ci: stop testing every commit twice on a runner there is one of

A branch push and its pull request are the same commit. With
`branches: ['**']` alongside `pull_request:`, opening a PR booked four
runs -- check and e2e for the branch, then both again for
refs/pull/N/head -- and this host has capacity 1, shared with an index
build that can hold it for three hours. PR #1's own checks queued two
runs deep behind exactly that.

`pull_request` covers feature branches. `main` stays because a
post-merge run is the record of the trunk's health, and now that main
refuses direct pushes it happens exactly once per merge.

The trade is that a branch pushed with no PR open gets no CI. That
matches the workflow this repo just committed to, and the signal returns
on the same commit the moment a PR exists.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AfVYUVExXsx1nSWrXN8mAh

.gitea/workflows/ci.yml

@@ -9,9 +9,23 @@ name: CI
 # before being written here, so every step below is a transcription of
 # something observed working rather than something expected to.
 
+# **A branch push and its PR are the same commit, and testing it twice
+# costs the only runner there is.**  `branches: ['**']` here meant every
+# PR booked four runs — `check` and `e2e` for the branch push, then both
+# again for `refs/pull/N/head` — on a host with capacity 1, where the
+# queue is shared with an index build that can hold it for three hours.
+#
+# `pull_request` covers feature branches, and `main` is kept because a
+# post-merge run is the record of the trunk's health.  Since main now
+# refuses direct pushes, that run happens exactly once per merge.
+#
+# The trade is explicit: a branch pushed with **no** PR open gets no CI.
+# That is consistent with the workflow this repo committed to — every
+# change goes through a PR — and the signal returns the moment one is
+# opened, on the same commit.
 on:
   push:
-    branches: ['**']
+    branches: [main]
   pull_request:
   workflow_dispatch:
 

.gitea/workflows/index-artifact.yml

@@ -7,11 +7,29 @@ name: Search index maintenance
 #   import older than 6mo -> rebuild (re-import from the newest dump)
 #   otherwise             -> refresh (fold in new incremental listens)
 #
-# A refresh is cheap and no-ops when nothing new has been published, so
-# running it on every push to main is safe.
+# **There is deliberately no `push` trigger, and restoring one is a
+# decision rather than a cleanup.** A refresh is individually cheap, so
+# running it on every push to main looked free; what it actually does is
+# put an unattended job that mutates the only copy of a ~205 GB catalog
+# on the same trigger as an ordinary code change, on a runner with
+# capacity 1.
+#
+# That is not hypothetical. On 2026-08-17 `fix(database): retire a table
+# whose shape the schema moved past` landed on main, green — the CI
+# database is deliberately in the older encoding, so the stale-shape
+# repair judged its `explore_index` stale and dropped it, and this job
+# fell back to a full import from the dumps. `fix(database): never
+# retire the catalog the index build derives` stops that specific repair
+# and cannot undo it. Every push to main then booked another `budget`
+# (3h) of the one runner while ordinary CI queued behind it.
+#
+# So the rule this file is an instance of: **a job that mutates state
+# which cannot be rebuilt in ten minutes is triggered deliberately, not
+# by a push.** The weekly cron keeps the catalog current, and
+# workflow_dispatch resumes or forces a build — indexbuild picks up from
+# its checkpoint either way, so nothing is lost by not running on every
+# merge. See docs/index-cache.md for the snapshot and the restore.
 on:
-  push:
-    branches: [main]
   schedule:
     # Weekly update pass. The 6-month rebuild is triggered by the same
     # command when it notices the import has aged out.
@@ -33,6 +51,10 @@ on:
 
 # Runs share one persistent working directory, so they must not overlap.
 # A push landing mid-build waits rather than corrupting the checkpoint.
+#
+# That directory holds the only copy of a catalog nothing can cheaply
+# re-derive: see docs/index-cache.md for the snapshot it takes and the
+# restore, which is minutes against the hours a rebuild costs.
 concurrency:
   group: search-index
   cancel-in-progress: false

.planning/NOTES.md

@@ -3338,3 +3338,110 @@ Four things measured or corrected in the doing:
 - **The gate belongs before the first status write.** Declining is a
   no-op — no job in the indicator, no error tier to dismiss — which is
   what makes the refusal safe to have on by default.
+
+## The stale-shape repair dropped the CI catalog (2026-08-17)
+
+Not our change, but it is the operational state everything else now runs
+in, and the restore condition needs to be written down somewhere that is
+not a commit message.
+
+`fix(database): retire a table whose shape the schema moved past` added
+`staleshape.go`: before `applySchema`, drop any non-Authored table whose
+live shape disagrees with the schema. That is the right rule for an
+install — a client's catalog is *downloaded*, so a stale one costs a
+minute of re-fetching the artifact, and keeping it costs every Explore
+read.
+
+It runs inside `database.NewDB`, which `cmd/indexbuild` also calls. On
+the first run after it landed, 19 seconds in:
+
+```
+16:15:51 retiring a table ... table=explore_index
+           reason="column entity_type is TEXT, schema declares INTEGER"
+16:16:05 index maintenance mode=build reason="no completed import yet"
+           lastImported=never baselineSeries=0
+```
+
+**The premise was false for the one database where it was expensive.**
+That catalog is not stale; it is deliberately kept in the older text
+encoding, which `artifactStoresText` and `sourceColumns` exist to
+tolerate — so it would have been judged stale and dropped on *every*
+run. And `retireLibraryTables`, in the same package, already documents
+the opposite rule for this database: drop everything the datamap does
+**not** call Cache.
+
+`fix(database): never retire the catalog the index build derives` makes
+the policy a build tag (`retireStaleCache`, false under `indexbuild`),
+which is how this project already separates the index tools. It prevents
+recurrence and cannot undo the drop: that volume was the only copy.
+
+**What it cost, and the shape of the cost.** A full re-import from the
+MetaBrainz dumps, resumed across runs from a checkpoint, at a rate that
+swung between 2 and 15 MB/s. The job runs on **every push to main** with
+a 3 h budget on a runner of capacity 1 — so until the import completes,
+every push books three hours and ordinary CI queues behind it. That is
+the real damage: not one lost job, but a repeating one.
+
+So the `push:` trigger in `index-artifact.yml` is **commented out**
+until a run reports `complete=true`; the weekly cron and
+`workflow_dispatch` still resume the build, which is all it needs.
+Restoring those two lines is the whole revert.
+
+Three things worth keeping from it:
+
+- **A repair belongs where its assumptions hold.** `NewDB` is the one
+  chokepoint every binary in this project shares, including the one
+  whose database cannot be re-derived cheaply. Anything destructive
+  there needs to ask which binary it is in — the build tag was available
+  and is what the fix used.
+- **The only copy of a 205 GB derived asset is one Docker volume.**
+  There is no snapshot, so the restore time is "however long
+  MetaBrainz takes today". A periodic copy would turn this class of
+  incident into twenty minutes.
+- **The fix's residual trade is now the thing to watch**: with Cache
+  tables never retired under `indexbuild`, a future `explore_index`
+  column fails that job loudly at build time instead of silently
+  rebuilding. That is the right default, and it means the next schema
+  change touching `explore_index` needs a deliberate plan for this one
+  database rather than none.
+
+## Two guards for the index cache, and what each one is worth (2026-08-17)
+
+Both come out of the incident above, and they protect different halves
+of it.
+
+**`TestNoCacheTableIsRetiredHere` asserts the outcome, not the
+mechanism.** The test that shipped with the fix pins one table in one
+wrong shape, which is the failure that happened; what actually cost the
+rebuild was a destructive repair added at `database.NewDB` — the
+chokepoint every binary here shares — without asking which binary it was
+in. The next one will have a different name and a different reason. So
+this puts *every* `datamap` Cache table into a shape the schema has
+moved past, opens the database the way `cmd/indexbuild` does, and
+requires all of them to still be there.
+
+Three things it got right by being written this way. The table list is
+`datamap.ByKind(Cache)`, so the two credit tables added the same day
+were covered without anyone adding them — flipping the policy back fails
+on **five** tables including `artist_credit_part` and
+`artist_credit_ref`, where the single-table test fails on one. It
+asserts rows survive as well as the table, because SQLite does an
+implicit DELETE before a DROP and a repair that recreated the table
+would otherwise look identical. And it *accepts* an error from `NewDB`,
+because that is the trade the fix documents: loud failure instead of a
+silent day of downloading.
+
+**`scripts/index-cache-snapshot.sh` covers the half no test can.** The
+volume held the only copy of a catalog whose rebuild is hours of someone
+else's bandwidth. `VACUUM INTO` rather than `cp`, because a byte copy of
+a live SQLite file is a corrupt file of plausible size; the staging
+directory is deliberately not copied, since a build resumes without it;
+and the snapshot is reopened and asked for its catalog row count before
+any rotation happens. Both failure paths were exercised rather than
+argued: a corrupt source and an empty catalog each exit non-zero, delete
+their own output, and leave the previous snapshots in place.
+
+`docs/index-cache.md` is the restore procedure, and the number that
+makes it worth having: a restored snapshot resolves to `refresh` and
+folds in the incremental listens since — minutes, against the 3–23 h a
+rebuild was estimating.

.planning/plans/active/015-multi-artist-credits.md

@@ -0,0 +1,337 @@
+# 015 — Multi-artist credits, navigable
+
+## The problem
+
+A track credited to more than one artist has exactly one navigable
+artist in this app, and the others are punctuation.
+
+`audio_files` carries `artist_credit` (the credit as tagged, for
+display) and `artist_id` (one artist, for grouping and browsing).
+`primaryArtist()` (`backend/library/artistcredit.go:53`) resolves that
+one artist by *string-parsing* the credit: it strips a " feat. "
+clause, and deliberately does not split on `&`, `x`, `with` or `,`
+because those appear inside real artist names. So "Lana Del Rey ft.
+Sean Lennon" stores Lana Del Rey and discards Sean Lennon entirely,
+and "Alina Baraz & Galimatias" stores one artist whose name is the
+whole credit.
+
+### What the measurement says
+
+Measured 2026-08-16 against a real 26,069-file library (19,840 mp3,
+6,229 flac; 57 unreadable, m4a/ogg not examined), plus an 80+80
+MusicBrainz `inc=artist-credits` sample.
+
+- **13%** of a random sample of the library's recordings have more
+  than one credited artist in MusicBrainz (10 of 79 resolved).
+  Extrapolates to ~3,250 of the 24,989 files carrying a recording
+  MBID.
+- **0.86%** of files (224) carry any structured multi-artist signal in
+  their own tags. mp3 carries **zero** files with multiple
+  `MUSICBRAINZ_ARTISTID` values across 19,840 files; flac has 87.
+- **1,286** files say "feat." in `ARTIST`; **1,159 of them (90%)**
+  have nothing structured behind it. A sample of 80 such files was
+  multi-artist in MB **80 of 80 times**.
+
+CLAUDE.md currently justifies plan 013's removal of `artist_credit` /
+`artist_credit_artist` with "3 credits of 2,823 listed more than one
+artist". That figure measured **our own writer**, not the library:
+`cachedLinkArtist` was called exactly once per credit
+(`e7748f1^:backend/library/library.go:1842`), so a collaboration could
+never have been recorded, and the three were resolution collisions on
+shared credit text. Dropping the join table was still correct — it only
+ever held one row, so it was pure join cost — but the stated evidence
+does not support "multi-artist is rare". Correcting that claim is part
+of this plan.
+
+### Why the tags cannot answer it
+
+Deriving the decomposition locally, with no network, works **79% of the
+time** (169 of 215 files with a multi-value `ARTISTS` tag: mp3 69/105,
+flac 100/110), and the failures are systematic rather than random:
+
+```
+ARTIST  = '2Pac feat. Snoop Dogg, Nate Dogg, Hussein Fatal & Yaki Kadafi'
+ARTISTS = ['2Pac', 'Snoop Doggy Dogg', 'Nate Dogg', 'Fatal', 'Yaki Kadafi']
+```
+
+`ARTISTS` holds **canonical** artist names; `ARTIST` holds
+**as-credited** names. Locating one inside the other fails on
+"Snoop Doggy Dogg" vs "Snoop Dogg", on "Fatal" vs "Hussein Fatal", and
+on Unicode (`Michel'le` vs `Michel’le`, `K-Ci` vs `K‐Ci` — U+2010, not
+a hyphen). That distinction is precisely what a join phrase encodes,
+and it is why this cannot be a tag-parsing feature.
+
+Two format details that will mislead anyone re-running the probe:
+Picard writes `ARTISTS` **slash-joined into one TXXX frame** on mp3 and
+as **true repeated Vorbis keys** on flac, so a probe splitting only on
+NUL undercounts mp3 to zero.
+
+## The shape
+
+MusicBrainz models a credit as ordered parts, and the credit *string*
+is derived from them — `artist_credit.name` is a cached render, nothing
+more. Each participant is `(position, artist, name, join_phrase)`,
+where `artist` is the MBID (canonical, what you navigate to) and `name`
+is the credited spelling (what you display).
+
+**Join phrases are assembly instructions, not disassembly
+instructions.** Rendering is a concatenation, never a search:
+
+```
+for each (position, artist_mbid, credited_name, join_phrase):
+    emit link(credited_name -> artist_mbid)
+    emit text(join_phrase)
+```
+
+The link positions are known **by construction**. This is load-bearing:
+if we instead located each `credited_name` inside the stored
+`artist_credit` text, we would reintroduce the mismatch above — the
+stored string may have come from the tags while the parts come from the
+catalog, and those **disagree for ~1 in 3 multi-artist files** (61 of
+90 sampled credits rendered exactly equal to the tag string).
+Divergences seen: `'Skrillex feat. Swae Lee'` tagged vs
+`'Skrillex & Swae Lee'` in MB; `'STRFKR'` vs `'Starfucker'`;
+`'Zedd feat. Hayley Williams'` vs `'... of Paramore'`. Either MB was
+edited after tagging or Picard versions differ; either way the search
+would miss or match the wrong span.
+
+So `audio_files.artist_credit` stops being the source of truth and
+becomes the **fallback**, used only where there are no parts.
+
+## Where the data comes from
+
+The catalog carries the decomposition; no user ever makes a
+per-recording call. Two sources were ruled out first, both cheaply:
+
+- **The canonical dump — which is what CI already pulls
+  (`dumpimport.go:84-85`) — does not have it.**
+  `canonical_musicbrainz_data.csv` gives `artist_mbids` (ordered list)
+  and `artist_credit_name`, but that last column is the *rendered*
+  string. Splitting it on CI needs the as-credited names, so CI would
+  fail exactly the way a local parse does.
+- **The JSON dumps do not cover the catalog.**
+  `json-dumps/recording.tar.xz` is 31 MB / 368 MB uncompressed and
+  holds **153,691 recordings**, not ~35M. Measured against the test
+  library's 24,885 recording MBIDs: **0.00% overlap, zero rows**. It is
+  some other subset and is not usable.
+
+That leaves the core dump, **`mbdump.tar.bz2`** (7.1 GB compressed at
+the 20260815 export), from
+`https://data.metabrainz.org/pub/musicbrainz/data/fullexport/`. Four
+members are needed:
+
+| member | why | approx rows |
+| --- | --- | --- |
+| `mbdump/artist_credit_name` | `(artist_credit, position, artist, name, join_phrase)` — the payload | ~4M |
+| `mbdump/artist` | `id -> gid`, since the above references artist *row ids* | ~2.6M |
+| `mbdump/recording` | `gid -> artist_credit`, to key credits by recording MBID | ~35M |
+| `mbdump/release_group` | same, for album credits | ~2M |
+
+### Coverage is not a concern
+
+Of 24,885 distinct recording MBIDs in the test library, **24,808
+(99.7%)** already have an `explore_index` recording row, measured
+against a database at 2,052,200 rows — i.e. shipped-artifact coverage,
+not a local build's. The popularity filter does not strand the long
+tail here.
+
+## Status
+
+- **Phase 1 — done.** `backend/explore/dumpcredits.go` +
+  `dumpcreditswrite.go`, wired into `dumpimport.go`'s `run` behind its
+  own `credits_import_done` marker.
+- **Phase 2 — done.** `cmd/indexexport` writes the two tables;
+  `artifactimport.go` reads them behind `artifactHasCredits()`.
+- **Phase 4 — done, and it does not need Phase 3.** `explore.GetCredits`
+  reads the catalog tables keyed on the *recording* MBID, which both
+  sides of the app already carry — a catalog row has one and so does a
+  local file (`library.Track.RecordingMBID`). So one binding serves the
+  Explore pages and the library's own lists, and all ten artist-link
+  call sites render credits today without a local table.
+- **Phase 3 (`file_artists`) — not started, and now an
+  offline-resilience task rather than a prerequisite.** The table is
+  deliberately *not* declared yet: nothing writes or reads it, and a
+  schema file plus a datamap note describing behaviour that does not
+  exist is a claim the code cannot back. Its remaining
+  value is that credits currently vanish when the catalog is absent or
+  still downloading, which is precisely the `no-index` state
+  `ShelfPage.State` exists to describe. Materialising into
+  `file_artists` is what makes a library stand on its own.
+
+**Nothing renders yet in practice**, because no published artifact
+carries credit tables — every credit falls back to its single link
+until an index build with Phase 1 runs and is exported.
+
+**Column layouts are verified against the real 20260815 export**, not
+taken from the schema docs — `artist(id, gid, …)`,
+`artist_credit(id, name, artist_count, …)`,
+`artist_credit_name(credit, position, artist, name, join_phrase)` and
+`recording(id, gid, name, artist_credit, …)` were each read out of the
+dump. `release_group` shares `recording`'s first four columns and is
+the one layout still taken on trust; `ErrDumpShape` turns a wrong guess
+into a loud failure rather than a quietly wrong catalog.
+
+**Still unrun: the ingest against the real 7.1 GB dump.** Everything is
+covered by tests over a synthetic tar, which cannot catch a surprise in
+the other ~35M rows.
+
+### Phase 1 — Ingest credits on CI
+
+New dump stage in `cmd/indexbuild`, behind the `indexbuild` tag with
+the rest of `dumpimport.go`'s stages.
+
+**Constraint from `b98840e`:** `cmd/indexbuild` is built
+`CGO_ENABLED=0` in a plain `golang` container and must not reach the
+Wails `application` package — `TestIndexToolsDoNotImportWails` walks
+`go list -deps -tags indexbuild`. Nothing here should need it, but a
+new `ServiceStartup` hook on a package this imports is how it comes
+back. Go's `compress/bzip2` is pure Go and decompress-only, which is
+all this needs.
+
+**Measured, 20260815 export.** Tar members are **alphabetical**, and
+that is favourable: `artist` (435 MB), `artist_credit` (414 MB) and
+`artist_credit_name` (237 MB) all fall inside the first ~900 MB
+compressed, while `recording` and `release_group` come later. So the
+maps are complete before the rows that consume them arrive, and no
+recording data is ever buffered.
+
+Pure-Go `compress/bzip2` decompresses at **26 MB/s uncompressed /
+8.7 MB/s compressed** (measured on a 250 MB prefix, 3.01x ratio) —
+**~13.7 min** for the whole file single-threaded, and less because the
+stream can stop after `release_group` rather than reading the
+`series`/`tag`/`track`/`url`/`work` tail. The 2 MB/s origin throttle
+dominates, as it already does for every other dump here.
+
+Do not, however, *depend* on the ordering: assert it and fall back to
+buffering if a future export reorders, rather than silently emitting
+nothing.
+
+- `artist` -> `map[int32]uuid16` (~2.6M x ~20 B = ~60 MB)
+- `artist_credit_name` -> `map[int32][]creditPart` (~4M x ~40 B =
+  ~200 MB)
+- `recording` / `release_group` -> emit `gid -> credit_id` **only for
+  MBIDs already in `explore_index`** (the kept set is ~1.4M x 16 B =
+  ~22 MB), which is what keeps 35M rows from being held
+
+Peak ~300 MB, one sequential pass.
+
+**Only multi-artist credits are stored.** A single-artist credit is
+`(name, "")` and is already fully described by `explore_index`'s
+`artist_name` / `artist_mbid`; storing it would triple the table for
+nothing. Post-filter after loading, once the row count per credit is
+known.
+
+New tables (and `datamap` entries, or `TestCatalogCoversSchema` fails
+the build — both are `Cache`, matching `explore_index`):
+
+```
+artist_credit_part(credit_id, position, artist_mbid, credited_name, join_phrase)
+```
+
+with `explore_index.artist_credit_id` as the link. Credits are
+**shared** — an album's twelve tracks by one artist share one credit
+row — which is the opposite of 013's local verdict, and correctly so:
+1:1 in a local library, genuinely many-to-one at 2M-row catalog scale.
+
+### Phase 2 — Ship them in the artifact
+
+`cmd/indexexport` currently creates exactly two tables in the artifact
+(`explore_index`, `artifact_meta`, at `cmd/indexexport/*.go:147,170`),
+so this is a structural addition, not a column.
+
+Estimated size: ~13% of 1.4M recordings, deduplicated by shared credit,
+at ~2.3 parts each — order 400k rows, ~18 MB uncompressed. Against a
+~0.6 GB install that is acceptable; it must be measured rather than
+assumed before merge.
+
+`artifactimport.go` must read it **only if present**, on the writer
+handle where `core` is attached — the `artifactHasTotals()` /
+`artifactStoresText()` pattern (`artifactimport.go:145-175`), one step
+up from a column to a table. An artifact published before this exists
+is still a perfectly good catalog and must import as one that declines
+to answer. Adding this to the importer's SELECT list without the probe
+is how every already-published artifact starts failing.
+
+`artifactCatalogColumns` gains `artist_credit_id`; it is kept in sync
+with the exporter by `TestArtifactColumnsMatchExporter`.
+
+### Phase 3 — Materialize locally
+
+```
+file_artists(audio_file_id, position, artist_id, credited_name, join_phrase)
+```
+
+`credited_name` is stored **per row**, not looked up from
+`artists.name` — that is the Snoop-Doggy-Dogg distinction, and it is
+the whole point.
+
+Filled at scan/import time by joining `audio_files.recording_mbid`
+against the catalog. **Materialized rather than resolved live**,
+because the catalog is a downloaded artifact that can be absent or
+still arriving — that is why `ShelfPage.State` has a `no-index` value —
+and a library whose track rows lose their artists when the catalog is
+missing is worse than today.
+
+That implies a backfill for the case where the catalog arrives *after*
+the library was scanned. It registers with `jobs` (progress, cancel)
+like every other long pass, and takes a **distinct kind** from
+`index-build`, since `job-controls.ts` keys its "you will discard hours
+of downloading" confirmation on that kind.
+
+`artists` gains rows for guests who own no files. **This changes what
+the artists grid shows** and is an open question below.
+
+### Phase 4 — Render
+
+`utils/explore-link.ts` gains a credit-rendering entry point taking
+ordered parts and returning a `TemplateResult`. Every row and detail
+view already renders artist names through it, so they inherit
+multi-artist links without individually knowing credits exist — the
+property that made centralising it worthwhile.
+
+Its existing fallback philosophy already covers the no-parts case: "a
+list where some rows are clickable and others silently are not reads as
+a bug, not as a statement about metadata." Where there are no parts
+(no recording MBID, or no catalog row — ~4% of the test library) render
+today's behaviour: the flat `artist_credit` string with one link to the
+primary artist. **Do not split the string there.** There is genuinely
+no information to split on, and that is the one place the temptation
+returns.
+
+`primaryArtist()` stays exactly as it is. It remains the fallback and
+is still what `artist_id` means.
+
+## Open questions
+
+1. **Catalog credit vs tagged credit, when they disagree** (~1 in 3
+   multi-artist files). Rendering the catalog's decomposition is what
+   makes names navigable; preserving the file's is what makes the app
+   reflect the user's files. Leaning toward: render the catalog
+   decomposition, keep `artist_credit` as the fallback string. Wants a
+   deliberate decision, not an accident.
+2. **Do guest artists appear in the artists grid?** Phase 3 creates
+   `artists` rows for people who own no files. The grid currently means
+   "artists in your library" and joins `audio_files`. A guest on one
+   track is arguably in the library and arguably not. Whichever way,
+   the ownership question stays "is there a file" — that rule does not
+   bend.
+3. **`release_group` credits** are ingested in the same pass for
+   nearly nothing, but album-artist rendering is a separate surface.
+   Ship the data in phase 1, render in a follow-up rather than widening
+   phase 4.
+4. **Our own `tagwriter`** does not write `ARTISTS` or multiple
+   `MUSICBRAINZ_ARTISTID` frames, so autotagging a folder degrades the
+   very field this rests on — the same shape as the existing
+   track-totals note. Out of scope here; worth recording.
+
+## Verification
+
+- Coverage: re-run the library probe and assert `file_artists` is
+  populated for ~13% of files, not ~0.9%.
+- `TestCatalogCoversSchema` / `TestLifetimesMatchSchema` for the new
+  tables.
+- `TestIndexToolsDoNotImportWails` still passes with the new stage.
+- An artifact **without** the credits table imports cleanly (the
+  `artifactHasTotals` regression shape).
+- Round-trip: a known multi-artist recording renders each name as a
+  separate link with the correct join phrases between them.

