immutable) |
| `TrackName` | `string` | |
| `TrackNumber` | `int` | 1-based |
| `OriginalFileName?` | `string` | |

*Removed from `TrackEntity`:* `Artist`, `Album`, `Genre`, `ReleaseDate`, `ImagePath`,
`ReleaseType`, `CreatedByUserId` — these now live on the Release.

### 0.2 Why

The current flat schema duplicates release-level metadata (artist, album, genre, release date,
cover art) on every track row. This was fine at launch but creates consistency hazards: updating an
album's cover art or artist name means rewriting every track row. Phase 8 introduces album-mode
editing (Batch Edit, album-scoped delete), so the data model should match the domain — a Release is
a first-class entity; Tracks belong to a Release.

The normalization also collapses several open questions in the UI spec below:

- `AlbumSummaryDto` becomes a simplified `ReleaseDto` (query the Release table, not `GROUP BY`).
- `CmsAlbumBrowser` parent rows map directly to `ReleaseEntity` rows — no derivation needed.
- The `AlbumSummaryDto` widening question (old §6 option ii / §12.3) **dissolves entirely** — the
  Release table just has all the fields.
- Batch Edit's "header" = the Release record; "track list" = the Track records.

### 0.3 Shape (direction, not implementation prescription)

**Migration path (breaking):**

1. Create `releases` table (from `ReleaseEntity`).
2. Populate `releases` from the distinct albums in `tracks` — one release row per distinct
   `(album, artist)` identity (see open question 5). Tracks with a null album get a nullable FK or a
   synthetic "Uncategorized" release (open question 1).
3. Add `release_id` FK column to `tracks`; populate from the newly created release rows.
4. Drop the now-redundant columns from `tracks` (`artist`, `album`, `genre`, `release_date`,
   `image_path`, `release_type`, `created_by_user_id`). **This step is the breaking part.**

**DTO impact:**

- New `ReleaseDto` mirroring `ReleaseEntity`.
- `TrackDto` slims down: remove `Artist`, `Album`, `Genre`, `ReleaseDate`, `ImagePath`,
  `ReleaseType`, `CreatedByUserId`. Add `ReleaseId` (long) and, for read paths that display release
  data alongside track data, either a nested `Release` (`ReleaseDto`) or a flat denormalized read
  model (open question 3).
- `AlbumSummaryDto` is retired / collapsed into `ReleaseDto` (open question 4).

**Consumer-impact list (the most consumer-visible part of this change):**

The existing `GET api/track/page` response shape changes. Callers that read `TrackDto.Artist` /
`.Album` / `.Genre` / `.ReleaseDate` / `.ImagePath` must adjust to read those from the Release
(`TrackDto.Release.Artist` / `.Release.Title` / …) or from a flat joined read model. Known
consumers:

- **`DeepDrftPublic.Client`** — the public track gallery (`TrackCard` and the views feeding it) reads
  `Artist`, `Album`, `Genre`, `ReleaseDate`, `ImagePath` for display. **This is the highest-risk
  consumer** — a public-facing surface, a different host, not part of the CMS phase otherwise.
- **`DeepDrftManager`** — the CMS (this phase's surface).

**Service / repository impact (list, not prescription):**

- `TrackRepository` — queries gain joins / `Include`s to load the Release.
- `TrackManager.GetPaged` — sorting by release fields (artist, album, genre, date) needs a join.
- `UnifiedTrackService.UploadAsync` — must find-or-create the Release before inserting the Track.
- `TrackContentService.AddTrackAsync` / the BatchUpload flow — accepts a Release (or release id)
  rather than embedding album/artist/etc. on the track upload form.
- The upload API endpoint form fields split: separate "release" fields from "track" fields, or the
  upload auto-creates the Release on first track and links subsequent tracks (open question 2).

### 0.4 Open questions (need Daniel before this lands)

1. **Tracks with no album.** Nullable `ReleaseId` (a "standalone track" concept), or does every
   track require a Release (a release with null/empty title)? **Recommend nullable FK** — some tracks
   genuinely have no album context.
2. **Upload flow change.** Does uploading now require the caller to first create/identify a Release,
   or does the endpoint auto-create a Release if none exists for the given album name?
   **Recommend auto-create-or-find** (upsert on `(title, artist)`) — keeps upload-form ergonomics
   close to today's.
3. **`TrackDto` shape for the public API.** The public gallery reads flat `Artist` / `Album` /
   `Genre` / `ReleaseDate` / `ImagePath`. Keep these queryable post-normalization via either a flat
   joined read model (no nested object) or a nested `Release` property. **Recommend the flat
   read-model for the public API** — no response-shape break if the flat fields stay populated from
   the join. Call this out as the most consumer-visible impact.
4. **`AlbumSummaryDto` fate.** With a Release table, `GetAlbumSummariesAsync` becomes
   `GetReleasesAsync` returning `List<ReleaseDto>`. **Recommend retiring `AlbumSummaryDto`** in favor
   of `ReleaseDto` — a direct replacement.
5. **Migration release identity.** Existing tracks are flat. The migration derives Release rows from
   them. If two tracks share an album name but different artists, they'd map to two releases (or one
   ambiguous one). **Recommend grouping by `(album, artist)`** as the release identity key — flag
   this edge case explicitly in the migration.

### 0.5 Impact on the Phase 8 UI sections below

With normalization landed:

- `CmsAlbumBrowser` parent rows **are** Release rows — the Release table has all the fields, so the
  old §6 "derive artist/genre/date/type" problem and the §12(3) "widen the DTO" question both
  disappear.
- Batch Edit's "header" is a `ReleaseDto` form; its "track list" is `List<TrackDto>` (slim).
- The `GetPagedAsync` album/genre filter still works — it joins through the `releases` table.

The UI sections (§1–§13) below are written against the pre-normalization vocabulary in places;
where they mention `AlbumSummaryDto` widening or deriving release fields from children, **read that
as resolved by §0** — the Release table supplies those fields directly. Net of §0, those UI
sections get *simpler*, not harder.

---

## 1. Summary

The CMS `/tracks` page today is a single `MudTable<TrackDto>` (the "Tracks" tab) plus a
"Waveform Pre-Processing" tab, both inside one `MudTabs`. This phase keeps that page and adds
**three browse modes** for the catalogue — **Track**, **Album**, **Genre** — selected by a
toggle, each addressable by a distinct URL so the public home page can deep-link into a given
mode.

**Pre-requisite gate:** all UI work in this phase sits on top of **§0 — the `TrackEntity`
normalization** (split into `ReleaseEntity` + slimmed `TrackEntity`). §0 is a breaking data-model
change that must land first; it also dissolves several of the open questions this spec originally
carried (notably the `AlbumSummaryDto` widening question — the Release table supplies those fields
directly). Read the UI sections below as written against the post-normalization schema.

The Waveform Pre-Processing tab is **removed** (§9): waveform-profile status becomes an in-grid
status column on `CmsTrackGrid`, with per-row "Generate" actions and a page-level "Generate All
Missing" button replacing the old tab.

The architectural spine is **one view-model feeding three renderings**. This matches Daniel's
standing preference (memory: *One source, multiple views* — "same data, different uses and
ergonomics"): the divergence lives in layout, not in data paths. All three modes read from the
**same `ICmsTrackService.GetPagedAsync` path**; Album and Genre modes add summary calls
(`GetAlbumSummariesAsync` / `GetGenreSummariesAsync`) for their *parent* rows but reuse the same
paged-tracks call for their *detail* rows. No new API endpoint is introduced.

This phase also **supersedes the deferred §6.2** ("card-contextual filtering"). §6.2 was deferred
precisely because filtering needed an intermediate browse surface (a list of albums, a list of
genres) for the admin to pick from — "a CMS analogue of the public `AlbumsView`/`GenresView`."
Album Mode and Genre Mode *are* that surface. §6.2 should be marked superseded-by-§8, not left
dangling.

It also introduces one new page — **Batch Edit** (`/tracks/album/{albumName}/edit`) — reached
from an Album-Mode row's Edit action. This is `BatchUpload.razor`'s master-detail mechanics with
existing data preloaded and the submit path swapped from upload to metadata update.

### What this is NOT

- Not a public-site feature. This is CMS-only (`DeepDrftManager`, InteractiveServer). The public
  home page only *links in*; it renders nothing from this phase. **Caveat:** §0's normalization
  changes the shared `TrackDto` contract, which *does* touch the public client — see §0.3 consumer
  impact. The browse-mode UI is CMS-only; the data-model change is not.
- **Is** a data-model change — §0 introduces `ReleaseEntity` and slims `TrackEntity`. (The browse
  *modes* introduce no new model; the pre-requisite does.)
- Not a replacement for the existing single-track edit page (`/tracks/{id}`) — that survives,
  reached from Track Mode's per-row Edit. Batch Edit is a *second*, album-scoped editor.
- The Waveform Pre-Processing tab **is removed** — its function moves into the grid (§9).

---

## 2. Grounding — what exists today (verified, not paraphrased)

Read against the live source on 2026-06-11. Two facts diverge from the brief and are
load-bearing:

1. **`ICmsTrackService.GetPagedAsync` does NOT currently take album/genre filter parameters.**
   Its signature is `GetPagedAsync(int page, int pageSize, string? sortColumn, bool sortDescending, CancellationToken ct)`.
   The *underlying* `GET api/track/page` endpoint accepts `album=` / `genre=` query filters, and
   `CmsTrackService` (the impl) could pass them — but the interface does not expose them today.
   **This phase requires extending that signature** (see §6, data contracts). This is the one real
   contract change. Flag it as the first implementation prerequisite.

2. **An existing single-track edit page already lives at `/tracks/{id}`.** Track Mode's per-row
   Edit button (`Href="/tracks/{context.Id}"`) goes there. Album Mode's batch Edit is a *separate*
   page. Do not conflate them; do not route the album edit through `/tracks/{id}`.