AGENTS.md

@@ -0,0 +1 @@
+CLAUDE.md

CLAUDE.md

@@ -233,6 +233,74 @@ rather than renaming them.
   the drift it caused before — `sql/schemas/` and the migrations
   disagreed, and sqlc generated against the stale one.
 
+  **What that costs an existing database is repaired once, at open.**
+  `CREATE ... IF NOT EXISTS` reaches an existing table only if its shape
+  already matches and otherwise silently no-ops, so a *changed* table
+  never migrates. Plan 014 added `total_tracks` to `explore_index` and
+  to `indexRowFields` — the projection every explore read uses — and no
+  database that already existed grew the column: **every** Explore
+  search, browse, artist and album page on such an install failed with
+  `no such column: total_tracks`, while a fresh install was perfectly
+  healthy, which is exactly why no test saw it. Plan 013 was worse on
+  the same install: `applySchema` could not be applied at all over a
+  pre-013 `audio_files`, so the app did not open.
+
+  `backend/database/staleshape.go` runs before `applySchema` and
+  retires what is stale, so the create is a create. Five things about
+  it are load-bearing:
+  - **It parses `sql/schemas/` for the expectation** rather than
+    writing the column list down a second time, because a second list
+    is a second thing to forget — the fault it exists to repair.
+  - **It notices a changed *type*, not just a missing column.** 013
+    moved `mbid` from TEXT to BLOB, and SQLite does not coerce between
+    them: a comparison against 16 raw bytes returns no rows rather than
+    an error. `ALTER TABLE ADD COLUMN` would have handled
+    `total_tracks` alone and cannot express this at all, which is why
+    the repair drops rather than migrates.
+  - **`Authored` is never retired**, and that boundary is a test
+    (`TestAuthoredTablesAreNeverRetired`), not a comment. Everything
+    else is rebuildable: `Cache` by definition, `Owned` by a rescan —
+    plan 013's stated "delete and rescan" — and `Derived` from Owned.
+    A table the schema no longer describes at all goes too; 013 left
+    seven behind plus `schema_migrations`.
+  - **Whether a stale `Cache` table may be rebuilt is a build tag**, and
+    it is the most expensive thing in this file to get wrong. In the app
+    the catalog is *downloaded*, so a wrong shape costs a minute of
+    re-fetching the artifact and keeping it costs every Explore read. In
+    `cmd/indexbuild` the catalog is *derived*, and the only way back is
+    the ~205 GB dump stream the `/cache` volume exists to avoid — so
+    `retireStaleCache` is false there (`staleshape_policy_indexbuild.go`)
+    and `TestTheCatalogSurvivesAStaleShape` fails the moment it is not.
+    `TestNoCacheTableIsRetiredHere` is the same assertion made of *every*
+    `datamap` Cache table rather than one, because the risk is not that
+    shape recurring — it is the next destructive repair added to
+    `database.NewDB`, the chokepoint every binary here shares, without
+    asking which binary it is in.
+    This is written down because it already happened: the repair shipped
+    without the distinction and dropped the real CI catalog on its first
+    run, with `reason="column entity_type is TEXT, schema declares
+    INTEGER"`. The mismatch was genuine — that database is deliberately
+    kept in the older encoding, which `fix(indexexport): read an index
+    older than the binary` exists to tolerate — so it would have been
+    dropped on *every* run. The consequence is that a future
+    `explore_index` column fails the index job loudly on `applySchema`
+    rather than silently costing it a rebuild, which is the trade a
+    human should get to make.
+  - **The drops are one transaction with `defer_foreign_keys`.** Those
+    legacy tables reference each other, so dropping them in any order
+    fails on whichever goes first, and turning foreign keys *off*
+    instead would silently take `playlist_tracks.audio_file_id`'s
+    ON DELETE SET NULL with it — leaving playlist entries pointing at
+    ids a rescan reissues to *different songs*. Nulled entries are
+    empty; stale ones are wrong, and wrong quietly.
+  - **The order is sorted, so a failure reproduces.** Map order is
+    random, and the foreign-key bug above passed its own regression
+    test on two runs in three until the order was fixed.
+
+  Retiring `explore_index` takes its FTS and its meta with it, because
+  the `dump_import_done` marker is what would otherwise stop the
+  artifact ever being fetched again.
+
   **What that costs an existing database is that it does not open**, and
   "delete and rescan" is the answer (plan 013, open question 1) — free
   for everyone except one machine. The index job's `/cache` volume is a
@@ -508,6 +576,56 @@ selected as a literal `0`. Adding the column to the importer's SELECT
 list without that is how a published artifact — which nobody can re-cut
 retroactively — starts failing with `no such column`.
 
+**A credit is ordered parts, and the string is derived from them.** A
+track credited to several artists had exactly one navigable artist and
+the rest were punctuation: `primaryArtist()` string-parses the credit,
+strips a " feat. " clause and discards the guest, and deliberately does
+not split on `&`, `with` or `,` because those live inside real artist
+names ("Simon & Garfunkel"). Measured on a real 26,069-file library,
+**13%** of recordings are multi-artist upstream while only **0.86%** of
+files carry a structured multi-artist tag — mp3 carries *zero* files
+with multiple `MUSICBRAINZ_ARTISTID` across 19,840 — so this cannot be
+a tag-parsing feature. (The "3 credits of 2,823" figure that justified
+plan 013's removal of the credit tables measured our own *writer*:
+`cachedLinkArtist` ran once per credit, so a collaboration could never
+have been recorded. Dropping the join table was still right on cost.)
+
+`artist_credit_part` / `artist_credit_ref` carry the decomposition for
+multi-artist credits only — a single-artist credit is already
+`explore_index`'s own `artist_name`, and storing those would triple the
+table to say nothing. Five things about it are load-bearing:
+
+- **Join phrases are assembly instructions, not disassembly ones.**
+  `creditLink` concatenates parts, so link boundaries are known by
+  construction. Locating a `credited_name` *inside* the stored credit
+  string would reintroduce the fault this exists to fix: that string may
+  come from the file's tags while the parts come from the catalog, and
+  the two disagree for ~1 in 3 multi-artist credits (`'Skrillex feat.
+  Swae Lee'` tagged against `'Skrillex & Swae Lee'` upstream).
+- **`credited_name` is stored per row**, never joined from `artists`:
+  MusicBrainz credits "Snoop Dogg" on a track by the artist called
+  "Snoop Doggy Dogg". Display follows the credit, navigation the MBID.
+- **The lookup is keyed on the recording MBID**, which the catalog and a
+  local file both carry (`library.Track.RecordingMBID`), so one binding
+  serves Explore and the library's own lists — which is why this needed
+  no local table. `file_artists` remains the offline-resilience step and
+  is deliberately *not* declared until something writes it.
+- **Absence is cached as an answer.** `credit-store.ts` stores `[]` for
+  a single-artist credit — *asked*, not *answered* — or the ~87% that
+  have nothing to decompose are re-requested on every render forever.
+  `request()` is per-row and coalesces into one call per frame, because
+  a virtualized list cannot hand over "the whole list": 50,000 rows is
+  100 queries for the ~30 on screen.
+- **The dump is a third source, and it had to be.** The canonical dump
+  CI already streams has no join phrases and no as-credited names, and
+  the JSON dumps cover 153,691 recordings of ~35M with *zero* overlap
+  against a real library. So `mbdump.tar.bz2` — 7.1 GB, ~13.7 min in
+  pure-Go bzip2, whose members are alphabetical, which is what lets one
+  pass resolve an entity's credit without buffering 35M recordings. The
+  pass runs on **every** mode, because a complete import means
+  `refresh`, which never enters the importer at all, and it reports
+  whether it populated anything so `changed` republishes the artifact.
+
 **A 0.6 GB download asks about the connection first.** `explore`'s
 catalog artifact had no network awareness at all, which on a phone is a
 month's data allowance spent without being asked (plan 016 B4).

Makefile

@@ -179,10 +179,12 @@ bindings-check: ## Fail if the generated bindings are stale
 css-check: ## Fail if a css`` literal was ended early by a backtick in a comment
 	@cd frontend && node scripts/check-css-literals.mjs
 
-# .pi/ documents commands, and a skill that documents a command wrongly
-# is worse than no skill: an agent runs it confidently.  Every command
-# in there is a make target on purpose, so this is checkable.
-skill-check: ## Fail if .pi/ documents a make target that does not exist
+# .pi/ and CLAUDE.md document commands, and a doc that documents a
+# command wrongly is worse than no doc: an agent runs it confidently.
+# Every command in them is a make target on purpose, so this is
+# checkable.  It also asserts AGENTS.md is a symlink to CLAUDE.md, so the
+# two harnesses cannot drift onto two descriptions of one project.
+skill-check: ## Fail if the agent docs name a missing make target, or AGENTS.md is not a symlink
 	@./scripts/skill-check.sh
 
 # Conventional Commits, which CLAUDE.md claimed CI enforced for a long

backend/database/database.go

@@ -89,6 +89,14 @@ func NewDB(logger *slog.Logger) (*DB, error) {
 		return nil, fmt.Errorf("could not apply PRAGMAs: %w", err)
 	}
 
+	// Before the schema is applied, not after: applySchema is
+	// CREATE ... IF NOT EXISTS, which no-ops against a table that
+	// already exists in an older shape.  Retiring the stale one first is
+	// what turns that no-op into a create.
+	if err := retireStaleTables(dbCtx, db, logger); err != nil {
+		return nil, err
+	}
+
 	if err := applySchema(dbCtx, db); err != nil {
 		return nil, err
 	}

backend/database/sql/schemas/artist_credit_part.sql

@@ -0,0 +1,56 @@
+-- The decomposition of a multi-artist credit, from the MusicBrainz
+-- dump.  One row per credited artist, in credit order.
+--
+-- A credit is ordered parts, and the credit *string* is derived from
+-- them -- MusicBrainz's own `artist_credit.name` is a cached render and
+-- nothing more.  Rendering is a concatenation:
+--
+--     for each part in position order:
+--         emit link(credited_name -> artist_mbid)
+--         emit text(join_phrase)
+--
+-- so the link boundaries are known by construction.  That is the whole
+-- reason this table exists, and it is why nothing may reconstruct a
+-- credit by *searching* for a name inside a credit string: the stored
+-- string may have come from a file's tags while the parts come from the
+-- catalog, and measured on a real library those disagree for about one
+-- in three multi-artist credits ("Skrillex feat. Swae Lee" tagged
+-- against "Skrillex & Swae Lee" upstream).  A search would miss, or
+-- match the wrong span.
+--
+-- `credited_name` is the name *as credited*, which is not the artist's
+-- canonical name: MusicBrainz credits "Snoop Dogg" on a track by the
+-- artist whose name is "Snoop Doggy Dogg".  It is stored per row rather
+-- than joined from an artist table for exactly that reason.
+--
+-- Only *multi-artist* credits are stored.  A single-artist credit is
+-- (name, "") and is already fully described by explore_index's
+-- artist_name and artist_mbid; storing those would roughly triple the
+-- table to say nothing new.
+--
+-- Credits are shared: an album's twelve tracks by one artist reference
+-- one credit_id.  That is the opposite of the local library's verdict
+-- in plan 013, and correctly so -- credit sharing is 1:1 in one
+-- person's files and genuinely many-to-one across a 2M-row catalog.
+--
+-- MBIDs are the same 16 raw bytes explore_index stores, for the same
+-- size reason and with the same CHECK, so a stringly write fails at the
+-- insert that made it rather than reading back as no rows at all.  See
+-- backend/explore/mbid.go.
+CREATE TABLE IF NOT EXISTS artist_credit_part (
+    credit_id     INTEGER NOT NULL,
+    position      INTEGER NOT NULL,
+    artist_mbid   BLOB NOT NULL CHECK(length(artist_mbid) = 16),
+
+    -- The name as credited on this release, which may differ from the
+    -- artist's canonical name.  Display uses this; navigation uses the
+    -- MBID above.
+    credited_name TEXT NOT NULL,
+
+    -- The literal connector that follows this part -- " feat. ", " & ",
+    -- ", ", or "" on the last part.  Rendered as plain text between two
+    -- links.
+    join_phrase   TEXT NOT NULL DEFAULT '',
+
+    PRIMARY KEY (credit_id, position)
+) WITHOUT ROWID;

backend/database/sql/schemas/artist_credit_ref.sql

@@ -0,0 +1,30 @@
+-- Which credit a catalog entity is credited to.  One row per recording
+-- or release group whose credit names more than one artist.
+--
+-- This is a table rather than an `explore_index.artist_credit_id`
+-- column, and that is a deliberate consequence of how this app applies
+-- its schema.  `applySchema` is CREATE ... IF NOT EXISTS and there is
+-- no migration chain (plan 013), so a *column* added to an existing
+-- table never reaches a database that already has it -- while a new
+-- *table* is created on every install, old or new, for free.
+-- explore_index is the one table nobody can afford to drop and rebuild
+-- on a schema change: it is the artifact users download rather than
+-- derive.
+--
+-- Only multi-artist credits are referenced here, matching
+-- artist_credit_part.  An entity with no row is credited to exactly one
+-- artist, which explore_index's own artist_name and artist_mbid already
+-- describe -- so absence is the common case and means "nothing to
+-- decompose", not "unknown".
+--
+-- `credit_id` is opaque and is only meaningful against the
+-- artist_credit_part rows built or imported alongside it.  The two are
+-- always written together; nothing persists a credit_id anywhere else.
+-- The local library stores resolved parts, never this id.
+CREATE TABLE IF NOT EXISTS artist_credit_ref (
+    mbid      BLOB NOT NULL PRIMARY KEY CHECK(length(mbid) = 16),
+    credit_id INTEGER NOT NULL
+) WITHOUT ROWID;
+
+CREATE INDEX IF NOT EXISTS idx_artist_credit_ref_credit
+    ON artist_credit_ref(credit_id);

backend/database/sql/sqlcgen/models.go

@@ -27,6 +27,19 @@ type Artist struct {
 	Mbid sql.NullString
 }
 
+type ArtistCreditPart struct {
+	CreditID     int64
+	Position     int64
+	ArtistMbid   []byte
+	CreditedName string
+	JoinPhrase   string
+}
+
+type ArtistCreditRef struct {
+	Mbid     []byte
+	CreditID int64
+}
+
 type ArtistEnrichment struct {
 	ArtistMbid string
 	BrowsedAt  sql.NullTime