Other verified facts:

- `/tracks` (`TrackList.razor`) is `@attribute [Authorize]`, InteractiveServer, two
  `MudTabPanel`s inside one `MudTabs`: "Tracks" (the `MudTable<TrackDto>` with `ServerData=LoadServerData`,
  `RowsPerPage=20`) and "Waveform Pre-Processing".
- Current Track table columns: Track Name, Artist, Album, Genre, Release Date (`yyyy-MM-dd`),
  Entry Key (monospace caption), File Name (monospace caption), Actions (Edit→`/tracks/{id}`,
  Delete→confirm dialog → `DeleteTrackAsync`).
- `BatchUpload.razor` (`/tracks/upload`): album-header fields (Album Name, Artist, Genre,
  Release Date as `YYYY-MM-DD` string, Release Type `MudSelect`, Cover Art `InputFile`), then a
  master list of `BatchTrackRow` (WavFile, TrackName, Status) with move-up/down/remove + a
  detail pane editing the selected row. Submit: optional one-shot cover-art upload
  (`UploadImageAsync`) → per-row `UploadTrackAsync` → per-row `UpdateAsync` to link the image
  (the upload endpoint takes no imagePath). `TrackNumber` is the 1-based list position (`i + 1`).
- `ICmsTrackService` already has: `GetPagedAsync`, `GetByIdAsync`, `UploadTrackAsync`,
  `UpdateAsync(id, trackName, artist, album, genre, releaseDate, imagePath?, releaseType?, trackNumber?)`,
  `DeleteTrackAsync(id)`, `UploadImageAsync`, `GetAlbumSummariesAsync`, `GetGenreSummariesAsync`,
  `GetTrackCountAsync`, plus the waveform pair.
- `TrackDto` fields: `Id, EntryKey, TrackName, Artist, Album?, Genre?, ReleaseDate? (DateOnly),
  ImagePath?, OriginalFileName?, ReleaseType, TrackNumber (int, 1-based)`.
- `AlbumSummaryDto`: `Album (string), TrackCount (int), CoverImageKey (string?)`.
- `GenreSummaryDto`: `Genre (string), TrackCount (int)`.
- Image URL pattern (from public `TrackCard.razor`): if `ImagePath` non-empty,
  `background-image: url('api/image/{Uri.EscapeDataString(imagePath)}')`; else a CSS fallback
  swatch. Images served unauthenticated at `api/image/{entryKey}`.

---

## 3. URL scheme — RECOMMENDATION: route segments, not query params

**Recommend: `/tracks` (Track, default), `/tracks/albums`, `/tracks/genres`** as route segments,
with `/tracks?mode=…` accepted as a tolerated alias if cheap.

Three options considered:

| Option | Form | Verdict |
|---|---|---|
| A. Query param | `/tracks?mode=albums` | Single `@page "/tracks"`, read `mode` from query. Simplest routing. |
| B. Route segment | `/tracks/albums` | Multiple `@page` directives on one component, or sub-routes. Cleanest URL, best deep-link target. |
| C. Separate pages | `/tracks`, `/albums`, `/genres` | Three components. Defeats the one-VM spine. Rejected. |

**Why B over A.** The brief's load-bearing requirement is that the *public home page hard-codes*
these URLs. A hard-coded link wants to be a stable, clean, semantically-obvious path. A route
segment (`/tracks/albums`) reads as a permanent address; a query param (`?mode=albums`) reads as
transient view state. Route segments are also the convention already in this app — `/tracks`,
`/tracks/upload`, `/tracks/{id}` are all segment-based, none use query params for navigation
state. Matching that convention keeps the URL space coherent.

**Why tolerate A as an alias.** Blazor lets one component carry multiple `@page` directives.
Adding `@page "/tracks"`, `@page "/tracks/albums"`, `@page "/tracks/genres"` to the one
`TrackList` component (the toggle just sets which mode renders) costs nothing and lets the
existing `?mode=` form (if any external link already uses it) keep working. If no such link
exists, drop the alias — don't carry dead surface.

**Concrete routing:**

```razor
@page "/tracks"
@page "/tracks/albums"
@page "/tracks/genres"
```

The component reads which segment routed it (via `NavigationManager.Uri` parse, or three thin
wrapper pages each setting an initial-mode parameter — see §10.1 for which). The toggle switches
mode *and* pushes the corresponding URL (`NavigationManager.NavigateTo("/tracks/albums")`) so the
address bar always reflects the current mode and deep-links round-trip.

**Decision needed (§10.1):** one multi-`@page` component reading the segment, vs. three trivial
wrapper pages passing an `InitialMode` enum into a shared body component. Recommend the latter
for clarity (each wrapper is two lines; the body owns the VM) — but it's a small call.

---

## 4. View-model design — `CmsTrackBrowserViewModel`

**Registered in DI as a scoped service and injected into the page** — matching the established
project pattern (`TracksViewModel` in `DeepDrftPublic.Client` is registered scoped in
`Startup.ConfigureDomainServices` and injected into `TracksView.razor`). The reason is dependency
hygiene: backend service dependencies (`ICmsTrackService` and the waveform service) live on the VM,
not in the component's `@inject` chain. The component injects only the VM; the VM holds the
services. It owns mode state and the load logic for all three modes. The *views* are dumb; the VM is
the single source of truth.

(This resolves the earlier open question that floated a page-owned `@code` object — see §12.5. The
DI-scoped VM is the decided pattern, not the alternative.)

### 4a. State

```
enum BrowseMode { Tracks, Albums, Genres }

class CmsTrackBrowserViewModel
{
    BrowseMode Mode { get; private set; }              // current mode (drives which view renders)

    // Track mode: the MudTable owns its own ServerData paging; the VM only holds the
    //             active filter (null for plain Track mode).
    string? ActiveAlbumFilter { get; private set; }    // set by Album-mode child expansion? No —
    string? ActiveGenreFilter { get; private set; }    //   filters belong to CmsTrackGrid params, not VM state. See note.

    // Album mode:
    IReadOnlyList<AlbumSummaryDto> Albums { get; }      // parent rows, loaded once on entering Albums mode
    bool AlbumsLoading { get; }
    // child tracks are NOT held in VM; each expanded album row owns its own lazy GetPagedAsync(album:) call.

    // Genre mode:
    IReadOnlyList<GenreSummaryDto> Genres { get; }      // cards, loaded once on entering Genres mode
    bool GenresLoading { get; }
    string? ExpandedGenre { get; }                      // accordion: at most one expanded at a time
}
```

**Note on filters as VM state vs. component params.** There is a real design choice here. Two
readings:

- **VM-as-truth (heavier):** the VM holds `ActiveGenreFilter`, and `CmsTrackGrid` reads it. Then
  the accordion's expansion is VM state.
- **Param-as-truth (lighter, recommended):** `CmsTrackGrid` takes `AlbumFilter`/`GenreFilter` as
  `[Parameter]`s; the *view* passes the expanded genre/album down. The VM holds only `Mode`, the
  summary lists, and `ExpandedGenre` (which genre card is open). The grid filter is derived from
  `ExpandedGenre`, not stored separately.

**Recommend param-as-truth.** It keeps the VM small and keeps `CmsTrackGrid` genuinely reusable
(its filter is an input, not a read of ambient VM state). The VM owns *mode + summaries + which
thing is expanded*; the grid owns its own paging given a filter. This is the cleanest expression
of "one source, multiple views" — the *source* is `GetPagedAsync`, and the grid is the one view
component, parameterised.

### 4b. Which service calls happen where

| Trigger | Call | Owner |
|---|---|---|
| Enter Track mode | none up-front; `MudTable.ServerData` pages on demand | `CmsTrackGrid` |
| Track-mode page/sort | `GetPagedAsync(page, size, sort, desc)` | `CmsTrackGrid.LoadServerData` |
| Enter Album mode | `GetAlbumSummariesAsync()` once | VM |
| Expand an album row | `GetPagedAsync(album: name, …)` lazily, first expansion only | the album row (cached after first load) |
| Album row Edit | navigate to `/tracks/album/{name}/edit` | view |
| Album row Delete | confirm → N× `DeleteTrackAsync` (the album's track ids) | VM/view |
| Enter Genre mode | `GetGenreSummariesAsync()` once | VM |
| Expand a genre card | `CmsTrackGrid` mounts with `GenreFilter=name` → `GetPagedAsync(genre: name, …)` | `CmsTrackGrid` |

State transitions: mode change → push URL → if entering Albums/Genres and the summary list is
unloaded, fire the summary call; collapse any expanded child/genre. Re-entering a mode reuses the
already-loaded summaries (cache for the page lifetime; a fresh page nav reloads).

---

## 5. Component tree

```
TrackList.razor  (@page "/tracks" + "/tracks/albums" + "/tracks/genres")
│   owns: (injects) CmsTrackBrowserViewModel [DI-scoped], the mode toggle, page header
│
├── page header
│     ├── mode toggle  (Track | Album | Genre)  — MudToggleGroup
│     └── "Generate All Missing" button  (Track mode only)              ← §9
│
├── [Track mode]   CmsTrackGrid              (no filter)                 ← DEFAULT
│       └── per-row: status icon + Generate (if no profile) + Edit/Delete/Info  ← §9
├── [Album mode]   CmsAlbumBrowser           (parent rows + lazy children)
└── [Genre mode]   CmsGenreBrowser           (card grid + accordion → CmsTrackGrid)

(navigates to) BatchEdit.razor  (/tracks/album/{albumName}/edit)
```

The old `MudTabs` / "Waveform Pre-Processing" `MudTabPanel` is gone (§9). Whether a `MudTabs` shell
survives at all depends on whether any other panel needs it — likely not.

New components:

### 5a. `CmsTrackGrid.razor` — the single reusable flat track table (THE DRY core)

The extracted Track-mode `MudTable<TrackDto>`. Single source of truth for the track-table layout.

| Parameter | Type | Default | Notes |
|---|---|---|---|
| `AlbumFilter` | `string?` | `null` | When set, passed to `GetPagedAsync(album:)`. |
| `GenreFilter` | `string?` | `null` | When set, passed to `GetPagedAsync(genre:)`. |
| `ShowAddButton` | `bool` | `true` | The "Add Track" button. Off when embedded under a genre. |
| `PageSize` | `int` | `20` | Mirrors today's `RowsPerPage`. |
| `OnTracksChanged` | `EventCallback` | — | Raised after a delete, so a parent can refresh a count. |

Owns its own `MudTable` + `LoadServerData` + delete-confirm dialog (lifted verbatim from today's
`TrackList`). When `AlbumFilter`/`GenreFilter` is set, `LoadServerData` passes it through. This is
the component consumed by **both** Track mode (no filter) and Genre mode (`GenreFilter` set) — no
duplicated table markup, exactly the brief's DRY requirement.

**Columns (new layout, §8 for detail):** Track # | Art (40×40) | Track Name | Artist | Album |
Genre | Release Date (`d MMMM, yyyy`) | **Waveform Status** | Actions (Edit, Delete, per-row
**Generate** when no profile, **Info** tooltip carrying Entry Key + File Name). Entry Key and File
Name columns are *removed* from the grid and moved into the Info tooltip.