backend/database/staleshape.go

@@ -0,0 +1,560 @@
+package database
+
+import (
+	"context"
+	"database/sql"
+	"fmt"
+	"io/fs"
+	"log/slog"
+	"maps"
+	"path"
+	"slices"
+	"strings"
+
+	"yellowjacket/backend/datamap"
+)
+
+// This file repairs the one thing `CREATE TABLE IF NOT EXISTS` cannot.
+//
+// `sql/schemas/` is the single description of the schema and there is no
+// migration chain (plan 013): a schema change is one edit to one file.
+// That works perfectly for a *new* table, which every install then
+// creates, and not at all for a changed one -- `IF NOT EXISTS` reaches
+// an existing table only if its shape already matches, and otherwise
+// silently no-ops.  The user's answer to that is "delete and rescan"
+// (plan 013, open question 1), which is free for everything a rescan
+// rebuilds.
+//
+// It is not free for the catalog.  explore_index is a *downloaded
+// artifact*, not something derived from the user's files, and it is the
+// largest thing this app stores.  So it went stale instead: plan 014
+// added `total_tracks` to the schema and to `indexRowFields` -- the one
+// projection every explore read uses -- and no database that already
+// existed ever grew the column.  Every Explore search, browse, artist
+// page and album page on such an install fails with
+// "no such column: total_tracks", while a fresh install is perfectly
+// healthy, which is why the tests did not see it.  The same databases
+// are stale a second way, from the same plan: their `mbid` columns are
+// still TEXT where the schema now declares BLOB, and SQLite does not
+// coerce between the two -- a comparison against 16 raw bytes simply
+// returns no rows.
+//
+// The repair is to notice and drop, not to migrate.  A dropped catalog
+// costs one artifact download (about a minute); the alternative --
+// ALTER TABLE ADD COLUMN, which would handle `total_tracks` alone
+// cheaply -- cannot express the TEXT-to-BLOB half at all, and would
+// leave those installs quietly broken while reporting success.
+//
+// Everything except `Authored` is eligible.  `Cache` is rebuildable by
+// definition; `Owned` is a projection of the user's files and a rescan
+// rebuilds it, which is plan 013's stated answer to exactly this
+// situation ("delete and rescan", open question 1); `Derived` is
+// computed from Owned.  No `Authored` table is ever dropped here --
+// that is the whole point of the datamap, and it is asserted by
+// TestAuthoredTablesAreNeverRetired rather than only stated.
+//
+// What that does *not* buy is immunity for authored rows that reference
+// a retired table.  `audio_files` is MIXED KIND: `play_count`,
+// `last_played` and `tag_status` are authored columns on an Owned
+// table, and they go with it.  Playlists survive as playlists, and
+// their entries survive pointing at nothing.  That cost was weighed and
+// accepted rather than overlooked -- the alternative is to carry the
+// authored columns across the rebuild keyed on file_path, which stays a
+// real option if this ever bites harder than it is worth.
+//
+// **This relies on foreign_keys being ON**, which applyPRAGMAs has
+// already done by the time NewDB calls it, and the dependency is not
+// cosmetic.  SQLite performs an implicit DELETE before dropping a table
+// when foreign keys are enabled, so `playlist_tracks.audio_file_id` --
+// declared ON DELETE SET NULL -- is nulled.  With foreign keys off, no
+// action fires and those rows keep the ids they had, which a rescan
+// then reissues starting from 1: every playlist would silently fill
+// with *different songs*.  Nulled entries are merely empty; stale ones
+// are wrong, and wrong quietly.  TestRetiringOwnedTablesDoesNotDangle
+// is what stops a future reordering turning one into the other.
+
+// retireGroups are tables that must be retired together.  A catalog
+// whose rows are gone must not keep the full-text index built over
+// them, nor the metadata claiming the import that produced them
+// finished -- that marker is exactly what stops the artifact being
+// fetched again.  applySchema recreates all three empty immediately
+// afterwards, and the ordinary "no index yet" path takes over.
+var retireGroups = [][]string{
+	{
+		"explore_index",
+		"explore_index_fts",
+		"explore_index_meta",
+		"explore_champion_fts",
+	},
+}
+
+// schemaColumn is one column as the schema file declares it.
+type schemaColumn struct {
+	name string
+	typ  string
+}
+
+// retireStaleTables drops every non-authored table whose live shape no
+// longer matches what sql/schemas/ declares, plus any table the schema
+// no longer describes at all, so applySchema can create the current
+// shape afresh.  It runs before applySchema and is a no-op on a new
+// database, where the tables do not exist yet.
+func retireStaleTables(
+	ctx context.Context, db *sql.DB, logger *slog.Logger,
+) error {
+	declared, err := declaredTables()
+	if err != nil {
+		return err
+	}
+
+	stale := make(map[string]string)
+
+	for table, columns := range declared {
+		entry, ok := datamap.Lookup(table)
+		if !ok || entry.Kind == datamap.Authored || entry.FTS {
+			continue
+		}
+
+		// Whether a stale Cache table may be rebuilt is decided per
+		// binary, at compile time: the app re-downloads its catalog in
+		// about a minute, cmd/indexbuild would re-derive it from ~205 GB
+		// of dumps.  See staleshape_policy.go.
+		if entry.Kind == datamap.Cache && !retireStaleCache {
+			continue
+		}
+
+		reason, err := staleReason(ctx, db, table, columns)
+		if err != nil {
+			return err
+		}
+
+		if reason != "" {
+			stale[table] = reason
+		}
+	}
+
+	obsolete, err := obsoleteTables(ctx, db)
+	if err != nil {
+		return err
+	}
+
+	maps.Copy(stale, obsolete)
+
+	if len(stale) == 0 {
+		return nil
+	}
+
+	return retireGroupsFor(ctx, db, logger, stale)
+}
+
+// obsoleteTables are live tables the schema no longer describes at all.
+// TestCatalogCoversSchema makes the datamap a complete description of
+// the current schema, so a table it does not know is one a past version
+// created and this one does not -- plan 013 alone left seven behind
+// (recordings, release_groups, artist_credit, artist_credit_artist,
+// release_group_recordings, recording_genres) plus the
+// schema_migrations table that squashing the chain retired.  They are
+// dead weight, and one of them holding a foreign key into a table being
+// rebuilt is worse than dead weight.
+//
+// SQLite's own bookkeeping and FTS shadow tables are not obsolete:
+// datamap.Lookup resolves a shadow table to its parent, and IsInternal
+// covers the rest.
+func obsoleteTables(ctx context.Context, db *sql.DB) (map[string]string, error) {
+	rows, err := db.QueryContext(
+		ctx, "SELECT name FROM sqlite_master WHERE type = 'table'",
+	)
+	if err != nil {
+		return nil, fmt.Errorf("could not list tables: %w", err)
+	}
+
+	defer func() { _ = rows.Close() }()
+
+	out := make(map[string]string)
+
+	for rows.Next() {
+		var name string
+
+		if err := rows.Scan(&name); err != nil {
+			return nil, fmt.Errorf("could not scan table name: %w", err)
+		}
+
+		if datamap.IsInternal(name) {
+			continue
+		}
+
+		if _, known := datamap.Lookup(name); !known {
+			out[name] = "the schema no longer describes this table"
+		}
+	}
+
+	if err := rows.Err(); err != nil {
+		return nil, fmt.Errorf("could not read table list: %w", err)
+	}
+
+	return out, nil
+}
+
+// retireGroupsFor drops each stale table along with everything its
+// retire group says must go with it.
+func retireGroupsFor(
+	ctx context.Context, db *sql.DB, logger *slog.Logger,
+	stale map[string]string,
+) error {
+	drop := make(map[string]string)
+
+	for table, reason := range stale {
+		drop[table] = reason
+
+		for _, group := range retireGroups {
+			if !slices.Contains(group, table) {
+				continue
+			}
+
+			for _, member := range group {
+				if _, already := drop[member]; !already {
+					drop[member] = "retired with " + table
+				}
+			}
+		}
+	}
+
+	return dropDeferred(ctx, db, logger, drop)
+}
+
+// dropDeferred drops every named table in one transaction with foreign
+// key enforcement deferred to the commit.
+//
+// The deferral is required and the two obvious alternatives are both
+// wrong.  These tables reference each other -- pre-013 `audio_files`
+// has a foreign key into `recordings`, which is itself being retired --
+// so dropping them one at a time in an arbitrary order fails with
+// "FOREIGN KEY constraint failed" on whichever is unlucky enough to go
+// first, and there is no order that is safe in general.  Turning
+// foreign keys *off* for the duration would fix that and silently take
+// the ON DELETE SET NULL on `playlist_tracks.audio_file_id` with it,
+// leaving playlist entries pointing at ids a rescan reissues to
+// different songs -- the exact failure
+// TestRetiringOwnedTablesDoesNotDangle exists to prevent.
+//
+// Deferring keeps the actions firing while tolerating the inconsistency
+// in the middle, and the commit then checks that the end state is
+// sound.  It is set inside the transaction because SQLite resets it at
+// every commit.
+func dropDeferred(
+	ctx context.Context, db *sql.DB, logger *slog.Logger,
+	drop map[string]string,
+) error {
+	tx, err := db.BeginTx(ctx, nil)
+	if err != nil {
+		return fmt.Errorf("could not begin the retire transaction: %w", err)
+	}
+
+	defer func() { _ = tx.Rollback() }()
+
+	if _, err := tx.ExecContext(ctx, "PRAGMA defer_foreign_keys = ON"); err != nil {
+		return fmt.Errorf("could not defer foreign keys: %w", err)
+	}
+
+	// Sorted, so a failure is reproducible.  Map order is random, and a
+	// bug that depends on which table happens to go first reproduces on
+	// one run in three and passes review on the other two -- which is
+	// exactly how the foreign-key ordering above reached a real
+	// database.  Sorting does not make any order *safe*; the deferral
+	// does that.
+	for _, table := range slices.Sorted(maps.Keys(drop)) {
+		logger.Warn(
+			"retiring a table the schema no longer describes",
+			"table", table,
+			"reason", drop[table],
+		)
+
+		if _, err := tx.ExecContext(
+			ctx, "DROP TABLE IF EXISTS "+quoteIdent(table),
+		); err != nil {
+			return fmt.Errorf("could not retire stale table %s: %w", table, err)
+		}
+	}
+
+	if err := tx.Commit(); err != nil {
+		return fmt.Errorf("could not commit the retire: %w", err)
+	}
+
+	return nil
+}
+
+// staleReason reports why a live table disagrees with its declaration,
+// or "" when it agrees.  A column the live table does not have is the
+// additive case; a column whose declared type changed is the one an
+// ALTER could not fix anyway.  Columns the live table has and the
+// schema no longer declares are ignored: they cost nothing and dropping
+// the table over one would retire a healthy catalog.
+func staleReason(
+	ctx context.Context, db *sql.DB, table string, columns []schemaColumn,
+) (string, error) {
+	live, err := liveColumns(ctx, db, table)
+	if err != nil {
+		return "", err
+	}
+
+	if len(live) == 0 {
+		// Not present at all: applySchema is about to create it.
+		return "", nil
+	}
+
+	for _, col := range columns {
+		liveType, present := live[col.name]
+		if !present {
+			return "missing column " + col.name, nil
+		}
+
+		if !sameDeclaredType(col.typ, liveType) {
+			return fmt.Sprintf(
+				"column %s is %s, schema declares %s",
+				col.name, liveType, col.typ,
+			), nil
+		}
+	}
+
+	return "", nil
+}
+
+// liveColumns returns the live table's columns and their declared types,
+// empty when the table does not exist.
+func liveColumns(
+	ctx context.Context, db *sql.DB, table string,
+) (map[string]string, error) {
+	rows, err := db.QueryContext(
+		ctx, "SELECT name, type FROM pragma_table_info(?)", table,
+	)
+	if err != nil {
+		return nil, fmt.Errorf("could not inspect table %s: %w", table, err)
+	}
+
+	defer func() { _ = rows.Close() }()
+
+	out := make(map[string]string)
+
+	for rows.Next() {
+		var name, typ string
+
+		if err := rows.Scan(&name, &typ); err != nil {
+			return nil, fmt.Errorf("could not scan column of %s: %w", table, err)
+		}
+
+		out[name] = typ
+	}
+
+	if err := rows.Err(); err != nil {
+		return nil, fmt.Errorf("could not read columns of %s: %w", table, err)
+	}
+
+	return out, nil
+}
+
+// sameDeclaredType compares two SQLite type names.  They are compared
+// case-insensitively and only on the leading word, so INTEGER matches
+// INTEGER and VARCHAR(20) matches VARCHAR -- SQLite's affinity rules
+// make finer distinctions meaningless, and a difference that fine is
+// not worth retiring a catalog over.  An empty declared type matches
+// anything, which is what a column declared with only constraints has.
+func sameDeclaredType(declared, live string) bool {
+	d := strings.ToUpper(strings.Fields(declared + " ")[0])
+	l := strings.ToUpper(strings.Fields(live + " ")[0])
+
+	if d == "" || l == "" {
+		return true
+	}
+
+	if i := strings.IndexByte(d, '('); i >= 0 {
+		d = d[:i]
+	}
+
+	if i := strings.IndexByte(l, '('); i >= 0 {
+		l = l[:i]
+	}
+
+	return d == l
+}
+
+// declaredTables parses every CREATE TABLE in sql/schemas/ into its
+// column list.  Parsing the schema rather than writing the expectation
+// down a second time is the point: a second list is a second thing to
+// forget, which is the fault this whole file exists to repair.
+func declaredTables() (map[string][]schemaColumn, error) {
+	dirEntries, err := schemas.ReadDir("sql/schemas")
+	if err != nil {
+		return nil, fmt.Errorf("could not read schemas directory: %w", err)
+	}
+
+	out := make(map[string][]schemaColumn)
+
+	for _, dirEntry := range dirEntries {
+		if dirEntry.IsDir() {
+			continue
+		}
+
+		content, err := fs.ReadFile(schemas, path.Join("sql/schemas", dirEntry.Name()))
+		if err != nil {
+			return nil, fmt.Errorf("could not read %s: %w", dirEntry.Name(), err)
+		}
+
+		maps.Copy(out, parseCreateTables(string(content)))
+	}
+
+	return out, nil
+}
+
+// constraintKeywords begin a table constraint rather than a column.
+var constraintKeywords = map[string]bool{
+	"PRIMARY": true, "FOREIGN": true, "UNIQUE": true,
+	"CHECK": true, "CONSTRAINT": true,
+}
+
+// parseCreateTables extracts the column names and declared types of
+// every non-virtual CREATE TABLE in one schema file.
+func parseCreateTables(content string) map[string][]schemaColumn {
+	out := make(map[string][]schemaColumn)
+	rest := stripLineComments(content)
+
+	for {
+		idx := indexFold(rest, "CREATE TABLE ")
+		if idx < 0 {
+			return out
+		}
+
+		rest = rest[idx+len("CREATE TABLE "):]
+
+		head, body, ok := splitTableBody(rest)
+		if !ok {
+			return out
+		}
+
+		if name := tableName(head); name != "" {
+			out[name] = parseColumns(body)
+		}
+	}
+}
+
+// tableName pulls the table name out of the text between "CREATE TABLE"
+// and its opening parenthesis, dropping an IF NOT EXISTS and any
+// quoting.
+func tableName(head string) string {
+	head = strings.TrimSpace(head)
+	head = strings.TrimPrefix(head, "IF NOT EXISTS ")
+	head = strings.TrimPrefix(head, "if not exists ")
+
+	fields := strings.Fields(head)
+	if len(fields) == 0 {
+		return ""
+	}
+
+	return strings.Trim(fields[len(fields)-1], `"'`+"`")
+}
+
+// splitTableBody returns the text before the table's opening paren and
+// the balanced text inside it.
+func splitTableBody(s string) (head, body string, ok bool) {
+	open := strings.IndexByte(s, '(')
+	if open < 0 {
+		return "", "", false
+	}
+
+	depth := 0
+
+	for i := open; i < len(s); i++ {
+		switch s[i] {
+		case '(':
+			depth++
+		case ')':
+			depth--
+
+			if depth == 0 {
+				return s[:open], s[open+1 : i], true
+			}
+		}
+	}
+
+	return "", "", false
+}
+
+// parseColumns splits a table body on its top-level commas and keeps
+// the parts that are columns rather than table constraints.
+func parseColumns(body string) []schemaColumn {
+	var (
+		out   []schemaColumn
+		depth int
+		start int
+	)
+
+	parts := make([]string, 0, 8)
+
+	for i := range len(body) {
+		switch body[i] {
+		case '(':
+			depth++
+		case ')':
+			depth--
+		case ',':
+			if depth == 0 {
+				parts = append(parts, body[start:i])
+				start = i + 1
+			}
+		}
+	}
+
+	parts = append(parts, body[start:])
+
+	for _, part := range parts {
+		fields := strings.Fields(part)
+		if len(fields) == 0 {
+			continue
+		}
+
+		// A table constraint need not be followed by a space --
+		// "UNIQUE(mbid)" is one field, and reading it as a column name
+		// makes an entirely healthy table look stale, which retires a
+		// catalog nobody asked to lose.
+		head := fields[0]
+		if i := strings.IndexByte(head, '('); i >= 0 {
+			head = head[:i]
+		}
+
+		if constraintKeywords[strings.ToUpper(head)] {
+			continue
+		}
+
+		col := schemaColumn{name: strings.Trim(head, `"'`+"`")}
+		if len(fields) > 1 {
+			col.typ = fields[1]
+		}
+
+		out = append(out, col)
+	}
+
+	return out
+}
+
+// stripLineComments removes -- comments, which otherwise contribute
+// stray parentheses and commas to the parse.
+func stripLineComments(s string) string {
+	lines := strings.Split(s, "\n")
+	for i, line := range lines {
+		if idx := strings.Index(line, "--"); idx >= 0 {
+			lines[i] = line[:idx]
+		}
+	}
+
+	return strings.Join(lines, "\n")
+}
+
+// indexFold is a case-insensitive strings.Index.
+func indexFold(s, substr string) int {
+	return strings.Index(strings.ToUpper(s), strings.ToUpper(substr))
+}
+
+// quoteIdent quotes a table name for interpolation into DDL, which
+// cannot take a bound parameter.
+func quoteIdent(name string) string {
+	return `"` + strings.ReplaceAll(name, `"`, `""`) + `"`
+}

backend/database/staleshape_policy.go

@@ -0,0 +1,15 @@
+//go:build !indexbuild
+
+package database
+
+// retireStaleCache reports whether a Cache table whose shape no longer
+// matches the schema may be dropped and rebuilt.
+//
+// In the app: yes.  The only Cache table large enough to care about is
+// the catalog, and the app does not derive it — it downloads it.  A
+// stale one costs about a minute of re-fetching the artifact, and
+// keeping it costs every Explore read on the install, because a
+// projection naming a column the table does not have fails outright.
+//
+// In cmd/indexbuild: no, and the file next to this one says why.
+const retireStaleCache = true

backend/database/staleshape_policy_indexbuild.go

@@ -0,0 +1,37 @@
+//go:build indexbuild
+
+package database
+
+// retireStaleCache is false here, and this is the whole reason the
+// policy is a build tag rather than a rule inside retireStaleTables.
+//
+// The index database is the one place in this project where the catalog
+// is *derived* rather than downloaded.  Rebuilding it is a ~205 GB dump
+// stream over hours, resumed across runs from a checkpoint on a
+// persistent volume; that volume exists for no other purpose.  The app's
+// answer to a stale catalog — drop it, fetch the artifact again — is
+// not available here, because this database *is* what the artifact is
+// cut from.
+//
+// This was not hypothetical.  The repair shipped without it and dropped
+// the CI catalog on its first run:
+//
+//	retiring a table ... table=explore_index
+//	  reason="column entity_type is TEXT, schema declares INTEGER"
+//	index maintenance mode=build reason="no completed import yet"
+//
+// The shape mismatch was real and the drop was correct by the app's
+// rule.  It was still wrong here: that database is deliberately kept in
+// the older encoding, which is what `fix(indexexport): read an index
+// older than the binary` exists to tolerate.  A rule that is right for
+// every install and catastrophic for one database has to be told which
+// one it is in, and a build tag is how this project already tells the
+// index tools apart (backend/events/runtime_indexbuild.go,
+// backend/explore/servicestartup.go, dumpbuild_stub.go).
+//
+// cmd/indexbuild has its own repair for the half it *can* safely
+// discard: retireLibraryTables drops every table the datamap does not
+// classify as Cache, which is empty by construction in that database.
+// Between the two, the library half is repaired and the catalog is
+// never touched.
+const retireStaleCache = false