### 5b. `CmsAlbumBrowser.razor` — Album mode

Parent album rows (from `AlbumSummaryDto`) that expand to child track rows (lazy `GetPagedAsync`).
See §6 for the data flow. Uses an **expandable `MudTable` with a `RowTemplate` + per-row child
content**, not `MudTreeView` (see §10.2 recommendation). Each parent row: art thumb, album name,
artist (derived), track count, genre (derived), earliest release date, release-type chip, Edit +
Delete actions. Child rows: track # + track name only (a minimal inline template, not a full
`CmsTrackGrid` — the brief is explicit that album children are simpler).

### 5c. `CmsGenreBrowser.razor` — Genre mode

Responsive `MudCard` grid, one card per `GenreSummaryDto` (genre name + track count). Accordion:
clicking a card expands it (and collapses any other) to reveal a `CmsTrackGrid` with
`GenreFilter=genre, ShowAddButton=false` beneath/within. See §7.

### 5d. `BatchEdit.razor` — new page (§ batch edit below)

Reached from `CmsAlbumBrowser` row Edit. See dedicated section.

---

## 6. Album mode data flow — parent eager, children lazy

**Parent rows (eager):** on entering Album mode, the VM calls `GetAlbumSummariesAsync()` once.
Each `AlbumSummaryDto` renders one parent row. The cover thumb uses `CoverImageKey` with the same
`api/image/{Uri.EscapeDataString(key)}` fallback pattern.

**Derived parent-row fields.** `AlbumSummaryDto` gives only `Album`, `TrackCount`, `CoverImageKey`.
The brief asks the parent row to also show artist, genre, earliest release date, and a
release-type chip — **none of which are on the summary DTO.** Two ways to get them:

- **(i) Derive from children on expand only.** Parent row shows album name, track count, cover at
  rest; artist/genre/date/type fill in *after* the row is expanded and the children load. Cheap,
  but the parent row is informationally thin until expanded.
- **(ii) Widen `AlbumSummaryDto`.** Add `Artist?` (or "Various"), `Genre?` (or null→"—"),
  `EarliestReleaseDate?`, `ReleaseType?` computed server-side in the albums-summary query. The
  parent row is fully populated without expansion.

**Recommend (ii) — widen the summary DTO.** The parent row is meant to be a scannable album
catalogue; showing artist/genre/date there is the point of Album mode, and lazy-filling them on
expand makes the resting view feel broken. Computing them in the existing albums-summary SQL is a
modest server change (a `GROUP BY album` with `MIN(release_date)`, a uniform-artist check, etc.),
and it keeps the parent render synchronous. **This is a data-contract change to flag (§ open
questions / §10.3).** If Daniel would rather not touch the DTO, fall back to (i) and accept the
thin resting row.

**"Various" / "—" derivation rules** (wherever they're computed):

- Artist: if all tracks in the album share one artist → that artist; else `"Various"`.
- Genre: if uniform across the album → that genre; else `"—"`.
- Release date: earliest (`MIN`) track `ReleaseDate` in the album.
- Release-type chip: the brief says "infer from TrackCount + ReleaseType, or from first track."
  Simplest defensible rule: use the album's tracks' `ReleaseType` if uniform; else fall back to
  `TrackCount` heuristic (1 = Single, 2–5 = EP, 6+ = LP). Recommend: trust the stored
  `ReleaseType` of the first track (it's set at upload for the whole release) and only use the
  count heuristic if `ReleaseType` is absent. Flag as a small decision (§10.4).

**Child rows (lazy).** On first expansion of an album row, call
`GetPagedAsync(album: albumName, page: 1, pageSize: <large enough for one album>, sort: "TrackNumber")`.
Albums are small (a release is a handful of tracks), so a single page suffices — pick a page size
that comfortably exceeds any real album (e.g. 100) rather than paging within an album. Cache the
result on the row so re-expanding doesn't re-fetch. Child rows render **track number + track
name only** — no per-track Edit/Delete (the brief: editing is via album-level Batch Edit).

**This reuses the same `GetPagedAsync` path as every other mode** — no new endpoint, honouring the
one-source spine. The only new seam is the lazy-on-expand trigger.

---

## 7. Genre mode accordion

- `GetGenreSummariesAsync()` once on entering Genre mode → a responsive `MudCard` grid
  (`MudItem xs=12 sm=6 md=4` or similar), one card per genre: genre name (`Typo.h6`), track count
  (`Typo.body2`), and a fallback visual (genres have no cover art — use a neutral swatch or a
  genre glyph).
- **Accordion, one open at a time.** Clicking a card sets `VM.ExpandedGenre = genre` (clicking the
  open card again clears it). When a genre is expanded, render a `CmsTrackGrid` with
  `GenreFilter=genre, ShowAddButton=false` **below the card grid** (a full-width panel under the
  expanded card's row), not inside the card (a track table doesn't fit a card cell). The
  expanded-card visual state (elevation/border) signals which genre's tracks are showing.
- Layout pattern borrowed from a **master-detail / expanding-panel accordion** (cf. macOS Finder
  column expand, or MudBlazor's own `MudExpansionPanels` — though here the panel hosts a grid, so a
  manual expand region under the card row reads better than nesting a table inside
  `MudExpansionPanel`). Recommend a manual `@if (VM.ExpandedGenre == genre)` panel rendered after
  the card grid, keyed to the expanded genre.
- The grid inside is the *same* `CmsTrackGrid` as Track mode — DRY satisfied: zero duplicated table
  markup, the genre filter is just a parameter.

---

## 8. Track-mode grid changes (the `CmsTrackGrid` layout)

Applied in `CmsTrackGrid` (so Genre mode's embedded grid gets them for free):

1. **Column order (left to right):** Track # → Art thumb → Track Name → Artist → Album → Genre →
   Release Date → **Waveform Status** → Actions. (Track # is the very first column; thumb second;
   per the brief. Waveform Status sits just before Actions — see item 7 and §9.)
2. **Track # column:** plain `@context.TrackNumber`. Header sortable on `TrackNumber` is optional
   (default sort stays Track Name asc to match today).
3. **Art thumb (40×40):** reuse the public `TrackCard` fallback pattern.
   - If `ImagePath` non-empty: a `div` 40×40 with
     `background-image: url('api/image/@(Uri.EscapeDataString(context.ImagePath))'); background-size: cover; border-radius: 4px;`.
   - Else: a neutral fallback swatch (mirror the public `deepdrft-track-row-thumb--fallback`
     class — a CMS-side equivalent in `CmsTrackGrid.razor.css`). Do **not** depend on the public
     client's CSS; define the CMS fallback locally (the public class lives in `DeepDrftPublic.Client`,
     a different host). If the thumb/fallback is genuinely identical to the public one, that's a
     candidate for a shared `DeepDrftShared.Client` component later — flag, don't force now (§10.6).
4. **Release Date format:** display `@context.ReleaseDate?.ToString("d MMMM, yyyy")` (→ "9 June,
   2026"); `"—"` when null. Sorting is unchanged — the sort key is the raw `DateOnly`
   (`SortLabel="ReleaseDate"` still maps to the date column server-side); the format is
   presentation-only and does not touch the sort.
5. **Entry Key + File Name → Info tooltip.** Remove both columns. Add an **Info icon** in the
   Actions cell (`Icons.Material.Filled.InfoOutlined`, `Size.Small`) wrapped in a `MudTooltip`
   whose content shows both values monospaced:
   ```razor
   <MudTooltip>
       <TooltipContent>
           <div style="font-family: monospace; text-align: left;">
               <div>Entry: @context.EntryKey</div>
               <div>File: @(context.OriginalFileName ?? "—")</div>
           </div>
       </TooltipContent>
       <ChildContent>
           <MudIconButton Icon="@Icons.Material.Filled.InfoOutlined" Size="Size.Small" />
       </ChildContent>
   </MudTooltip>
   ```
   (Brief offered row-level tooltip *or* an info icon; recommend the **info icon in Actions** —
   a row-wide tooltip fights the existing per-cell `DataLabel` hovers and is awkward to trigger
   precisely. The icon is an explicit, discoverable affordance.)
6. **Unchanged:** sort labels, paging (`PageSizeOptions { 10, 20, 50 }`), Add Track button (gated
   by `ShowAddButton`), Edit (→`/tracks/{id}`) and Delete (confirm dialog) actions.
7. **Waveform Status column (new — replaces the Waveform tab, see §9).** A status icon showing
   whether a waveform profile exists for the track: `Icons.Material.Filled.CheckCircle` colored
   `Color.Success` when present, `Icons.Material.Filled.Cancel` colored `Color.Warning` when absent.
   This maps to the existing `HasProfile` boolean on `WaveformStatusDto`. **Data-contract decision
   (flagged):** the grid binds `TrackDto`, which has no waveform status today. Two options —
   (a) add `HasWaveformProfile` (bool) to `TrackDto`, or (b) the grid does a separate waveform-status
   lookup and merges by `EntryKey`. **Recommend (a)** — one field, keeps the grid a single data
   source rather than fanning out a second call per page. When `HasWaveformProfile` is false, the
   Actions cell also shows a per-row **Generate** action button (only rendered when the profile is
   missing). See §9.

## 9. Waveform processing — tab removed, folded into the grid

**Decision (review, 2026-06-11): the "Waveform Pre-Processing" `MudTabPanel` is removed entirely.**
Its three responsibilities relocate:

1. **Per-track status** → the **Waveform Status column** in `CmsTrackGrid` (§8 item 7). A
   success/warning icon driven by the profile-exists boolean. Every track's status is visible inline
   while browsing, instead of behind a separate tab.
2. **Per-track generate** → a per-row **Generate** action in `CmsTrackGrid`, rendered only when the
   track has no profile (`!HasWaveformProfile`). Generating refreshes the row's status.
3. **Bulk generate** → a **"Generate All Missing"** button in the **main tracks-page header area**,
   visible when the user is in **Track mode**. This replaces the bulk-run button that lived in the
   Waveform tab.

With the tab gone, the `MudTabs` shell may collapse to a single surface (the mode toggle + the
rendered mode + the page header). Confirm whether `MudTabs` is retained for any other panel or
removed outright. The mode toggle and the "Generate All Missing" button both live in the Track-mode
page header.

Because the status + generate affordances live in `CmsTrackGrid`, **Genre mode's embedded grid gets
them for free** — a genre's tracks show waveform status and per-row generate identically to Track
mode. The "Generate All Missing" page-level button is Track-mode-scoped (it operates over the whole
catalogue, not a filtered slice); whether it should also appear scoped-to-filter in Genre/Album mode
is a minor follow-up, not required for this phase.