backend/database/staleshape_test.go

@@ -0,0 +1,499 @@
+package database
+
+import (
+	"context"
+	"database/sql"
+	"log/slog"
+	"path"
+	"testing"
+
+	_ "modernc.org/sqlite"
+)
+
+// testLogger discards the repair's warnings; the tests assert on the
+// database, not on the log.
+func testLogger() *slog.Logger {
+	return slog.New(slog.DiscardHandler)
+}
+
+// openRaw opens a scratch database file with no schema applied, so a
+// test can build an *old* shape and then let NewDB's repair meet it.
+func openRaw(t *testing.T, dir string) *sql.DB {
+	t.Helper()
+
+	db, err := sql.Open("sqlite", path.Join(dir, "yj.db"))
+	if err != nil {
+		t.Fatalf("open: %v", err)
+	}
+
+	t.Cleanup(func() { _ = db.Close() })
+
+	return db
+}
+
+// TestRetiresIndexMissingAColumn is plan 014's bug, symptom first: an
+// explore_index created before `total_tracks` existed, met by the
+// projection every explore read uses.  Before the repair this failed
+// with "no such column: total_tracks" on every install that already had
+// a catalog, while a fresh one was perfectly healthy.
+func TestRetiresIndexMissingAColumn(t *testing.T) {
+	ctx := context.Background()
+	dir := t.TempDir()
+	db := openRaw(t, dir)
+
+	// The pre-014 shape: the columns the projection needs, minus the
+	// one the plan added.
+	if _, err := db.ExecContext(ctx, `
+		CREATE TABLE explore_index (
+			id INTEGER PRIMARY KEY AUTOINCREMENT,
+			entity_type INTEGER NOT NULL,
+			mbid BLOB NOT NULL,
+			title TEXT NOT NULL,
+			artist_name TEXT NOT NULL,
+			artist_mbid BLOB NOT NULL
+		);
+		INSERT INTO explore_index (entity_type, mbid, title, artist_name, artist_mbid)
+			VALUES (1, x'00112233445566778899aabbccddeeff', 'x', 'y', x'');
+	`); err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+
+	if err := retireStaleTables(ctx, db, testLogger()); err != nil {
+		t.Fatalf("retire: %v", err)
+	}
+
+	if err := applySchema(ctx, db); err != nil {
+		t.Fatalf("applySchema: %v", err)
+	}
+
+	// The column the projection needs is there now.
+	var n int
+	if err := db.QueryRowContext(ctx,
+		`SELECT COUNT(*) FROM pragma_table_info('explore_index')
+		 WHERE name = 'total_tracks'`,
+	).Scan(&n); err != nil {
+		t.Fatalf("inspect: %v", err)
+	}
+
+	if n != 1 {
+		t.Fatalf("explore_index still has no total_tracks column")
+	}
+
+	// And the catalog really was retired rather than patched, so the
+	// artifact is fetched again instead of half a catalog being served.
+	if err := db.QueryRowContext(ctx,
+		"SELECT COUNT(*) FROM explore_index",
+	).Scan(&n); err != nil {
+		t.Fatalf("count: %v", err)
+	}
+
+	if n != 0 {
+		t.Fatalf("stale rows survived the retire: %d", n)
+	}
+}
+
+// TestRetiresIndexWithTextMBIDs is the half an ALTER could not have
+// repaired: plan 013 changed mbid from TEXT to BLOB, and SQLite does not
+// coerce between them, so a query against 16 raw bytes returns no rows
+// rather than an error.
+func TestRetiresIndexWithTextMBIDs(t *testing.T) {
+	ctx := context.Background()
+	dir := t.TempDir()
+	db := openRaw(t, dir)
+
+	if _, err := db.ExecContext(ctx, `
+		CREATE TABLE explore_index (
+			id INTEGER PRIMARY KEY AUTOINCREMENT,
+			entity_type TEXT NOT NULL,
+			mbid TEXT NOT NULL,
+			title TEXT NOT NULL,
+			artist_name TEXT NOT NULL,
+			artist_mbid TEXT NOT NULL,
+			total_tracks INTEGER NOT NULL DEFAULT 0
+		);
+	`); err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+
+	if err := retireStaleTables(ctx, db, testLogger()); err != nil {
+		t.Fatalf("retire: %v", err)
+	}
+
+	if err := applySchema(ctx, db); err != nil {
+		t.Fatalf("applySchema: %v", err)
+	}
+
+	var typ string
+	if err := db.QueryRowContext(ctx,
+		`SELECT type FROM pragma_table_info('explore_index') WHERE name = 'mbid'`,
+	).Scan(&typ); err != nil {
+		t.Fatalf("inspect: %v", err)
+	}
+
+	if typ != "BLOB" {
+		t.Fatalf("mbid is still %s, want BLOB", typ)
+	}
+}
+
+// TestRetiringTheIndexTakesItsMetaWithIt guards the thing that makes the
+// repair actually repair: the marker saying the import finished is what
+// stops the artifact being fetched again, so a catalog dropped without
+// it would stay empty forever.
+func TestRetiringTheIndexTakesItsMetaWithIt(t *testing.T) {
+	ctx := context.Background()
+	dir := t.TempDir()
+	db := openRaw(t, dir)
+
+	if _, err := db.ExecContext(ctx, `
+		CREATE TABLE explore_index (
+			id INTEGER PRIMARY KEY AUTOINCREMENT,
+			entity_type INTEGER NOT NULL,
+			mbid BLOB NOT NULL
+		);
+		CREATE TABLE explore_index_meta (key TEXT PRIMARY KEY, value TEXT NOT NULL);
+		INSERT INTO explore_index_meta VALUES ('dump_import_done', '1');
+	`); err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+
+	if err := retireStaleTables(ctx, db, testLogger()); err != nil {
+		t.Fatalf("retire: %v", err)
+	}
+
+	if err := applySchema(ctx, db); err != nil {
+		t.Fatalf("applySchema: %v", err)
+	}
+
+	var n int
+	if err := db.QueryRowContext(ctx,
+		"SELECT COUNT(*) FROM explore_index_meta WHERE key = 'dump_import_done'",
+	).Scan(&n); err != nil {
+		t.Fatalf("meta: %v", err)
+	}
+
+	if n != 0 {
+		t.Fatalf("the import-done marker survived a retired catalog")
+	}
+}
+
+// TestHealthyDatabaseIsUntouched is the other half, and the one that
+// would make this dangerous if it failed: a current schema must survive
+// a launch with its catalog intact.  A repair that retires a healthy
+// catalog costs every user an artifact download on every start.
+func TestHealthyDatabaseIsUntouched(t *testing.T) {
+	ctx := context.Background()
+	dir := t.TempDir()
+	db := openRaw(t, dir)
+
+	if err := applySchema(ctx, db); err != nil {
+		t.Fatalf("applySchema: %v", err)
+	}
+
+	if _, err := db.ExecContext(ctx, `
+		INSERT INTO explore_index (entity_type, mbid, title, artist_name, artist_mbid)
+			VALUES (1, x'00112233445566778899aabbccddeeff', 'x', 'y', x'')
+	`); err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+
+	if err := retireStaleTables(ctx, db, testLogger()); err != nil {
+		t.Fatalf("retire: %v", err)
+	}
+
+	var n int
+	if err := db.QueryRowContext(ctx,
+		"SELECT COUNT(*) FROM explore_index",
+	).Scan(&n); err != nil {
+		t.Fatalf("count: %v", err)
+	}
+
+	if n != 1 {
+		t.Fatalf("a healthy catalog was retired: %d rows left", n)
+	}
+}
+
+// TestAuthoredTablesAreNeverRetired states the boundary in a test rather
+// than only in a comment: this mechanism deletes data, and the only
+// thing standing between it and a user's playlists is the Kind filter.
+func TestAuthoredTablesAreNeverRetired(t *testing.T) {
+	ctx := context.Background()
+	dir := t.TempDir()
+	db := openRaw(t, dir)
+
+	// A playlists table missing most of its current columns.
+	if _, err := db.ExecContext(ctx, `
+		CREATE TABLE playlists (id INTEGER PRIMARY KEY, name TEXT NOT NULL);
+		INSERT INTO playlists (name) VALUES ('irreplaceable');
+	`); err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+
+	if err := retireStaleTables(ctx, db, testLogger()); err != nil {
+		t.Fatalf("retire: %v", err)
+	}
+
+	var n int
+	if err := db.QueryRowContext(ctx,
+		"SELECT COUNT(*) FROM playlists",
+	).Scan(&n); err != nil {
+		t.Fatalf("count: %v", err)
+	}
+
+	if n != 1 {
+		t.Fatalf("an authored table was retired; rows left: %d", n)
+	}
+}
+
+// TestRetiresTablesTheSchemaNoLongerDescribes covers what plan 013 left
+// behind on every database that predates it: seven tables the schema
+// stopped describing, plus the schema_migrations table that squashing
+// the chain retired.  They are not stale in shape — they are simply not
+// ours any more.
+func TestRetiresTablesTheSchemaNoLongerDescribes(t *testing.T) {
+	ctx := context.Background()
+	db := openRaw(t, t.TempDir())
+
+	if _, err := db.ExecContext(ctx, `
+		CREATE TABLE recordings (id INTEGER PRIMARY KEY, name TEXT);
+		CREATE TABLE artist_credit (id INTEGER PRIMARY KEY, text TEXT);
+		CREATE TABLE schema_migrations (version INTEGER PRIMARY KEY);
+	`); err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+
+	if err := retireStaleTables(ctx, db, testLogger()); err != nil {
+		t.Fatalf("retire: %v", err)
+	}
+
+	for _, table := range []string{"recordings", "artist_credit", "schema_migrations"} {
+		var n int
+
+		if err := db.QueryRowContext(ctx,
+			"SELECT COUNT(*) FROM sqlite_master WHERE type='table' AND name = ?",
+			table,
+		).Scan(&n); err != nil {
+			t.Fatalf("inspect %s: %v", table, err)
+		}
+
+		if n != 0 {
+			t.Errorf("%s survived; the schema no longer describes it", table)
+		}
+	}
+}
+
+// TestFTSShadowTablesAreNotObsolete is the sweep's sharp edge: an FTS5
+// virtual table is backed by four shadow tables that appear in
+// sqlite_master under their own names and are in no schema file.
+// Dropping one destroys the index it belongs to.
+func TestFTSShadowTablesAreNotObsolete(t *testing.T) {
+	ctx := context.Background()
+	db := openRaw(t, t.TempDir())
+
+	if err := applySchema(ctx, db); err != nil {
+		t.Fatalf("applySchema: %v", err)
+	}
+
+	obsolete, err := obsoleteTables(ctx, db)
+	if err != nil {
+		t.Fatalf("obsoleteTables: %v", err)
+	}
+
+	if len(obsolete) != 0 {
+		t.Fatalf("a freshly created schema reported obsolete tables: %v", obsolete)
+	}
+}
+
+// TestRetiringOwnedTablesDoesNotDangle pins the one behaviour that is
+// silently wrong rather than loudly broken.
+//
+// Retiring audio_files leaves playlist entries behind.  With
+// foreign_keys ON — which applyPRAGMAs has done before NewDB gets here —
+// SET NULL fires and they point at nothing.  With it OFF they keep ids
+// that the rescan reissues from 1, so every playlist quietly fills with
+// different songs.  Nothing about the schema makes that ordering
+// obvious, so it is asserted rather than assumed.
+func TestRetiringOwnedTablesDoesNotDangle(t *testing.T) {
+	ctx := context.Background()
+	db := openRaw(t, t.TempDir())
+
+	if _, err := db.ExecContext(ctx, "PRAGMA foreign_keys = ON"); err != nil {
+		t.Fatalf("pragma: %v", err)
+	}
+
+	if err := applySchema(ctx, db); err != nil {
+		t.Fatalf("applySchema: %v", err)
+	}
+
+	// Break audio_files' shape so it is retired, keeping a playlist
+	// entry that references it.
+	if _, err := db.ExecContext(ctx, `
+		INSERT INTO playlists (id, name) VALUES (1, 'keepme');
+		INSERT INTO libraries (id, name, path) VALUES (0, 'test', '/music');
+		INSERT INTO audio_files (id, file_path, file_type_id, length_milliseconds)
+			VALUES (7, '/music/a.flac', 1, 1000);
+		INSERT INTO playlist_tracks (playlist_id, audio_file_id, position)
+			VALUES (1, 7, 0);
+		DROP VIEW IF EXISTS track_metadata;
+		ALTER TABLE audio_files DROP COLUMN artist_credit;
+	`); err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+
+	if err := retireStaleTables(ctx, db, testLogger()); err != nil {
+		t.Fatalf("retire: %v", err)
+	}
+
+	if err := applySchema(ctx, db); err != nil {
+		t.Fatalf("applySchema: %v", err)
+	}
+
+	var dangling int
+	if err := db.QueryRowContext(ctx,
+		"SELECT COUNT(*) FROM playlist_tracks WHERE audio_file_id IS NOT NULL",
+	).Scan(&dangling); err != nil {
+		t.Fatalf("count: %v", err)
+	}
+
+	if dangling != 0 {
+		t.Fatalf(
+			"%d playlist entries still point at retired audio_files ids; "+
+				"a rescan will reissue those ids to different tracks",
+			dangling,
+		)
+	}
+
+	// The playlist itself is authored and must be untouched.
+	var playlists int
+	if err := db.QueryRowContext(ctx,
+		"SELECT COUNT(*) FROM playlists",
+	).Scan(&playlists); err != nil {
+		t.Fatalf("playlists: %v", err)
+	}
+
+	if playlists != 1 {
+		t.Fatalf("authored playlist lost: %d", playlists)
+	}
+}
+
+// TestRetiringInterlinkedLegacyTables is the bug the unit tests missed
+// and a real database found.
+//
+// The tables plan 013 retired reference each other -- pre-013
+// audio_files has a foreign key into recordings -- so with foreign keys
+// ON, dropping them one at a time fails with "FOREIGN KEY constraint
+// failed" on whichever goes first, and map iteration order decides
+// which that is.  Every other test in this file ran with foreign keys
+// off and passed happily; the app enables them in applyPRAGMAs before
+// the repair runs, so only the real launch path showed it.
+func TestRetiringInterlinkedLegacyTables(t *testing.T) {
+	ctx := context.Background()
+	db := openRaw(t, t.TempDir())
+
+	if _, err := db.ExecContext(ctx, "PRAGMA foreign_keys = ON"); err != nil {
+		t.Fatalf("pragma: %v", err)
+	}
+
+	// The pre-013 shape, with the reference that makes ordering matter.
+	// release_group_recordings sorts *after* recordings and references
+	// it, so the deterministic order retires the parent while the child
+	// still holds rows pointing at it -- which is the case that fails
+	// without the deferral, rather than one that fails on some runs.
+	if _, err := db.ExecContext(ctx, `
+		CREATE TABLE recordings (id INTEGER PRIMARY KEY, name TEXT);
+		CREATE TABLE artist_credit (id INTEGER PRIMARY KEY, text TEXT);
+		CREATE TABLE release_group_recordings (
+			id           INTEGER PRIMARY KEY,
+			recording_id INTEGER NOT NULL,
+			FOREIGN KEY(recording_id) REFERENCES recordings(id)
+		);
+		CREATE TABLE audio_files (
+			id           INTEGER PRIMARY KEY,
+			file_path    TEXT NOT NULL UNIQUE,
+			recording_id INTEGER,
+			FOREIGN KEY(recording_id) REFERENCES recordings(id)
+		);
+		INSERT INTO recordings (id, name) VALUES (1, 'x');
+		INSERT INTO release_group_recordings (id, recording_id) VALUES (1, 1);
+		INSERT INTO audio_files (id, file_path, recording_id)
+			VALUES (1, '/music/a.flac', 1);
+	`); err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+
+	if err := retireStaleTables(ctx, db, testLogger()); err != nil {
+		t.Fatalf("retire: %v", err)
+	}
+
+	if err := applySchema(ctx, db); err != nil {
+		t.Fatalf("applySchema: %v", err)
+	}
+
+	for _, table := range []string{"recordings", "artist_credit"} {
+		var n int
+
+		if err := db.QueryRowContext(ctx,
+			"SELECT COUNT(*) FROM sqlite_master WHERE type='table' AND name = ?",
+			table,
+		).Scan(&n); err != nil {
+			t.Fatalf("inspect %s: %v", table, err)
+		}
+
+		if n != 0 {
+			t.Errorf("%s survived the retire", table)
+		}
+	}
+
+	// And the rebuilt audio_files is the current shape, which is the
+	// whole reason the old one had to go.
+	var n int
+	if err := db.QueryRowContext(ctx,
+		`SELECT COUNT(*) FROM pragma_table_info('audio_files')
+		 WHERE name = 'artist_credit'`,
+	).Scan(&n); err != nil {
+		t.Fatalf("inspect audio_files: %v", err)
+	}
+
+	if n != 1 {
+		t.Fatal("audio_files was not rebuilt in the current shape")
+	}
+}
+
+// TestParseCreateTablesReadsTheRealSchema keeps the parser honest
+// against the files it actually runs on: a parser that silently found
+// no columns would report every table healthy and repair nothing.
+func TestParseCreateTablesReadsTheRealSchema(t *testing.T) {
+	declared, err := declaredTables()
+	if err != nil {
+		t.Fatalf("declaredTables: %v", err)
+	}
+
+	cols, ok := declared["explore_index"]
+	if !ok {
+		t.Fatal("explore_index was not parsed out of the schema files")
+	}
+
+	want := map[string]string{
+		"mbid":         "BLOB",
+		"total_tracks": "INTEGER",
+		"artist_name":  "TEXT",
+	}
+
+	got := make(map[string]string, len(cols))
+	for _, c := range cols {
+		got[c.name] = c.typ
+	}
+
+	for name, typ := range want {
+		if got[name] != typ {
+			t.Errorf("explore_index.%s parsed as %q, want %q", name, got[name], typ)
+		}
+	}
+
+	// A table constraint must not be mistaken for a column.
+	for _, c := range cols {
+		switch c.name {
+		case "PRIMARY", "FOREIGN", "UNIQUE", "CHECK", "CONSTRAINT":
+			t.Errorf("parsed table constraint %q as a column", c.name)
+		}
+	}
+}

backend/datamap/datamap.go

@@ -185,6 +185,21 @@ var tables = []Table{
 		Note: "Full-text index over the champion entities of the " +
 			"MusicBrainz dump. Rebuilt only by a full index build.",
 	},
+	{
+		Name: "artist_credit_part", Kind: Cache, Lifetime: Retained,
+		Note: "The decomposition of a multi-artist credit, from the " +
+			"MusicBrainz dump: one row per credited artist, with the " +
+			"name as credited and the join phrase that follows it. " +
+			"Arrives with the downloaded artifact, so rebuilding it " +
+			"costs a dump stream and it is never swept.",
+	},
+	{
+		Name: "artist_credit_ref", Kind: Cache, Lifetime: Retained,
+		Note: "Which credit a catalog recording or release group is " +
+			"credited to. Present only for multi-artist credits; " +
+			"absence means one artist, which explore_index already " +
+			"describes. Ships and dies with artist_credit_part.",
+	},
 	{
 		Name: "explore_index", Kind: Cache, Lifetime: Retained,
 		Note: "The offline MusicBrainz search index. Rebuilding costs a " +

backend/download/manager_test.go

@@ -199,24 +199,23 @@ func TestManagerEndToEndAutoPick(t *testing.T) {
 		t.Errorf("expected imported file at %s: %v", want, err)
 	}
 
-	// Staging was released only after a successful import.
-	entries, err := os.ReadDir(f.staging.Root())
-	if err != nil {
-		t.Fatalf("read staging root: %v", err)
-	}
+	// Staging release and the rescan happen *after* the state is
+	// recorded (manager.go sets StateComplete, then releases, then
+	// scans), so waiting on the state is not waiting on these.  Under
+	// load the worker is descheduled in between and asserting straight
+	// away reads the world one step too early -- which is exactly how
+	// this test failed on a busy machine while passing alone.
+	waitFor(t, func() bool {
+		entries, err := os.ReadDir(f.staging.Root())
+		if err != nil || len(entries) != 0 {
+			return false
+		}
 
-	if len(entries) != 0 {
-		t.Errorf("staging not released: %d dirs remain", len(entries))
-	}
+		f.lib.mu.Lock()
+		defer f.lib.mu.Unlock()
 
-	// The library was told to rescan.
-	f.lib.mu.Lock()
-	scanned := len(f.lib.scanned)
-	f.lib.mu.Unlock()
-
-	if scanned != 1 {
-		t.Errorf("library scans = %d, want 1", scanned)
-	}
+		return len(f.lib.scanned) == 1
+	}, "staging was never released, or the library was never rescanned")
 }
 
 // An ambiguous result set must park for the user rather than guess.

backend/download/service_test.go

@@ -20,6 +20,30 @@ func newServiceFixture(t *testing.T) serviceFixture {
 	mf := newManagerFixture(t)
 	svc := NewService(slogDiscard(), mf.manager, mf.store, NewMemSecretStore())
 