This depends on the §8 item-7 data-contract decision (recommend `HasWaveformProfile` on `TrackDto`)
so the grid has the status without a second per-page lookup — which dovetails cleanly with §0's
`TrackDto` rework (the field can be added as part of the normalization DTO pass).

---

## 10. Batch Edit page

`@page "/tracks/album/{AlbumName}/edit"` (URL-encoded album name as the route param).

**Component boundary — CONFIRMED (review, 2026-06-11): a new `BatchEdit.razor` page that shares
extracted sub-components with `BatchUpload.razor`, NOT a `BatchUpload` grown with an `isEdit`
flag.** Sub-components to extract: the album-header fields block (post-§0, edits the `ReleaseDto`),
the batch track list (master list with move-up/down/remove + status chips), and the track detail
pane. Both pages compose these with their own submit logic.

Why not an `isEdit` parameter on `BatchUpload`:

- The two pages diverge in *enough* places that a flag breeds conditionals: edit preloads header +
  rows; edit's track rows have no WAV slot for *existing* tracks but *do* allow adding new ones;
  edit's submit is per-row `UpdateAsync` (existing) + `UploadTrackAsync` (newly added), where
  upload is all `UploadTrackAsync`. A single component with `@if (isEdit)` scattered through the
  master list, the detail pane, and `SubmitAsync` becomes hard to read and easy to break.
- But the *mechanics* are highly shareable: the album-header field block, the master list (rows
  with move-up/down/remove + status chip), the detail pane, the cover-art picker, `FormatBytes`,
  the row model. Those should be **extracted into shared sub-components / a shared partial** that
  both `BatchUpload` and `BatchEdit` compose.

So: extract the reusable pieces (e.g. `AlbumHeaderFields.razor`, `BatchTrackList.razor`,
`BatchTrackDetail.razor`), and have both pages compose them with their own submit logic. This is
the same DRY-by-extraction move as `CmsTrackGrid`, applied to the editor. **Trade-off:** more files
and an upfront extraction cost on `BatchUpload` (which works today). If Daniel wants the smaller
diff, the `isEdit`-flag route is faster to ship but accrues the conditional-soup debt — call it
out, recommend extraction.

**Preload (edit mode):**

- Load the album's tracks via `GetPagedAsync(album: albumName, sort: "TrackNumber")` (same lazy
  call Album mode uses).
- Header fields pre-filled from the first track: Album, Artist, Genre, Release Date, ReleaseType,
  and cover art (`ImagePath` → preview via `api/image/{key}`).
- Master list shows existing tracks ordered by `TrackNumber`, each `TrackName` pre-filled, each
  flagged **existing** (carries its `Id`, `EntryKey`; no WAV slot, shows current file name read-only).
- Newly added rows (via the same WAV `InputFile`) are flagged **new** and carry a WAV file like
  today.

**Row model (extended):**

```
class BatchEditRow {
    long? Id;                 // null = new track (upload), set = existing (update)
    string? EntryKey;         // existing only
    string? OriginalFileName; // existing only, read-only display
    IBrowserFile? WavFile;    // new only
    string TrackName;
    int TrackNumber;          // 1-based, reorderable
    Status; ErrorMessage;
}
```

**Submit (edit mode):** for each row in order, `TrackNumber = position`:

- **Existing row (`Id` set):** `UpdateAsync(Id, TrackName, Artist, Album, Genre, ReleaseDate,
  imagePath, ReleaseType, TrackNumber)`. Note `imagePath` is **tri-state** in `UpdateAsync` (null =
  unchanged, "" = clear, value = set) — pass the (possibly newly uploaded) cover key to every row
  so the whole album shares one cover, or `null` to leave each unchanged if cover wasn't touched.
- **New row (`Id` null):** `UploadTrackAsync(...)` then the cover-link `UpdateAsync` — exactly
  today's `BatchUpload` path.
- If a new cover image was picked, upload it once first (`UploadImageAsync`), then apply its key
  across all rows' `UpdateAsync`. Mirrors the existing one-shot-image pattern.

**Reorder + delete within the album** are natural extensions (the master list already has
move-up/down/remove). Removing an *existing* row should delete the track (`DeleteTrackAsync`) —
flag whether remove-in-edit deletes or just detaches (§10.8); recommend an explicit confirm before
deleting an existing track from within the editor.

**Album-row Delete (from `CmsAlbumBrowser`, not the editor):** confirm dialog scoped to the album
("Delete all N tracks in '{album}'? This removes metadata and audio for every track."), then
`DeleteTrackAsync` for each track id in the album. The brief specifies album-scoped delete; make
the confirm copy unambiguous about the blast radius.