+	// Every test here is about the durable Request that `StartDownload`
+	// leaves behind, and none of them is about the download itself -- but
+	// their fixture is an anchored four-track request with a healthy
+	// provider, which is exactly what `AutoPickable` says yes to. So
+	// `Manager.Start` was firing `go m.grab(...)`, detached and with
+	// `context.WithoutCancel`, and the test then raced it.
+	//
+	// It lost, twice, in CI (`check` on c03c0b8, and nowhere locally):
+	//
+	//	service_test.go:66: state = "satisfied", want wanted
+	//	testing.go:1369: TempDir RemoveAll cleanup: ... directory not empty
+	//
+	// The first is the request reaching its *next* state before the
+	// assertion read it; the second is that same goroutine still writing
+	// into `t.TempDir()` after the test returned. One cause, two shapes.
+	//
+	// Putting the candidate outside the auto-pick size window stops the
+	// grab from ever starting, which is better than waiting for it: there
+	// is no goroutine to be slow, so the tests state what they mean
+	// ("the request exists, in this state") without a timing assumption
+	// underneath. A test that does want the download has `managerFixture`
+	// and sets its own preferences.
+	mf.manager.SetPreferences(AutoDownloadPrefs{MaxSizeMB: 1})
+
 	return serviceFixture{managerFixture: mf, svc: svc}
 }
 

backend/explore/artifactimport.go

@@ -283,6 +283,9 @@ func (si *SearchIndex) importCoreArtifact(ctx context.Context, path string) erro
 	}
 
 	merged, mergeErr := si.mergeArtifactRows(ctx, info.rows)
+	if mergeErr == nil {
+		si.mergeArtifactCredits(ctx)
+	}
 
 	if ftsSuspended {
 		start := time.Now()
@@ -473,3 +476,75 @@ func (si *SearchIndex) removeArtifactFile(path string) {
 		si.logger.Warn("core artifact: cleanup failed", "path", path, "error", err)
 	}
 }
+
+// artifactHasCredits reports whether the attached artifact carries the
+// multi-artist credit tables.
+//
+// The same shape, and the same handle, as artifactHasTotals above: an
+// artifact published before credits existed is still a perfectly good
+// catalog, and there is one already out there.  Selecting from a table
+// that is not in it would fail an import that should have succeeded, so
+// it is asked rather than assumed -- on the *writer*, because `core` is
+// attached to that one connection and the read pool cannot see it.
+func (si *SearchIndex) artifactHasCredits() bool {
+	var n int
+
+	err := si.db.QueryRowWriter(
+		`SELECT COUNT(*) FROM core.sqlite_master
+		 WHERE type = 'table' AND name IN ('artist_credit_part', 'artist_credit_ref')`,
+	).Scan(&n)
+
+	return err == nil && n == 2
+}
+
+// mergeArtifactCredits copies the credit decomposition out of the
+// attached artifact.
+//
+// Credits are replaced wholesale rather than merged: they are derived
+// entirely from one dump build, they are keyed by ids that are only
+// meaningful within the artifact that carried them, and a half-updated
+// credit renders as the wrong artists rather than as missing ones.
+//
+// A failure here is logged and not returned.  The catalog has already
+// merged at this point, and a catalog without credits is the catalog
+// this app had before them -- every credit falls back to its single
+// artist, which is the same fallback an untagged file already gets.
+func (si *SearchIndex) mergeArtifactCredits(ctx context.Context) {
+	if !si.artifactHasCredits() {
+		si.logger.Info("core artifact: no credit tables, keeping single-artist credits")
+
+		return
+	}
+
+	start := time.Now()
+
+	for _, stmt := range []string{
+		"DELETE FROM artist_credit_part",
+		"DELETE FROM artist_credit_ref",
+		`INSERT OR REPLACE INTO artist_credit_part
+			(credit_id, position, artist_mbid, credited_name, join_phrase)
+		 SELECT credit_id, position, artist_mbid, credited_name, join_phrase
+		 FROM core.artist_credit_part`,
+		`INSERT OR REPLACE INTO artist_credit_ref (mbid, credit_id)
+		 SELECT mbid, credit_id FROM core.artist_credit_ref`,
+	} {
+		if err := ctx.Err(); err != nil {
+			return
+		}
+
+		if _, err := si.db.ExecContext(stmt); err != nil {
+			si.logger.Warn("core artifact: credit merge failed", "error", err)
+
+			return
+		}
+	}
+
+	var refs int
+
+	_ = si.db.QueryRowWriter("SELECT COUNT(*) FROM artist_credit_ref").Scan(&refs)
+
+	si.logger.Info("core artifact: credits merged",
+		"entities", refs,
+		"elapsed", time.Since(start).Round(time.Millisecond),
+	)
+}

backend/explore/artifactimport_test.go

@@ -3,6 +3,7 @@ package explore
 import (
 	"context"
 	"database/sql"
+	"encoding/hex"
 	"os"
 	"path/filepath"
 	"strings"
@@ -597,3 +598,153 @@ func TestImportCoreArtifactReadsTotalsWhenPresent(t *testing.T) {
 		t.Errorf("TotalTracks = %d, want 0 (the catalog does not say)", old.TotalTracks)
 	}
 }
+
+// addArtifactCredits gives an artifact file the credit tables the
+// exporter now writes, so the import path can be exercised against one
+// that has them.
+func addArtifactCredits(t *testing.T, path string) {
+	t.Helper()
+
+	db, err := sql.Open("sqlite", "file:"+path)
+	if err != nil {
+		t.Fatalf("open artifact: %v", err)
+	}
+
+	defer func() { _ = db.Close() }()
+
+	for _, stmt := range []string{
+		`CREATE TABLE artist_credit_part (
+			credit_id     INTEGER NOT NULL,
+			position      INTEGER NOT NULL,
+			artist_mbid   BLOB NOT NULL,
+			credited_name TEXT NOT NULL,
+			join_phrase   TEXT NOT NULL DEFAULT '',
+			PRIMARY KEY (credit_id, position)
+		) WITHOUT ROWID`,
+		`CREATE TABLE artist_credit_ref (
+			mbid      BLOB NOT NULL PRIMARY KEY,
+			credit_id INTEGER NOT NULL
+		) WITHOUT ROWID`,
+	} {
+		if _, err := db.Exec(stmt); err != nil {
+			t.Fatalf("create credit tables: %v", err)
+		}
+	}
+
+	// The packed form the catalog stores.  uuid16/parseUUID live behind
+	// the indexbuild tag, so this file decodes for itself.
+	pack := func(mbid string) []byte {
+		raw, err := hex.DecodeString(strings.ReplaceAll(mbid, "-", ""))
+		if err != nil || len(raw) != 16 {
+			t.Fatalf("fixture MBID %q is not a UUID: %v", mbid, err)
+		}
+
+		return raw
+	}
+
+	a, b, rec := pack(artA), pack(artB), pack(recA)
+
+	for _, part := range [][]any{
+		{7, 0, a, "Artist A", " feat. "},
+		{7, 1, b, "Artist B", ""},
+	} {
+		if _, err := db.Exec(`INSERT INTO artist_credit_part
+			(credit_id, position, artist_mbid, credited_name, join_phrase)
+			VALUES (?, ?, ?, ?, ?)`, part...); err != nil {
+			t.Fatalf("insert part: %v", err)
+		}
+	}
+
+	if _, err := db.Exec(
+		"INSERT INTO artist_credit_ref (mbid, credit_id) VALUES (?, ?)", rec, 7,
+	); err != nil {
+		t.Fatalf("insert ref: %v", err)
+	}
+}
+
+// TestImportCoreArtifactMergesCredits is the positive half of the
+// compatibility pair: an artifact that carries credits delivers them,
+// rendering back to the credit string they decompose.
+func TestImportCoreArtifactMergesCredits(t *testing.T) {
+	db := database.NewTestDB(t)
+	si := NewSearchIndex(db, nil, nil, testLogger())
+
+	path := writeTestArtifact(t, validMeta(), []artifactRow{
+		{"recording", recA, "Song A", "Artist A feat. Artist B", artA, 2000},
+	})
+
+	addArtifactCredits(t, path)
+
+	if err := si.importCoreArtifact(context.Background(), path); err != nil {
+		t.Fatalf("importCoreArtifact: %v", err)
+	}
+
+	rows, err := db.QueryContext(
+		`SELECT p.credited_name, p.join_phrase
+		 FROM artist_credit_ref r
+		 JOIN artist_credit_part p ON p.credit_id = r.credit_id
+		 ORDER BY p.position`,
+	)
+	if err != nil {
+		t.Fatalf("query credits: %v", err)
+	}
+
+	defer func() { _ = rows.Close() }()
+
+	var rendered strings.Builder
+
+	for rows.Next() {
+		var name, join string
+
+		if err := rows.Scan(&name, &join); err != nil {
+			t.Fatalf("scan: %v", err)
+		}
+
+		rendered.WriteString(name)
+		rendered.WriteString(join)
+	}
+
+	if got := rendered.String(); got != "Artist A feat. Artist B" {
+		t.Errorf("rendered credit = %q, want %q", got, "Artist A feat. Artist B")
+	}
+}
+
+// TestImportCoreArtifactWithoutCredits is the regression that matters
+// most here: an artifact published before credits existed cannot be
+// re-cut retroactively, so it must import as a catalog that declines to
+// answer rather than failing outright.  writeTestArtifact deliberately
+// builds one without the tables.
+func TestImportCoreArtifactWithoutCredits(t *testing.T) {
+	db := database.NewTestDB(t)
+	si := NewSearchIndex(db, nil, nil, testLogger())
+
+	path := writeTestArtifact(t, validMeta(), []artifactRow{
+		{"recording", recA, "Song A", "Artist A", artA, 2000},
+	})
+
+	if err := si.importCoreArtifact(context.Background(), path); err != nil {
+		t.Fatalf("an artifact without credit tables must still import: %v", err)
+	}
+
+	var rows int
+	if err := db.QueryRowWriter(
+		"SELECT COUNT(*) FROM explore_index",
+	).Scan(&rows); err != nil {
+		t.Fatalf("count: %v", err)
+	}
+
+	if rows != 1 {
+		t.Errorf("catalog rows = %d, want 1", rows)
+	}
+
+	var refs int
+	if err := db.QueryRowWriter(
+		"SELECT COUNT(*) FROM artist_credit_ref",
+	).Scan(&refs); err != nil {
+		t.Fatalf("count refs: %v", err)
+	}
+
+	if refs != 0 {
+		t.Errorf("credit refs = %d, want 0", refs)
+	}
+}

backend/explore/credits.go

@@ -0,0 +1,132 @@
+package explore
+
+import (
+	"fmt"
+	"strings"
+)
+
+// Reading multi-artist credits back out of the catalog.
+//
+// The tables are filled centrally (backend/explore/dumpcredits.go, and
+// the artifact import) and hold only credits naming more than one
+// artist: an entity with no rows here is credited to one artist, which
+// explore_index's own artist_name and artist_mbid already describe.
+// Absence is the common case and means "nothing to decompose", never
+// "unknown".
+//
+// The lookup is keyed on the *recording* MBID, which both sides of the
+// app already have -- a catalog row carries it and so does a local
+// file (library.Track.RecordingMBID) -- so one query serves the Explore
+// pages and the library's own lists without either needing to know
+// where the other gets its rows.
+
+// CreditPart is one credited artist within a credit, in credit order.
+//
+// CreditedName is the name *as credited*, which is not the artist's own
+// name: MusicBrainz credits "Snoop Dogg" on a track by the artist
+// called "Snoop Doggy Dogg".  Display uses it; navigation uses
+// ArtistMBID.  JoinPhrase is the literal connector that follows this
+// part, so a credit renders by concatenation and never by searching a
+// name inside a credit string.
+type CreditPart struct {
+	Position     int    `json:"position"`
+	ArtistMBID   string `json:"artistMbid"`
+	CreditedName string `json:"creditedName"`
+	JoinPhrase   string `json:"joinPhrase"`
+}
+
+// creditLookupBatch bounds how many MBIDs go into one IN clause.  A
+// tracklist is the caller here, so the realistic ceiling is a few
+// hundred; the bound exists so a 50,000-row selection cannot build a
+// statement SQLite refuses to parse.
+const creditLookupBatch = 500
+
+// GetCredits returns the decomposition of every multi-artist credit
+// among the given entity MBIDs, keyed by MBID.
+//
+// MBIDs with a single-artist credit are simply absent from the result,
+// which is what the caller wants: it renders its existing single link
+// for those, and that is the same answer it would have rendered anyway.
+func (si *SearchIndex) GetCredits(mbids []string) (map[string][]CreditPart, error) {
+	out := make(map[string][]CreditPart)
+
+	for start := 0; start < len(mbids); start += creditLookupBatch {
+		end := min(start+creditLookupBatch, len(mbids))
+
+		if err := si.appendCredits(mbids[start:end], out); err != nil {
+			return nil, err
+		}
+	}
+
+	return out, nil
+}
+
+// appendCredits runs one batch into the accumulating result.
+func (si *SearchIndex) appendCredits(
+	mbids []string, out map[string][]CreditPart,
+) error {
+	args := make([]any, 0, len(mbids))
+	holders := make([]string, 0, len(mbids))
+
+	for _, mbid := range mbids {
+		if mbid == "" {
+			continue
+		}
+
+		args = append(args, dbMBID(mbid))
+		holders = append(holders, "?")
+	}
+
+	if len(args) == 0 {
+		return nil
+	}
+
+	// Ordered by position because that ordering *is* the credit's
+	// meaning; the caller concatenates in the order it receives.
+	rows, err := si.db.QueryContext(
+		`SELECT r.mbid, p.position, p.artist_mbid, p.credited_name, p.join_phrase
+		 FROM artist_credit_ref r
+		 JOIN artist_credit_part p ON p.credit_id = r.credit_id
+		 WHERE r.mbid IN (`+strings.Join(holders, ",")+`)
+		 ORDER BY r.mbid, p.position`,
+		args...,
+	)
+	if err != nil {
+		return fmt.Errorf("read artist credits: %w", err)
+	}
+
+	defer func() { _ = rows.Close() }()
+
+	for rows.Next() {
+		var (
+			entity dbMBID
+			artist dbMBID
+			part   CreditPart
+		)
+
+		if err := rows.Scan(
+			&entity, &part.Position, &artist, &part.CreditedName, &part.JoinPhrase,
+		); err != nil {
+			return fmt.Errorf("scan artist credit: %w", err)
+		}
+
+		part.ArtistMBID = string(artist)
+		out[string(entity)] = append(out[string(entity)], part)
+	}
+
+	if err := rows.Err(); err != nil {
+		return fmt.Errorf("read artist credits: %w", err)
+	}
+
+	return nil
+}
+
+// GetCredits is the bound form: the frontend asks for a tracklist's
+// worth of MBIDs at once rather than one per row.
+//
+// Batched for the reason every other per-row backend question here is:
+// asking on hover or on render turns a list into N IPC round trips, and
+// this one is asked about every row of every list in the app.
+func (e *Service) GetCredits(mbids []string) (map[string][]CreditPart, error) {
+	return e.index.GetCredits(mbids)
+}

backend/explore/credits_test.go

@@ -0,0 +1,128 @@
+package explore
+
+import (
+	"encoding/hex"
+	"fmt"
+	"strings"
+	"testing"
+
+	"yellowjacket/backend/database"
+)
+
+// seedCredit writes one multi-artist credit and points an entity at it,
+// the way the dump import and the artifact import both do.
+func seedCredit(t *testing.T, db *database.DB, entity string, id int, parts []CreditPart) {
+	t.Helper()
+
+	pack := func(mbid string) []byte {
+		raw, err := hex.DecodeString(strings.ReplaceAll(mbid, "-", ""))
+		if err != nil || len(raw) != 16 {
+			t.Fatalf("bad fixture mbid %q: %v", mbid, err)
+		}
+
+		return raw
+	}
+
+	if _, err := db.ExecContext(
+		"INSERT INTO artist_credit_ref (mbid, credit_id) VALUES (?, ?)",
+		pack(entity), id,
+	); err != nil {
+		t.Fatalf("seed ref: %v", err)
+	}
+
+	for _, p := range parts {
+		if _, err := db.ExecContext(
+			`INSERT INTO artist_credit_part
+				(credit_id, position, artist_mbid, credited_name, join_phrase)
+			 VALUES (?, ?, ?, ?, ?)`,
+			id, p.Position, pack(p.ArtistMBID), p.CreditedName, p.JoinPhrase,
+		); err != nil {
+			t.Fatalf("seed part: %v", err)
+		}
+	}
+}
+
+// TestGetCreditsDecomposes: the parts come back in position order and
+// concatenate to the credit they describe.
+func TestGetCreditsDecomposes(t *testing.T) {
+	db := database.NewTestDB(t)
+	si := NewSearchIndex(db, nil, nil, testLogger())
+
+	rec := testMBID("rec-1")
+	a, b := testMBID("artist-a"), testMBID("artist-b")
+
+	seedCredit(t, db, rec, 7, []CreditPart{
+		{Position: 0, ArtistMBID: a, CreditedName: "2Pac", JoinPhrase: " feat. "},
+		{Position: 1, ArtistMBID: b, CreditedName: "Snoop Dogg"},
+	})
+
+	got, err := si.GetCredits([]string{rec})
+	if err != nil {
+		t.Fatalf("GetCredits: %v", err)
+	}
+
+	parts := got[rec]
+	if len(parts) != 2 {
+		t.Fatalf("parts = %d, want 2", len(parts))
+	}
+
+	var rendered strings.Builder
+	for _, p := range parts {
+		rendered.WriteString(p.CreditedName)
+		rendered.WriteString(p.JoinPhrase)
+	}
+
+	if rendered.String() != "2Pac feat. Snoop Dogg" {
+		t.Errorf("rendered = %q, want %q", rendered.String(), "2Pac feat. Snoop Dogg")
+	}
+
+	// Dashed on the way out: a blob reaching the frontend is sixteen
+	// bytes of mojibake, and nothing above mbid.go speaks that.
+	if parts[0].ArtistMBID != a {
+		t.Errorf("artist mbid = %q, want %q", parts[0].ArtistMBID, a)
+	}
+}
+
+// TestGetCreditsOmitsSingleArtist: absence is the common case and means
+// "nothing to decompose", so the caller renders its existing one link.
+func TestGetCreditsOmitsSingleArtist(t *testing.T) {
+	db := database.NewTestDB(t)
+	si := NewSearchIndex(db, nil, nil, testLogger())
+
+	got, err := si.GetCredits([]string{testMBID("untagged"), ""})
+	if err != nil {
+		t.Fatalf("GetCredits: %v", err)
+	}
+
+	if len(got) != 0 {
+		t.Errorf("got %d credits, want none", len(got))
+	}
+}
+
+// TestGetCreditsBatches: the lookup is asked about whole tracklists, so
+// it must not build one statement per row or one SQLite refuses to
+// parse.
+func TestGetCreditsBatches(t *testing.T) {
+	db := database.NewTestDB(t)
+	si := NewSearchIndex(db, nil, nil, testLogger())
+
+	mbids := make([]string, 0, creditLookupBatch*2+7)
+	for i := range creditLookupBatch*2 + 7 {
+		mbids = append(mbids, testMBID(fmt.Sprintf("batch-%d", i)))
+	}
+
+	// One real credit somewhere past the first batch boundary.
+	seedCredit(t, db, mbids[creditLookupBatch+3], 9, []CreditPart{
+		{Position: 0, ArtistMBID: testMBID("a"), CreditedName: "A", JoinPhrase: " & "},
+		{Position: 1, ArtistMBID: testMBID("b"), CreditedName: "B"},
+	})
+
+	got, err := si.GetCredits(mbids)
+	if err != nil {
+		t.Fatalf("GetCredits: %v", err)
+	}
+
+	if len(got[mbids[creditLookupBatch+3]]) != 2 {
+		t.Errorf("a credit past the first batch boundary was not returned")
+	}
+}

backend/explore/dumpcredits.go

@@ -0,0 +1,637 @@
+//go:build indexbuild
+
+package explore
+
+import (
+	"archive/tar"
+	"bufio"
+	"compress/bzip2"
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"path"
+	"regexp"
+	"strconv"
+	"strings"
+)
+
+// Multi-artist credits, from the core MusicBrainz dump.
+//
+// A credit is ordered parts and the credit *string* is derived from
+// them; MusicBrainz's own artist_credit.name is a cached render.  What
+// this pass extracts is the decomposition: for each catalog recording
+// and release group whose credit names more than one artist, the
+// credited artists in order, each with the name *as credited* and the
+// join phrase that follows it.  See artist_credit_part.sql for why that
+// is stored rather than derived, and why nothing may reconstruct a
+// credit by searching a name inside a credit string.
+//
+// It is a separate dump from everything else here, and it has to be.
+// The canonical dump this importer already streams gives artist_mbids
+// (an ordered list) and artist_credit_name (the *rendered* string) --
+// no join phrases, and no per-artist as-credited names.  Splitting the
+// rendered string using canonical artist names fails on exactly the
+// credits that matter: measured on a real library, 21% of multi-artist
+// credits name an artist differently from the artist's own name
+// ("Snoop Dogg" credited on a track by "Snoop Doggy Dogg"), so the
+// substring is simply not there.  The JSON dumps were checked too and
+// cover 153,691 recordings of ~35M, with zero overlap against a real
+// library.  This dump is the only source.
+//
+// Cost, measured on the 20260815 export: 7.1 GB compressed, decompressed
+// by pure-Go compress/bzip2 at ~26 MB/s uncompressed (~13.7 min for the
+// whole file, single-threaded).  cmd/indexbuild is built CGO_ENABLED=0,
+// so the stdlib decompressor is what there is -- and it is fine, because
+// the 2 MB/s origin throttle dominates, as it does for every other dump
+// here.
+
+const (
+	// defaultMBDumpBaseURL is the core MusicBrainz export.  Only
+	// mbdump.tar.bz2 is fetched; the other tarballs there hold data this
+	// app has no use for.
+	defaultMBDumpBaseURL = "https://data.metabrainz.org/pub/musicbrainz/data/fullexport/"
+)
+
+var (
+	mbdumpDirRe  = regexp.MustCompile(`^\d{8}-\d+$`)
+	mbdumpFileRe = regexp.MustCompile(`^mbdump\.tar\.bz2$`)
+
+	// ErrDumpShape is returned when a dump member does not have the
+	// columns this code was written against.  It is deliberately fatal:
+	// reading the wrong column silently produces a catalog whose credits
+	// are subtly wrong, which is far worse than a failed build.
+	ErrDumpShape = errors.New("musicbrainz dump member has an unexpected shape")
+)
+
+// Column positions in the Postgres COPY output, verified against the
+// 20260815 export.  There is no header row to read them from, so they
+// are asserted instead -- see checkShape.
+const (
+	artistColID  = 0
+	artistColGID = 1
+	artistColMin = 2
+
+	creditColID          = 0
+	creditColArtistCount = 2
+	creditColMin         = 3
+
+	partColCredit   = 0
+	partColPosition = 1
+	partColArtist   = 2
+	partColName     = 3
+	partColJoin     = 4
+	partColMin      = 5
+
+	// recording and release_group share a layout in the columns this
+	// pass reads: id, gid, name, artist_credit, ...
+	entityColGID    = 1
+	entityColCredit = 3
+	entityColMin    = 4
+)
+
+// creditPart is one credited artist within a credit.
+type creditPart struct {
+	position int
+	artistID int32
+	name     string
+	join     string
+}
+
+// creditScan is what one pass over the dump collects.
+type creditScan struct {
+	// artistGIDs maps an artist row id to its MBID.  artist_credit_name
+	// references artists by row id, and the tar orders `artist` before
+	// it, so this is complete by the time it is read.
+	artistGIDs map[int32]uuid16
+
+	// multiCredits are the credit ids naming more than one artist, from
+	// artist_credit.artist_count.  Taking the count from the dump rather
+	// than counting parts means a credit can be rejected before its
+	// parts are stored.
+	multiCredits map[int32]struct{}
+
+	// parts are the decompositions of multiCredits, keyed by credit id.
+	parts map[int32][]creditPart
+
+	// refs maps a kept catalog entity to its credit.  Only entities in
+	// explore_index and only multi-artist credits: everything else is
+	// already described by explore_index's own artist_name/artist_mbid.
+	refs map[uuid16]int32
+
+	// used are the credits some ref actually points at, which is a small
+	// fraction of multiCredits -- the catalog keeps ~1.8M entities of
+	// MusicBrainz's tens of millions.
+	used map[int32]struct{}
+
+	skippedUnknownArtist int
+}
+
+// creditsImportDoneKey marks in explore_index_meta that the credit pass
+// has run against the current catalog.
+//
+// It is its own marker rather than part of the import's stage state for
+// a resume reason: the credit pass runs *after* the catalog is
+// assembled, and a failure in it must not send the next run back
+// through the ~205 GB it just finished.  Marking separately means a
+// retry retries only this.
+const creditsImportDoneKey = "credits_import_done"
+
+// ensureArtistCredits runs the credit pass unless it has already run
+// against this catalog, reporting whether it newly populated them.
+//
+// Called from both of run's paths -- the full import and the resume
+// that finds the rows already assembled -- and from the maintenance
+// entry point below, since a catalog built before credits existed is
+// otherwise never offered a chance to gain them: the index job picks
+// its mode from the index's own state, and a complete import means
+// "refresh", which never enters run() at all.
+//
+// The return value is what tells the job there is something new worth
+// publishing.  A refresh otherwise reports "changed" only when the
+// listens series advanced, so credits would sit in the CI database and
+// never reach an artifact.
+func (imp *dumpImporter) ensureArtistCredits(ctx context.Context) bool {
+	if imp.si.hasMeta(creditsImportDoneKey) {
+		return false
+	}
+
+	url, err := discoverDumpFile(
+		ctx, imp.httpClient, imp.mbdumpBaseURL, mbdumpDirRe, mbdumpFileRe,
+	)
+	if err != nil {
+		imp.logger.Warn("credit import: could not find the dump", "error", err)
+
+		return false
+	}
+
+	if err := imp.importArtistCredits(ctx, url); err != nil {
+		// A catalog without credits is the catalog this app shipped
+		// before them: every credit falls back to its single artist.
+		// That is worth far less than failing an import that otherwise
+		// succeeded.
+		imp.logger.Warn("credit import: failed", "error", err)
+
+		return false
+	}
+
+	imp.si.setMeta(creditsImportDoneKey, "1")
+
+	return true
+}
+
+// EnsureArtistCredits tops up the credit tables outside a full import.
+//
+// It exists because the index job's modes are decided from the index's
+// own state: a cache holding a completed import chooses `refresh`,
+// which folds in incremental listens and never enters the dump
+// importer.  Without this, a catalog built before the credit pass
+// existed could only gain credits from a `rebuild` -- and a rebuild
+// re-downloads ~205 GB to reproduce rows it already has, to add
+// something that costs 7 GB on its own.
+//
+// Reports whether credits were newly populated, so the caller knows
+// there is a new artifact worth publishing.
+func (e *Service) EnsureArtistCredits(ctx context.Context) bool {
+	imp, err := newDumpImporter(e.index, e.lb)
+	if err != nil {
+		e.index.logger.Warn("credit import: could not start", "error", err)
+
+		return false
+	}
+
+	return imp.ensureArtistCredits(ctx)
+}
+
+// importArtistCredits streams the core MusicBrainz dump and fills
+// artist_credit_part and artist_credit_ref for the entities the catalog
+// kept.
+//
+// It runs after assembleIndex because it asks explore_index which
+// entities those are: the popularity filter decides what is worth
+// carrying credits for, and asking the table rather than the kept sets
+// means this stays correct if that filter changes.
+func (imp *dumpImporter) importArtistCredits(ctx context.Context, url string) error {
+	kept, err := imp.keptEntityMBIDs(ctx)
+	if err != nil {
+		return err
+	}
+
+	if len(kept) == 0 {
+		imp.logger.Warn("credit import: no catalog entities, skipping")
+
+		return nil
+	}
+
+	imp.logger.Info("credit import: starting", "url", url, "entities", len(kept))
+	imp.logJob("Streaming MusicBrainz dump for artist credits")
+
+	scan, err := imp.scanCreditDump(ctx, url, kept)
+	if err != nil {
+		return err
+	}
+
+	imp.logger.Info("credit import: scanned",
+		"multiArtistCredits", len(scan.multiCredits),
+		"entitiesWithMultiArtistCredit", len(scan.refs),
+		"creditsUsed", len(scan.used),
+	)
+
+	return imp.writeCredits(ctx, scan)
+}
+
+// keptEntityMBIDs is every recording and release group in the catalog.
+// Artists are excluded: an artist is not credited to a credit.
+func (imp *dumpImporter) keptEntityMBIDs(ctx context.Context) (map[uuid16]struct{}, error) {
+	rows, err := imp.si.db.QueryContextWith(ctx,
+		`SELECT mbid FROM explore_index
+		 WHERE entity_type IN (2 /* release_group */, 3 /* recording */)`,
+	)
+	if err != nil {
+		return nil, fmt.Errorf("credit import: read catalog entities: %w", err)
+	}
+
+	defer func() { _ = rows.Close() }()
+
+	out := make(map[uuid16]struct{})
+
+	for rows.Next() {
+		var raw []byte
+
+		if err := rows.Scan(&raw); err != nil {
+			return nil, fmt.Errorf("credit import: scan mbid: %w", err)
+		}
+
+		if len(raw) != len(uuid16{}) {
+			continue
+		}
+
+		var id uuid16
+
+		copy(id[:], raw)
+
+		out[id] = struct{}{}
+	}
+
+	if err := rows.Err(); err != nil {
+		return nil, fmt.Errorf("credit import: read catalog entities: %w", err)
+	}
+
+	return out, nil
+}
+
+// scanCreditDump makes one sequential pass over mbdump.tar.bz2.
+//
+// The tar's members are alphabetical, which is what makes a single pass
+// possible without buffering the big ones: `artist` and
+// `artist_credit_name` both arrive before `recording` and
+// `release_group`, so by the time an entity names a credit, that
+// credit's parts and their artists' MBIDs are already known and the
+// entity can be resolved and dropped.  35M recording rows are never
+// held.
+//
+// The order is not depended on blindly: an entity naming a credit that
+// has not been seen is counted and reported rather than silently
+// producing an empty catalog, which is what a reordered export would
+// otherwise look like.
+func (imp *dumpImporter) scanCreditDump(
+	ctx context.Context, url string, kept map[uuid16]struct{},
+) (*creditScan, error) {
+	stream := imp.openDumpStream(ctx, url, 0)
+
+	defer func() { _ = stream.Close() }()
+
+	return imp.scanCreditTar(
+		ctx,
+		tar.NewReader(bzip2.NewReader(bufio.NewReaderSize(stream, 1<<20))),
+		kept,
+	)
+}
+
+// scanCreditTar is the parse, separated from the fetch so it can be
+// driven by a tar built in a test.  compress/bzip2 is decompress-only,
+// so a test cannot produce the real container.
+func (imp *dumpImporter) scanCreditTar(
+	ctx context.Context, tr *tar.Reader, kept map[uuid16]struct{},
+) (*creditScan, error) {
+	scan := &creditScan{
+		artistGIDs:   make(map[int32]uuid16),
+		multiCredits: make(map[int32]struct{}),
+		parts:        make(map[int32][]creditPart),
+		refs:         make(map[uuid16]int32),
+		used:         make(map[int32]struct{}),
+	}
+
+	for {
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
+
+		hdr, err := tr.Next()
+		if errors.Is(err, io.EOF) {
+			break
+		}
+
+		if err != nil {
+			return nil, fmt.Errorf("credit import: tar: %w", err)
+		}
+
+		if hdr.Typeflag != tar.TypeReg {
+			continue
+		}
+
+		done, err := imp.scanCreditMember(ctx, hdr.Name, tr, kept, scan)
+		if err != nil {
+			return nil, err
+		}
+
+		if done {
+			// Everything this pass needs has been read; the rest of the
+			// tarball is other entities' data and decompressing it would
+			// cost minutes for nothing.
+			break
+		}
+	}
+
+	if scan.skippedUnknownArtist > 0 {
+		imp.logger.Warn("credit import: credits dropped for unknown artists",
+			"count", scan.skippedUnknownArtist,
+		)
+	}
+
+	return scan, nil
+}
+
+// scanCreditMember dispatches one tar member, reporting whether the
+// pass has everything it needs.
+func (imp *dumpImporter) scanCreditMember(
+	ctx context.Context, name string, r io.Reader,
+	kept map[uuid16]struct{}, scan *creditScan,
+) (bool, error) {
+	switch path.Base(name) {
+	case "artist":
+		return false, imp.scanArtists(ctx, r, scan)
+	case "artist_credit":
+		return false, imp.scanCredits(ctx, r, scan)
+	case "artist_credit_name":
+		return false, imp.scanCreditParts(ctx, r, scan)
+	case "recording", "release_group":
+		if err := imp.scanCreditedEntities(ctx, r, kept, scan); err != nil {
+			return false, err
+		}
+
+		// release_group sorts after recording, so the pass is complete
+		// once it has been read.
+		return path.Base(name) == "release_group", nil
+	default:
+		return false, nil
+	}
+}
+
+// scanArtists records every artist's MBID by row id.
+func (imp *dumpImporter) scanArtists(
+	ctx context.Context, r io.Reader, scan *creditScan,
+) error {
+	return scanTSV(ctx, r, artistColMin, "artist", func(fields []string) error {
+		id, ok := parseInt32(fields[artistColID])
+		if !ok {
+			return nil
+		}
+
+		var gid uuid16
+
+		if !parseUUID(fields[artistColGID], gid[:]) {
+			return fmt.Errorf("%w: artist.gid is not a UUID: %q",
+				ErrDumpShape, truncate(fields[artistColGID]))
+		}
+
+		scan.artistGIDs[id] = gid
+
+		return nil
+	})
+}
+
+// scanCredits records which credits name more than one artist.
+func (imp *dumpImporter) scanCredits(
+	ctx context.Context, r io.Reader, scan *creditScan,
+) error {
+	return scanTSV(ctx, r, creditColMin, "artist_credit", func(fields []string) error {
+		id, ok := parseInt32(fields[creditColID])
+		if !ok {
+			return nil
+		}
+
+		count, ok := parseInt32(fields[creditColArtistCount])
+		if !ok {
+			return fmt.Errorf("%w: artist_credit.artist_count is not a number: %q",
+				ErrDumpShape, truncate(fields[creditColArtistCount]))
+		}
+
+		if count > 1 {
+			scan.multiCredits[id] = struct{}{}
+		}
+
+		return nil
+	})
+}
+
+// scanCreditParts records the decomposition of every multi-artist
+// credit.
+func (imp *dumpImporter) scanCreditParts(
+	ctx context.Context, r io.Reader, scan *creditScan,
+) error {
+	return scanTSV(ctx, r, partColMin, "artist_credit_name", func(fields []string) error {
+		credit, ok := parseInt32(fields[partColCredit])
+		if !ok {
+			return nil
+		}
+
+		if _, multi := scan.multiCredits[credit]; !multi {
+			return nil
+		}
+
+		position, ok := parseInt32(fields[partColPosition])
+		if !ok {
+			return nil
+		}
+
+		artist, ok := parseInt32(fields[partColArtist])
+		if !ok {
+			return nil
+		}
+
+		scan.parts[credit] = append(scan.parts[credit], creditPart{
+			position: int(position),
+			artistID: artist,
+			name:     fields[partColName],
+			join:     fields[partColJoin],
+		})
+
+		return nil
+	})
+}
+
+// scanCreditedEntities resolves recordings and release groups against
+// the catalog, keeping only those the catalog holds and whose credit
+// names more than one artist.
+func (imp *dumpImporter) scanCreditedEntities(
+	ctx context.Context, r io.Reader, kept map[uuid16]struct{}, scan *creditScan,
+) error {
+	return scanTSV(ctx, r, entityColMin, "recording/release_group",
+		func(fields []string) error {
+			var gid uuid16
+
+			if !parseUUID(fields[entityColGID], gid[:]) {
+				return fmt.Errorf("%w: entity gid is not a UUID: %q",
+					ErrDumpShape, truncate(fields[entityColGID]))
+			}
+
+			if _, want := kept[gid]; !want {
+				return nil
+			}
+
+			credit, ok := parseInt32(fields[entityColCredit])
+			if !ok {
+				return fmt.Errorf("%w: entity artist_credit is not a number: %q",
+					ErrDumpShape, truncate(fields[entityColCredit]))
+			}
+
+			if _, multi := scan.multiCredits[credit]; !multi {
+				return nil
+			}
+
+			scan.refs[gid] = credit
+			scan.used[credit] = struct{}{}
+
+			return nil
+		})
+}
+
+// scanTSV reads Postgres COPY output a line at a time, unescaping each
+// field and handing the row to fn.
+//
+// The shape is asserted on the first row rather than trusted: this dump
+// has no header, so a column that moved would otherwise be read as a
+// neighbouring one and produce a catalog that is quietly wrong.
+func scanTSV(
+	ctx context.Context, r io.Reader, minCols int, member string,
+	fn func(fields []string) error,
+) error {
+	sc := bufio.NewScanner(r)
+	sc.Buffer(make([]byte, 0, 1<<20), 1<<24)
+
+	checked := false
+	rows := 0
+
+	for sc.Scan() {
+		rows++
+
+		if rows%(1<<20) == 0 {
+			if err := ctx.Err(); err != nil {
+				return err
+			}
+		}
+
+		line := sc.Text()
+		if line == "" {
+			continue
+		}
+
+		fields := strings.Split(line, "\t")
+		if len(fields) < minCols {
+			if !checked {
+				return fmt.Errorf("%w: %s has %d columns, need at least %d",
+					ErrDumpShape, member, len(fields), minCols)
+			}
+
+			continue
+		}
+
+		checked = true
+
+		for i := range fields {
+			fields[i] = unescapeCopy(fields[i])
+		}
+
+		if err := fn(fields); err != nil {
+			return err
+		}
+	}
+
+	if err := sc.Err(); err != nil {
+		return fmt.Errorf("credit import: read %s: %w", member, err)
+	}
+
+	return nil
+}
+
+// unescapeCopy undoes Postgres COPY's text escaping.  A NULL (\N) is
+// returned as an empty string: every field this pass reads is either a
+// number it will reject or a name whose absence means the same as
+// empty.
+func unescapeCopy(s string) string {
+	if s == `\N` {
+		return ""
+	}
+
+	if !strings.ContainsRune(s, '\\') {
+		return s
+	}
+
+	var b strings.Builder
+
+	b.Grow(len(s))
+
+	for i := 0; i < len(s); i++ {
+		if s[i] != '\\' || i+1 >= len(s) {
+			b.WriteByte(s[i])
+
+			continue
+		}
+
+		i++
+
+		switch s[i] {
+		case 'n':
+			b.WriteByte('\n')
+		case 't':
+			b.WriteByte('\t')
+		case 'r':
+			b.WriteByte('\r')
+		case 'b':
+			b.WriteByte('\b')
+		case 'f':
+			b.WriteByte('\f')
+		case 'v':
+			b.WriteByte('\v')
+		case '\\':
+			b.WriteByte('\\')
+		default:
+			b.WriteByte('\\')
+			b.WriteByte(s[i])
+		}
+	}
+
+	return b.String()
+}
+
+func parseInt32(s string) (int32, bool) {
+	n, err := strconv.ParseInt(s, 10, 32)
+	if err != nil {
+		return 0, false
+	}
+
+	return int32(n), true
+}
+
+// truncate bounds an error message built from dump data, which is
+// attacker-free but can be long.
+func truncate(s string) string {
+	const limit = 64
+
+	if len(s) <= limit {
+		return s
+	}
+
+	return s[:limit] + "..."
+}