---

## 11. Data contracts — summary of what changes

| Contract | Change | Why | Risk |
|---|---|---|---|
| **`TrackEntity` / `ReleaseEntity`** (§0) | **Split** flat `TrackEntity` into `ReleaseEntity` + slim `TrackEntity` | normalization; pre-requisite gate | **High — breaking migration; see §0** |
| **`TrackDto`** (§0) | Slim down (drop release fields, add `ReleaseId` + Release access path) | follows the entity split | High — touches public + CMS consumers (§0.3) |
| **`ReleaseDto`** (§0) | **New**, mirrors `ReleaseEntity`; retires `AlbumSummaryDto` | Album mode reads the Release table directly | Medium |
| `ICmsTrackService.GetPagedAsync` | **Add** `string? album = null, string? genre = null` params (or an overload) | All three modes' filtered reads route through here; endpoint already supports the query filters. Post-§0 the filter joins through `releases` | Low — additive, default-null keeps callers compiling |
| `CmsTrackService` (impl) | Pass `album`/`genre` through to the `api/track/page` query string | wire the above | Low |
| `TrackDto.HasWaveformProfile` (bool) | **Add** (recommended over a second lookup, §8 item 7) | in-grid waveform status column without per-page fan-out | Low — additive; fold into the §0 DTO pass |
| `AlbumSummaryDto` | **Dissolved by §0** — replaced by `ReleaseDto`; `GetAlbumSummariesAsync` → `GetReleasesAsync`. (The old "widen the DTO" question is moot — the Release table has all the fields.) | normalization | — (replaced, not widened) |
| New row models (`BatchEditRow`) | new, CMS-internal | edit mode | None (internal) |
| Routes | `/tracks/albums`, `/tracks/genres`, `/tracks/album/{albumName}/edit` | mode deep-links + batch edit | Low |

No new API endpoints (existing endpoints' *shapes* change per §0). The data-model change **does**
reach the public client (§0.3) — the browse-mode UI does not, but the `TrackDto` rework does.

---

## 12. Open questions (need Daniel before build)

1. **(§3) Routing mechanism.** Three wrapper pages passing `InitialMode`, vs. one multi-`@page`
   component parsing the segment. Recommend wrapper pages. *Small call.*
2. **(§5b/§10.2) Album expand widget.** Recommend an expandable `MudTable` (parent `RowTemplate`
   + a child-row region toggled per row) over `MudTreeView` — the parent rows are tabular
   (multi-column: thumb, name, artist, count, genre, date, type, actions), which a tree node
   renders poorly. `MudTreeView` suits hierarchies of like items, not parent-rows-of-different-shape.
   Confirm.
3. **~~(§6/§11) Widen `AlbumSummaryDto`?~~ — DISSOLVED by §0.** The normalization gives Album mode a
   real `ReleaseEntity`/`ReleaseDto` with artist/genre/date/type on it directly; there is no summary
   DTO to widen and no derivation from children. The question is moot once §0 lands.
4. **(§6) Release-type chip derivation.** *Mostly dissolved by §0* — `ReleaseType` now lives on the
   Release record, so the chip reads `Release.ReleaseType` directly. The count heuristic is only a
   migration-time concern (deriving `ReleaseType` for legacy releases that had it per-track); flag in
   the §0 migration, not the UI.
5. **~~(§4/§10.5) VM scoping~~ — RESOLVED.** The VM is **DI-registered as a scoped service** and
   injected, matching the `TracksViewModel` pattern in `DeepDrftPublic.Client` (keeps backend service
   deps off the component's inject chain). Not page-owned. See §4.
6. **(§8) Shared thumb component.** The 40×40 art+fallback thumb is near-identical to the public
   `TrackCard` thumb but in a different host. Define locally in `CmsTrackGrid` now; flag a future
   `DeepDrftShared.Client` extraction (Phase 7 territory) rather than coupling the CMS to the
   public client's CSS. Confirm "local now, share later."
7. **~~(§9) Mode URL vs. tab selection~~ — DISSOLVED.** The Waveform tab is removed (§9), so there
   is no tab/mode independence question left. `/tracks/albums` simply lands the page in Album mode.
   If a `MudTabs` shell survives at all, confirm what (if anything) the second panel is — likely
   none.
8. **(§10) Remove-in-edit semantics.** Does removing an *existing* track row in Batch Edit delete
   the track (with confirm), or only detach it from the editing session? Recommend delete-with-
   confirm (a release editor that can't drop a track is half a tool), but it's a destructive
   action worth Daniel's explicit nod.
9. **~~Batch Edit component boundary (§10)~~ — RESOLVED.** Confirmed: a **new `BatchEdit.razor`
   page** sharing **extracted sub-components** with `BatchUpload.razor` — *not* an `isEdit` flag on
   `BatchUpload`. Sub-components to extract: the **album-header fields block**, the **batch track
   list** (master list with move-up/down/remove + status chips), and the **track detail pane**. Both
   `BatchUpload` and `BatchEdit` compose these with their own submit logic. (Post-§0, the
   "album-header block" edits the `ReleaseDto`.) See §10.
10. **Home-page links.** Out of scope for this CMS phase, but: the public home page will hard-code
    `/tracks`, `/tracks/albums`, `/tracks/genres` (cross-host links into the Manager app). Confirm
    the Manager base URL is stable/known to the public site at build time (the URL *paths* are
    settled here; the *host* resolution is a public-site config question, not this phase's).

---

## 13. Suggested sequencing (within the phase)

1. Extend `GetPagedAsync` with album/genre filters (unblocks everything).
2. Extract `CmsTrackGrid` from today's `TrackList` table + apply the §8 layout changes (Track
   mode reaches parity — shippable on its own).
3. Add the mode toggle + routing + `CmsGenreBrowser` (Genre mode reuses `CmsTrackGrid` — cheapest
   second mode).
4. `CmsAlbumBrowser` + the `AlbumSummaryDto` widening decision (§6) (the heaviest mode).
5. Extract `BatchUpload` sub-components + build `BatchEdit` (depends on Album mode's Edit action).

Steps 1–2 deliver visible value (cleaner Track grid) before any mode work lands, and each step is
independently mergeable.