backend/explore/dumpcredits_test.go

@@ -0,0 +1,406 @@
+//go:build indexbuild
+
+package explore
+
+import (
+	"archive/tar"
+	"bytes"
+	"context"
+	"errors"
+	"strings"
+	"testing"
+
+	"yellowjacket/backend/database"
+)
+
+// tarOf builds an uncompressed tar of the named members, in the order
+// given.  Order is the point of several of these tests: the real dump's
+// members are alphabetical, which is what lets one pass resolve an
+// entity's credit without buffering 35M recordings.
+func tarOf(t *testing.T, members ...[2]string) *tar.Reader {
+	t.Helper()
+
+	var buf bytes.Buffer
+
+	tw := tar.NewWriter(&buf)
+
+	for _, m := range members {
+		body := []byte(m[1])
+
+		if err := tw.WriteHeader(&tar.Header{
+			Name:     "mbdump/" + m[0],
+			Mode:     0o644,
+			Size:     int64(len(body)),
+			Typeflag: tar.TypeReg,
+		}); err != nil {
+			t.Fatalf("tar header: %v", err)
+		}
+
+		if _, err := tw.Write(body); err != nil {
+			t.Fatalf("tar write: %v", err)
+		}
+	}
+
+	if err := tw.Close(); err != nil {
+		t.Fatalf("tar close: %v", err)
+	}
+
+	return tar.NewReader(&buf)
+}
+
+func tsv(rows ...[]string) string {
+	var b strings.Builder
+
+	for _, r := range rows {
+		b.WriteString(strings.Join(r, "\t"))
+		b.WriteByte('\n')
+	}
+
+	return b.String()
+}
+
+// mustMBID is testMBID in the packed form the catalog stores.
+func mustMBID(label string) uuid16 {
+	var u uuid16
+
+	if !parseUUID(testMBID(label), u[:]) {
+		panic("testMBID did not produce a UUID for " + label)
+	}
+
+	return u
+}
+
+// The two artists of the worked example, and the entities they credit.
+var (
+	creditRecMBID = mustMBID("recording-1")
+	creditRGMBID  = mustMBID("release-group-1")
+)
+
+// sampleDump is the shape verified against the 20260815 export:
+// artist(id, gid, ...), artist_credit(id, name, artist_count, ...),
+// artist_credit_name(credit, position, artist, name, join_phrase),
+// recording/release_group(id, gid, name, artist_credit, ...).
+func sampleDump(t *testing.T) *tar.Reader {
+	t.Helper()
+
+	return tarOf(t,
+		[2]string{"artist", tsv(
+			[]string{"11", testMBID("artist-a"), "Snoop Doggy Dogg", "Snoop Doggy Dogg"},
+			[]string{"22", testMBID("artist-b"), "2Pac", "2Pac"},
+		)},
+		[2]string{"artist_credit", tsv(
+			[]string{"900", "2Pac feat. Snoop Dogg", "2", "1", "", "0", ""},
+			[]string{"901", "Solo Artist", "1", "1", "", "0", ""},
+		)},
+		[2]string{"artist_credit_name", tsv(
+			// Deliberately out of position order: the dump is not
+			// obliged to emit them sorted and the credit's meaning is
+			// the order, not the file's.
+			[]string{"900", "1", "11", "Snoop Dogg", ""},
+			[]string{"900", "0", "22", "2Pac", " feat. "},
+			[]string{"901", "0", "11", "Solo Artist", ""},
+		)},
+		[2]string{"recording", tsv(
+			[]string{"1", testMBID("recording-1"), "Some Song", "900", "180000"},
+			[]string{"2", testMBID("not-kept"), "Other", "900", "1"},
+			[]string{"3", testMBID("solo"), "Solo", "901", "1"},
+		)},
+		[2]string{"release_group", tsv(
+			[]string{"5", testMBID("release-group-1"), "Some Album", "900", "1"},
+		)},
+	)
+}
+
+func creditTestImporter(t *testing.T) *dumpImporter {
+	t.Helper()
+
+	db := database.NewTestDB(t)
+
+	return &dumpImporter{
+		si:     NewSearchIndex(db, nil, nil, testLogger()),
+		logger: testLogger(),
+	}
+}
+
+// TestScanCreditDumpDecomposes is the worked example end to end: the
+// credit's parts come back in position order, with the *credited*
+// names and the join phrase between them.
+func TestScanCreditDumpDecomposes(t *testing.T) {
+	imp := creditTestImporter(t)
+
+	kept := map[uuid16]struct{}{
+		creditRecMBID: {},
+		creditRGMBID:  {},
+	}
+
+	scan, err := imp.scanCreditTar(context.Background(), sampleDump(t), kept)
+	if err != nil {
+		t.Fatalf("scan: %v", err)
+	}
+
+	if got := len(scan.refs); got != 2 {
+		t.Fatalf("refs = %d, want 2 (the recording and the release group)", got)
+	}
+
+	if scan.refs[creditRecMBID] != 900 {
+		t.Errorf("recording credit = %d, want 900", scan.refs[creditRecMBID])
+	}
+
+	parts := scan.parts[900]
+	if len(parts) != 2 {
+		t.Fatalf("parts = %d, want 2", len(parts))
+	}
+
+	// Sorting happens on write, so assert the pieces are all present
+	// and let the render test below check the order.
+	byPos := map[int]creditPart{}
+	for _, p := range parts {
+		byPos[p.position] = p
+	}
+
+	if byPos[0].name != "2Pac" || byPos[0].join != " feat. " {
+		t.Errorf("position 0 = %q/%q, want \"2Pac\"/\" feat. \"",
+			byPos[0].name, byPos[0].join)
+	}
+
+	// The credited name, not the artist's own name: this is the whole
+	// reason credited_name is stored per row.
+	if byPos[1].name != "Snoop Dogg" {
+		t.Errorf("position 1 credited name = %q, want \"Snoop Dogg\"", byPos[1].name)
+	}
+}
+
+// TestSingleArtistCreditsAreNotStored: a one-artist credit is already
+// described by explore_index's artist_name/artist_mbid, and storing it
+// would roughly triple the table to say nothing new.
+func TestSingleArtistCreditsAreNotStored(t *testing.T) {
+	imp := creditTestImporter(t)
+
+	solo := mustMBID("solo")
+	kept := map[uuid16]struct{}{solo: {}}
+
+	scan, err := imp.scanCreditTar(context.Background(), sampleDump(t), kept)
+	if err != nil {
+		t.Fatalf("scan: %v", err)
+	}
+
+	if len(scan.refs) != 0 {
+		t.Fatalf("a single-artist credit was referenced: %v", scan.refs)
+	}
+
+	if _, ok := scan.multiCredits[901]; ok {
+		t.Error("credit 901 has artist_count 1 and should not be multi")
+	}
+}
+
+// TestOnlyKeptEntitiesAreReferenced: the catalog's popularity filter
+// decides what is worth carrying credits for, and an entity outside it
+// must not produce a row pointing at nothing.
+func TestOnlyKeptEntitiesAreReferenced(t *testing.T) {
+	imp := creditTestImporter(t)
+
+	kept := map[uuid16]struct{}{creditRecMBID: {}}
+
+	scan, err := imp.scanCreditTar(context.Background(), sampleDump(t), kept)
+	if err != nil {
+		t.Fatalf("scan: %v", err)
+	}
+
+	if _, ok := scan.refs[mustMBID("not-kept")]; ok {
+		t.Error("an entity outside the catalog was referenced")
+	}
+
+	if len(scan.used) != 1 {
+		t.Errorf("used credits = %d, want 1", len(scan.used))
+	}
+}
+
+// TestWriteCreditsRoundTrips checks what the frontend will actually
+// read: parts in position order, dashed MBIDs out of the 16 raw bytes,
+// and a rendered credit that reassembles to the tagged string.
+func TestWriteCreditsRoundTrips(t *testing.T) {
+	imp := creditTestImporter(t)
+
+	kept := map[uuid16]struct{}{creditRecMBID: {}, creditRGMBID: {}}
+
+	scan, err := imp.scanCreditTar(context.Background(), sampleDump(t), kept)
+	if err != nil {
+		t.Fatalf("scan: %v", err)
+	}
+
+	if err := imp.writeCredits(context.Background(), scan); err != nil {
+		t.Fatalf("writeCredits: %v", err)
+	}
+
+	rows, err := imp.si.db.QueryContext(
+		`SELECT p.position, p.artist_mbid, p.credited_name, p.join_phrase
+		 FROM artist_credit_ref r
+		 JOIN artist_credit_part p ON p.credit_id = r.credit_id
+		 WHERE r.mbid = ?
+		 ORDER BY p.position`,
+		creditRecMBID[:],
+	)
+	if err != nil {
+		t.Fatalf("query: %v", err)
+	}
+
+	defer func() { _ = rows.Close() }()
+
+	var rendered strings.Builder
+
+	names := []string{}
+
+	for rows.Next() {
+		var (
+			pos  int
+			mbid []byte
+			name string
+			join string
+		)
+
+		if err := rows.Scan(&pos, &mbid, &name, &join); err != nil {
+			t.Fatalf("scan row: %v", err)
+		}
+
+		if len(mbid) != 16 {
+			t.Fatalf("artist_mbid is %d bytes, want 16", len(mbid))
+		}
+
+		names = append(names, name)
+
+		rendered.WriteString(name)
+		rendered.WriteString(join)
+	}
+
+	if err := rows.Err(); err != nil {
+		t.Fatalf("rows: %v", err)
+	}
+
+	// Concatenation is the contract: names in order, join phrases
+	// between them, and no searching a name inside a credit string.
+	if got := rendered.String(); got != "2Pac feat. Snoop Dogg" {
+		t.Errorf("rendered credit = %q, want %q", got, "2Pac feat. Snoop Dogg")
+	}
+
+	if len(names) != 2 || names[0] != "2Pac" {
+		t.Errorf("parts came back out of position order: %v", names)
+	}
+}
+
+// TestCreditRefsNeverDangle: a ref whose parts were not stored renders
+// as a credit with no artists at all, which is worse than the
+// single-artist fallback it replaced.
+func TestCreditRefsNeverDangle(t *testing.T) {
+	imp := creditTestImporter(t)
+
+	kept := map[uuid16]struct{}{creditRecMBID: {}}
+
+	scan, err := imp.scanCreditTar(context.Background(), sampleDump(t), kept)
+	if err != nil {
+		t.Fatalf("scan: %v", err)
+	}
+
+	// An artist the dump never named: the credit cannot be navigated to
+	// and must be dropped whole, taking its ref with it.
+	scan.artistGIDs = map[int32]uuid16{}
+
+	if err := imp.writeCredits(context.Background(), scan); err != nil {
+		t.Fatalf("writeCredits: %v", err)
+	}
+
+	var refs, parts int
+
+	if err := imp.si.db.QueryRowWriter(
+		"SELECT COUNT(*) FROM artist_credit_ref",
+	).Scan(&refs); err != nil {
+		t.Fatalf("count refs: %v", err)
+	}
+
+	if err := imp.si.db.QueryRowWriter(
+		"SELECT COUNT(*) FROM artist_credit_part",
+	).Scan(&parts); err != nil {
+		t.Fatalf("count parts: %v", err)
+	}
+
+	if refs != 0 || parts != 0 {
+		t.Fatalf("refs=%d parts=%d, want 0/0 when the artists are unknown", refs, parts)
+	}
+}
+
+// TestCreditDumpShapeIsAsserted: the dump has no header row, so a
+// column that moved would be read as its neighbour and produce a
+// catalog that is quietly wrong.  Loud is the requirement.
+func TestCreditDumpShapeIsAsserted(t *testing.T) {
+	imp := creditTestImporter(t)
+
+	short := tarOf(t, [2]string{"artist", tsv([]string{"11", "only-two-columns"})})
+
+	_, err := imp.scanCreditTar(context.Background(), short, map[uuid16]struct{}{})
+	if err == nil {
+		t.Fatal("a member with a non-UUID gid was accepted")
+	}
+
+	if !errors.Is(err, ErrDumpShape) {
+		t.Errorf("error = %v, want ErrDumpShape", err)
+	}
+}
+
+// TestUnescapeCopy covers Postgres COPY's text escaping, which reaches
+// artist names routinely -- a tab or backslash in a name would
+// otherwise shift every field after it.
+func TestUnescapeCopy(t *testing.T) {
+	tests := []struct{ in, want string }{
+		{`plain`, `plain`},
+		{`\N`, ``},
+		{`a\tb`, "a\tb"},
+		{`a\nb`, "a\nb"},
+		{`back\\slash`, `back\slash`},
+		{`AC\/DC`, `AC\/DC`},
+		{`trailing\`, `trailing\`},
+	}
+
+	for _, tt := range tests {
+		if got := unescapeCopy(tt.in); got != tt.want {
+			t.Errorf("unescapeCopy(%q) = %q, want %q", tt.in, got, tt.want)
+		}
+	}
+}
+
+// TestEnsureArtistCreditsIsIdempotent pins what the index job depends
+// on to decide whether to publish.
+//
+// The pass runs on every mode, including the `refresh` that a complete
+// catalog always chooses — so it must be free when there is nothing to
+// do, and it must say so.  A `true` here republishes the artifact; a
+// `true` on every run would republish an identical one weekly, and a
+// permanent `false` would mean a catalog that never gains credits at
+// all.
+func TestEnsureArtistCreditsIsIdempotent(t *testing.T) {
+	imp := creditTestImporter(t)
+
+	// The marker is what "already done" means; with it set, the pass
+	// must not reach the network or report a change.
+	imp.si.setMeta(creditsImportDoneKey, "1")
+
+	if imp.ensureArtistCredits(context.Background()) {
+		t.Fatal("a second run reported new credits; the artifact would republish forever")
+	}
+}
+
+// TestEnsureArtistCreditsReportsFailureAsNoChange: a dump that cannot be
+// reached leaves the catalog exactly as it was, and must not claim
+// otherwise — publishing on it would ship an artifact with no credits
+// and mark the work done.
+func TestEnsureArtistCreditsReportsFailureAsNoChange(t *testing.T) {
+	imp := creditTestImporter(t)
+	imp.httpClient = newDumpHTTPClient()
+	imp.mbdumpBaseURL = "http://127.0.0.1:1/nonexistent/"
+
+	if imp.ensureArtistCredits(context.Background()) {
+		t.Fatal("an unreachable dump reported new credits")
+	}
+
+	if imp.si.hasMeta(creditsImportDoneKey) {
+		t.Error("a failed pass marked itself done; it would never retry")
+	}
+}

backend/explore/dumpcreditswrite.go

@@ -0,0 +1,164 @@
+//go:build indexbuild
+
+package explore
+
+import (
+	"context"
+	"database/sql"
+	"fmt"
+	"sort"
+)
+
+// writeCredits persists the scanned decompositions.
+//
+// Only credits some catalog entity actually points at are written: the
+// dump has millions of multi-artist credits and the catalog keeps ~1.8M
+// entities, so storing every credit would be most of a table nothing
+// can reach.
+//
+// The two tables are written in one transaction, because a ref pointing
+// at parts that are not there renders as a credit with no artists --
+// worse than the single-artist fallback it replaced.
+func (imp *dumpImporter) writeCredits(ctx context.Context, scan *creditScan) error {
+	tx, err := imp.si.db.BeginTx()
+	if err != nil {
+		return fmt.Errorf("credit import: begin: %w", err)
+	}
+
+	defer func() { _ = tx.Rollback() }()
+
+	// A rebuild replaces the previous pass wholesale.  These are Cache
+	// tables derived entirely from the dump, so there is nothing to
+	// merge and a stale row is a wrong credit.
+	for _, table := range []string{"artist_credit_part", "artist_credit_ref"} {
+		if _, err := tx.ExecContext(ctx, "DELETE FROM "+table); err != nil {
+			return fmt.Errorf("credit import: clear %s: %w", table, err)
+		}
+	}
+
+	written, err := imp.writeCreditParts(ctx, tx, scan)
+	if err != nil {
+		return err
+	}
+
+	refs, err := imp.writeCreditRefs(ctx, tx, scan, written)
+	if err != nil {
+		return err
+	}
+
+	if err := tx.Commit(); err != nil {
+		return fmt.Errorf("credit import: commit: %w", err)
+	}
+
+	imp.logger.Info("credit import: complete",
+		"credits", len(written),
+		"refs", refs,
+	)
+
+	return nil
+}
+
+// writeCreditParts inserts the parts of every used credit and returns
+// the set of credits that were actually stored.
+//
+// A credit is stored whole or not at all.  If any of its artists has no
+// MBID -- which should not happen, the dump being self-consistent, but
+// would leave a part that cannot be navigated to -- the credit is
+// dropped and the entity falls back to explore_index's single artist,
+// which is a worse answer rather than a broken one.
+func (imp *dumpImporter) writeCreditParts(
+	ctx context.Context, tx *sql.Tx, scan *creditScan,
+) (map[int32]struct{}, error) {
+	stmt, err := tx.PrepareContext(ctx,
+		`INSERT INTO artist_credit_part
+			(credit_id, position, artist_mbid, credited_name, join_phrase)
+		 VALUES (?, ?, ?, ?, ?)`,
+	)
+	if err != nil {
+		return nil, fmt.Errorf("credit import: prepare part insert: %w", err)
+	}
+
+	defer func() { _ = stmt.Close() }()
+
+	written := make(map[int32]struct{}, len(scan.used))
+
+	for credit := range scan.used {
+		parts := scan.parts[credit]
+		if len(parts) < 2 {
+			// artist_credit said more than one artist and
+			// artist_credit_name did not deliver them.  Nothing to
+			// decompose, so leave the entity to its single artist.
+			continue
+		}
+
+		// Position order is the credit's meaning, and the dump is not
+		// obliged to emit it sorted.
+		sort.Slice(parts, func(i, j int) bool {
+			return parts[i].position < parts[j].position
+		})
+
+		resolved := make([][]any, 0, len(parts))
+		ok := true
+
+		for _, part := range parts {
+			gid, found := scan.artistGIDs[part.artistID]
+			if !found {
+				scan.skippedUnknownArtist++
+				ok = false
+
+				break
+			}
+
+			resolved = append(resolved, []any{
+				credit, part.position, gid[:], part.name, part.join,
+			})
+		}
+
+		if !ok {
+			continue
+		}
+
+		for _, args := range resolved {
+			if _, err := stmt.ExecContext(ctx, args...); err != nil {
+				return nil, fmt.Errorf("credit import: insert part: %w", err)
+			}
+		}
+
+		written[credit] = struct{}{}
+	}
+
+	return written, nil
+}
+
+// writeCreditRefs points each kept entity at its credit, skipping any
+// whose credit was not stored so a ref never dangles.
+func (imp *dumpImporter) writeCreditRefs(
+	ctx context.Context, tx *sql.Tx, scan *creditScan, written map[int32]struct{},
+) (int, error) {
+	stmt, err := tx.PrepareContext(ctx,
+		"INSERT OR REPLACE INTO artist_credit_ref (mbid, credit_id) VALUES (?, ?)",
+	)
+	if err != nil {
+		return 0, fmt.Errorf("credit import: prepare ref insert: %w", err)
+	}
+
+	defer func() { _ = stmt.Close() }()
+
+	count := 0
+
+	for mbid, credit := range scan.refs {
+		if _, stored := written[credit]; !stored {
+			continue
+		}
+
+		id := mbid
+
+		if _, err := stmt.ExecContext(ctx, id[:], credit); err != nil {
+			return 0, fmt.Errorf("credit import: insert ref: %w", err)
+		}
+
+		count++
+	}
+
+	return count, nil
+}

backend/explore/dumpimport.go

@@ -104,6 +104,7 @@ type dumpImporter struct {
 
 	canonicalBaseURL string
 	listensBaseURL   string
+	mbdumpBaseURL    string
 
 	// Disk safety floors (fields so tests can relax them).
 	minStartFreeBytes uint64
@@ -144,6 +145,7 @@ func newDumpImporter(si *SearchIndex, lb *ListenBrainzClient) (*dumpImporter, er
 		stagingDir:        stagingDir,
 		canonicalBaseURL:  defaultCanonicalBaseURL,
 		listensBaseURL:    defaultListensBaseURL,
+		mbdumpBaseURL:     defaultMBDumpBaseURL,
 		minStartFreeBytes: dumpMinStartFreeBytes,
 		abortFreeBytes:    dumpAbortFreeBytes,
 	}, nil
@@ -171,6 +173,7 @@ func (imp *dumpImporter) run(ctx context.Context) error {
 	// Fast path: rows already assembled, only patch passes remain.
 	if state.Stage == dumpStageAssembled {
 		imp.si.MarkReadyIfPopulated()
+		imp.ensureArtistCredits(ctx)
 		imp.runPatchPasses(ctx)
 
 		if err := ctx.Err(); err != nil {
@@ -305,6 +308,11 @@ func (imp *dumpImporter) run(ctx context.Context) error {
 	imp.si.MarkReadyIfPopulated()
 	imp.si.refreshStatusCounts()
 
+	// Multi-artist credits, from a different dump.  After the catalog,
+	// because it asks explore_index which entities are worth carrying
+	// credits for.
+	imp.ensureArtistCredits(ctx)
+
 	// Stage 4: API patch passes (idempotent).
 	imp.runPatchPasses(ctx)
 

cmd/indexbuild/main.go

@@ -177,12 +177,25 @@ func run(o opts) error {
 
 	complete := svc.IndexImportComplete() && !errors.Is(err, errIncomplete)
 
+	// Credits are maintenance, not part of any one mode.  They come from
+	// a different dump, they are keyed on entities the catalog already
+	// holds, and a catalog built before the pass existed would otherwise
+	// only gain them from a rebuild — which re-downloads ~205 GB to
+	// re-derive rows it already has.  Skipped when the import is not
+	// complete, because there is nothing to key them against yet.
+	creditsAdded := false
+	if complete {
+		creditsAdded = svc.EnsureArtistCredits(context.Background())
+	}
+
 	// "Changed" means there is something new worth publishing, so it is
 	// only ever true for a finished import: a build stamps the listens
 	// series early, long before its rows are assembled, and reporting a
 	// change off that would be a lie about a half-built index.
 	changed := complete &&
-		(svc.IndexBaselineSeries() != seriesBefore || chosen != modeRefresh)
+		(svc.IndexBaselineSeries() != seriesBefore ||
+			chosen != modeRefresh ||
+			creditsAdded)
 
 	report(logger, svc, chosen, complete, changed)
 

cmd/indexbuild/staleschema_test.go

@@ -12,6 +12,7 @@ import (
 	_ "modernc.org/sqlite"
 
 	"yellowjacket/backend/database"
+	"yellowjacket/backend/datamap"
 	"yellowjacket/backend/system"
 )
 
@@ -53,10 +54,19 @@ func TestRetireLibraryTables(t *testing.T) {
 		CREATE TABLE recordings (id INTEGER PRIMARY KEY, title TEXT);
 	`)
 
-	// The symptom, before the repair: the schema cannot be applied over
-	// a table whose shape has moved on.
-	if _, err := database.NewDB(logger); err == nil {
-		t.Fatal("expected the stale shape to fail to open; it did not")
+	// This used to assert the symptom -- that the schema cannot be
+	// applied over a table whose shape has moved on -- because at the
+	// time nothing repaired it and only this job did.  The app-side
+	// repair (backend/database/staleshape.go) now retires a stale
+	// non-authored table before applySchema meets it, so opening
+	// succeeds and the symptom no longer reproduces from here.
+	//
+	// That does not make retireLibraryTables redundant, and the rest of
+	// this test is why: the app-side repair only removes what is *stale*,
+	// while this database wants its library half gone entirely, healthy
+	// or not, because nothing here scans, plays or authors.
+	if _, err := database.NewDB(logger); err != nil {
+		t.Fatalf("the app-side repair should have opened this: %v", err)
 	}
 
 	if err := retireLibraryTables(context.Background(), logger); err != nil {
@@ -116,3 +126,186 @@ func count(t *testing.T, dbPath, query string) int {
 
 	return n
 }
+
+// TestTheCatalogSurvivesAStaleShape is the accident written down.
+//
+// The app repairs a stale Cache table by dropping it: its catalog is
+// downloaded, so a wrong shape costs about a minute of re-fetching and
+// keeping it costs every Explore read. Applied here that rule is
+// catastrophic — this database is what the artifact is *cut from*, so
+// there is nothing to re-fetch and the only way back is the ~205 GB
+// dump stream the /cache volume exists to avoid.
+//
+// It shipped without that distinction and dropped the real CI catalog
+// on the first run:
+//
+//	retiring a table ... table=explore_index
+//	  reason="column entity_type is TEXT, schema declares INTEGER"
+//	index maintenance mode=build reason="no completed import yet"
+//
+// The mismatch was real: that database is deliberately kept in the
+// older encoding, which `fix(indexexport): read an index older than the
+// binary` exists to tolerate. So the shape will not match, every run,
+// by design — and the catalog must survive it anyway.
+func TestTheCatalogSurvivesAStaleShape(t *testing.T) {
+	logger := slog.New(slog.DiscardHandler)
+
+	t.Setenv("YJ_HOME", t.TempDir())
+
+	dataDir, err := system.GetUserDataDirPath()
+	if err != nil {
+		t.Fatalf("resolve data dir: %v", err)
+	}
+
+	dbPath := filepath.Join(dataDir, "yj.db")
+
+	if _, err := database.NewDB(logger); err != nil {
+		t.Fatalf("first open: %v", err)
+	}
+
+	// The shape the real index database is in: every current column,
+	// but the ids and the entity type still text.  That is what the
+	// exporter's backward-compatibility fix tolerates, and it is what
+	// the repair saw and called stale.
+	exec(t, dbPath, `
+		DROP TABLE explore_index;
+		CREATE TABLE explore_index (
+			id                     INTEGER PRIMARY KEY AUTOINCREMENT,
+			entity_type            TEXT NOT NULL,
+			mbid                   TEXT NOT NULL,
+			title                  TEXT NOT NULL,
+			artist_name            TEXT NOT NULL,
+			artist_mbid            TEXT NOT NULL,
+			aliases                TEXT NOT NULL DEFAULT '',
+			popularity             INTEGER NOT NULL DEFAULT 0,
+			listener_count         INTEGER NOT NULL DEFAULT 0,
+			duration               INTEGER NOT NULL DEFAULT 0,
+			caa_release_mbid       TEXT NOT NULL DEFAULT '',
+			release_name           TEXT NOT NULL DEFAULT '',
+			primary_type           TEXT NOT NULL DEFAULT '',
+			secondary_types        TEXT NOT NULL DEFAULT '',
+			release_date           TEXT NOT NULL DEFAULT '',
+			total_tracks           INTEGER NOT NULL DEFAULT 0,
+			artist_type            TEXT NOT NULL DEFAULT '',
+			country                TEXT NOT NULL DEFAULT '',
+			disambiguation         TEXT NOT NULL DEFAULT '',
+			sort_name              TEXT NOT NULL DEFAULT '',
+			in_library             INTEGER NOT NULL DEFAULT 0,
+			is_similar             INTEGER NOT NULL DEFAULT 0,
+			local_artist_id        INTEGER,
+			local_release_group_id INTEGER,
+			local_recording_id     INTEGER,
+			discog_fetched         INTEGER NOT NULL DEFAULT 0,
+			UNIQUE(mbid)
+		);
+		INSERT INTO explore_index
+			(entity_type, mbid, title, artist_name, artist_mbid)
+		VALUES ('artist', 'a-b-c', 'A Catalog Row', 'An Artist', 'd-e-f');
+	`)
+
+	if _, err := database.NewDB(logger); err != nil {
+		t.Fatalf("open with a stale catalog shape: %v", err)
+	}
+
+	if got := count(t, dbPath, "SELECT COUNT(*) FROM explore_index"); got != 1 {
+		t.Fatalf(
+			"explore_index rows = %d, want 1 — the catalog was retired, "+
+				"which costs this database a ~205GB rebuild",
+			got,
+		)
+	}
+}
+
+// TestNoCacheTableIsRetiredHere is the general form of the accident
+// above, and it exists because the specific one is not the risk.
+//
+// `TestTheCatalogSurvivesAStaleShape` pins one table in one wrong shape,
+// which is the failure that happened. What cost the ~205 GB was not that
+// shape: it was a destructive repair added to `database.NewDB` -- the
+// one chokepoint every binary in this project shares -- without asking
+// which binary it was running in. The next such repair will have a
+// different name and a different reason, and this database still cannot
+// afford it.
+//
+// So the assertion is about the *outcome* rather than the mechanism: put
+// every Cache table in a shape the schema has certainly moved past, open
+// the database the way cmd/indexbuild does, and require that all of them
+// are still there afterwards. Any future repair that drops one fails
+// here regardless of how it decides to.
+//
+// Two things about it are deliberate.
+//
+// The table list comes from `datamap.ByKind(Cache)` rather than being
+// written out, so a Cache table added next year is covered by this test
+// on the day it is added -- the same reason `TestCatalogCoversSchema`
+// reads the schema instead of a list.
+//
+// And `NewDB` returning an error is *accepted*, because that is the
+// trade the fix documents: with Cache tables no longer rebuilt here, a
+// shape the schema moved past now fails this job loudly instead of
+// silently costing it a day of downloading. Loud is fine. Gone is not.
+func TestNoCacheTableIsRetiredHere(t *testing.T) {
+	logger := slog.New(slog.DiscardHandler)
+
+	t.Setenv("YJ_HOME", t.TempDir())
+
+	dataDir, err := system.GetUserDataDirPath()
+	if err != nil {
+		t.Fatalf("resolve data dir: %v", err)
+	}
+
+	dbPath := filepath.Join(dataDir, "yj.db")
+
+	if _, err := database.NewDB(logger); err != nil {
+		t.Fatalf("first open: %v", err)
+	}
+
+	// An FTS table is four shadow tables and cannot be given a "wrong
+	// shape" meaningfully; the repair skips them for the same reason and
+	// retires them with their parent, which the parents below cover.
+	var cache []string
+
+	for _, table := range datamap.ByKind(datamap.Cache) {
+		if table.FTS {
+			continue
+		}
+
+		cache = append(cache, table.Name)
+	}
+
+	if len(cache) == 0 {
+		t.Fatal("no Cache tables to check: the datamap or this test is wrong")
+	}
+
+	for _, name := range cache {
+		// A shape nothing in the current schema describes. What matters
+		// is only that it disagrees; the real mismatch was one column's
+		// type.
+		exec(t, dbPath, `
+			DROP TABLE IF EXISTS `+name+`;
+			CREATE TABLE `+name+` (id INTEGER PRIMARY KEY, moved_past TEXT);
+			INSERT INTO `+name+` (moved_past) VALUES ('irreplaceable');
+		`)
+	}
+
+	// The error is not the assertion: see the note above.
+	_, _ = database.NewDB(logger)
+
+	for _, name := range cache {
+		rows := count(t, dbPath,
+			`SELECT count(*) FROM sqlite_master WHERE type = 'table' AND name = '`+name+`'`)
+		if rows == 0 {
+			t.Errorf("%s was retired: in this database a Cache table is derived, "+
+				"not downloaded, and dropping one costs the ~205 GB dump stream", name)
+
+			continue
+		}
+
+		// Present but emptied is the same loss wearing a different
+		// shape: SQLite does an implicit DELETE before a DROP, and a
+		// repair that recreated the table would look identical here.
+		if n := count(t, dbPath, `SELECT count(*) FROM `+name); n == 0 {
+			t.Errorf("%s survived but was emptied", name)
+		}
+	}
+}

cmd/indexexport/main.go

@@ -207,6 +207,26 @@ func createSchema(db *sql.DB) error {
 			key   TEXT PRIMARY KEY,
 			value TEXT NOT NULL
 		)`,
+		// Multi-artist credits.  Shipped as their own tables rather than
+		// as an explore_index column because a credit is a variable
+		// number of ordered parts, and because credits are *shared* --
+		// an album's tracks by one artist reference one credit, which is
+		// what keeps this to a few hundred thousand rows.
+		//
+		// An importer that predates these reads an artifact without
+		// them; artifactHasCredits is what asks.
+		`CREATE TABLE core.artist_credit_part (
+			credit_id     INTEGER NOT NULL,
+			position      INTEGER NOT NULL,
+			artist_mbid   BLOB NOT NULL,
+			credited_name TEXT NOT NULL,
+			join_phrase   TEXT NOT NULL DEFAULT '',
+			PRIMARY KEY (credit_id, position)
+		) WITHOUT ROWID`,
+		`CREATE TABLE core.artist_credit_ref (
+			mbid      BLOB NOT NULL PRIMARY KEY,
+			credit_id INTEGER NOT NULL
+		) WITHOUT ROWID`,
 	}
 
 	for _, stmt := range stmts {
@@ -283,6 +303,63 @@ func copyRows(db *sql.DB, artists, perArtistRGs, perArtistRecs int) error {
 		fmt.Printf("  %-15s %d\n", sel.label+":", n)
 	}
 
+	return copyCredits(db)
+}
+
+// copyCredits ships the credit decomposition for the entities that made
+// it into the artifact, and only those.
+//
+// The refs go first and the parts follow *from* the refs, so a credit is
+// carried only if something in the artifact points at it.  The source
+// index holds credits for every catalog entity, while the artifact is a
+// windowed subset -- copying all of them would carry a large table most
+// of which nothing in the artifact can reach.
+//
+// A source index built before the credit pass simply has no rows here,
+// which is not an error: the artifact then carries the tables empty, and
+// every credit falls back to its single artist exactly as before.
+func copyCredits(db *sql.DB) error {
+	// Asked, not assumed.  A source index built before the credit pass
+	// has no such table, and "no such table" would fail an export whose
+	// catalog is otherwise complete.
+	for _, table := range []string{"artist_credit_ref", "artist_credit_part"} {
+		var n int
+
+		if err := db.QueryRow(
+			`SELECT COUNT(*) FROM main.sqlite_master
+			 WHERE type = 'table' AND name = ?`, table,
+		).Scan(&n); err != nil {
+			return fmt.Errorf("probe %s: %w", table, err)
+		}
+
+		if n == 0 {
+			fmt.Printf("  %-15s none in source\n", "credits:")
+
+			return nil
+		}
+	}
+
+	refs, err := insertSelect(db, `
+		INSERT INTO core.artist_credit_ref (mbid, credit_id)
+		SELECT r.mbid, r.credit_id
+		FROM main.artist_credit_ref r
+		WHERE r.mbid IN (SELECT mbid FROM core.explore_index)`)
+	if err != nil {
+		return err
+	}
+
+	parts, err := insertSelect(db, `
+		INSERT INTO core.artist_credit_part
+			(credit_id, position, artist_mbid, credited_name, join_phrase)
+		SELECT p.credit_id, p.position, p.artist_mbid, p.credited_name, p.join_phrase
+		FROM main.artist_credit_part p
+		WHERE p.credit_id IN (SELECT credit_id FROM core.artist_credit_ref)`)
+	if err != nil {
+		return err
+	}
+
+	fmt.Printf("  %-15s %d refs, %d parts\n", "credits:", refs, parts)
+
 	return nil
 }
 

docs/index-cache.md

@@ -0,0 +1,99 @@
+# The index cache, and why it has a snapshot
+
+`/srv/yellowjacket/index-cache` on the Gitea host is the `YJ_HOME` the
+search-index job keeps between runs — `.gitea/workflows/index-artifact.yml`
+mounts it at `/cache`. It holds the catalog every user eventually
+downloads, and it is the one database in this project that is
+**derived rather than downloaded**.
+
+That is the whole reason this document exists. An install with a broken
+catalog re-fetches the ~0.6 GB artifact and is fine in a minute. This
+database *is* what that artifact is cut from, so its only route back is
+re-streaming the MetaBrainz dumps: hours, at a rate that belongs to
+someone else's server, holding a runner of capacity 1 the entire time.
+
+## What happened on 2026-08-17
+
+A schema repair (`fix(database): retire a table whose shape the schema
+moved past`) dropped every table whose live shape disagreed with the
+schema, before `applySchema`. Correct for the app. Applied here it
+deleted the catalog 19 seconds into the first run:
+
+```
+retiring a table ... table=explore_index
+  reason="column entity_type is TEXT, schema declares INTEGER"
+index maintenance mode=build reason="no completed import yet"
+```
+
+The mismatch was real and deliberate: this database is kept in the older
+text encoding, which `artifactStoresText` and `sourceColumns` exist to
+tolerate. So it would have been judged stale on *every* run.
+
+Two things came out of it. `retireStaleCache` is now a build tag —
+false under `indexbuild`, true in the app — and
+`TestNoCacheTableIsRetiredHere` asserts the outcome rather than the
+mechanism, so the next destructive repair fails a test instead of a
+production volume. And the volume got the snapshot it should always have
+had, below.
+
+## Taking snapshots
+
+```sh
+scripts/index-cache-snapshot.sh [SOURCE_HOME] [DEST_DIR] [KEEP]
+```
+
+Defaults: `/srv/yellowjacket/index-cache`, `/srv/yellowjacket/index-snapshots`,
+keep 2. On the Gitea host, daily and away from the Monday 04:00 build:
+
+```
+30 5 * * * /path/to/index-cache-snapshot.sh >> /var/log/yj-index-snapshot.log 2>&1
+```
+
+Three properties worth knowing before trusting it:
+
+- **It uses `VACUUM INTO`, not `cp`.** The database may be open, and a
+  byte copy of a live SQLite file is a corrupt file of plausible size.
+  `VACUUM INTO` takes a read lock and writes a consistent, compacted
+  copy; it is safe to run while a build is in progress.
+- **It does not copy `data/explore-staging`.** That is a resumable
+  checkpoint of work in flight — large, constantly changing, and a build
+  resumes without it. What cannot be cheaply re-derived is the finished
+  catalog, which is in the database.
+- **It verifies before it rotates.** Each snapshot is reopened and asked
+  for its catalog row count; a run that produces an unreadable or empty
+  file fails loudly, deletes its own output, and leaves the previous
+  snapshots alone. Both paths are exercised, not assumed.
+
+## Restoring
+
+Stop anything that might be using the volume first — the job holds it
+for the length of a build, and the concurrency group (`search-index`)
+means a queued run will start the moment one ends.
+
+```sh
+cd /srv/yellowjacket
+mv index-cache/data/yj.db index-cache/data/yj.db.broken   # keep it until you are sure
+cp index-snapshots/yj-index-<stamp>.db index-cache/data/yj.db
+chown --reference=index-cache/data/yj.db.broken index-cache/data/yj.db
+```
+
+Then dispatch the workflow with `mode=auto`. A restored snapshot is
+older than the dumps, so `indexbuild` resolves to `refresh` and folds in
+the incremental listens since — which is minutes, not hours.
+
+Two notes on what a restore does *not* need. The staging directory can
+be deleted; it will be rebuilt if a build is needed. And the published
+artifact is untouched by any of this: users keep downloading the last
+good one until a run reports `complete=true` and `changed=true`
+republishes.
+
+## The trade this leaves open
+
+With Cache tables no longer retired under `indexbuild`, a future
+`explore_index` column change will fail this job **loudly** — at
+`applySchema`, or at the first query naming the column — rather than
+silently rebuilding. That is the right default: loud is recoverable and
+a silent day of downloading is not. It does mean the next schema change
+touching `explore_index` needs a deliberate plan for this one database:
+take a snapshot, apply the change to a copy, or accept a rebuild
+knowingly.

frontend/bindings/yellowjacket/backend/explore/index.ts

@@ -12,6 +12,7 @@ export {
 
 export type {
     AlbumCompleteFunc,
+    CreditPart,
     IndexStatus,
     LBSimilarArtist,
     LBTopRecording,

frontend/bindings/yellowjacket/backend/explore/models.ts

@@ -7,6 +7,23 @@
  */
 export type AlbumCompleteFunc = any;
 
+/**
+ * CreditPart is one credited artist within a credit, in credit order.
+ * 
+ * CreditedName is the name *as credited*, which is not the artist's own
+ * name: MusicBrainz credits "Snoop Dogg" on a track by the artist
+ * called "Snoop Doggy Dogg".  Display uses it; navigation uses
+ * ArtistMBID.  JoinPhrase is the literal connector that follows this
+ * part, so a credit renders by concatenation and never by searching a
+ * name inside a credit string.
+ */
+export interface CreditPart {
+    "position": number;
+    "artistMbid": string;
+    "creditedName": string;
+    "joinPhrase": string;
+}
+
 /**
  * IndexStatus is the full index build status, exposed to the frontend.
  */

frontend/bindings/yellowjacket/backend/explore/service.ts

@@ -225,6 +225,18 @@ export function GetCandidateThumbnail(releaseMBID: string, releaseGroupMBID: str
     return $Call.ByID(1946932424, releaseMBID, releaseGroupMBID);
 }
 
+/**
+ * GetCredits is the bound form: the frontend asks for a tracklist's
+ * worth of MBIDs at once rather than one per row.
+ * 
+ * Batched for the reason every other per-row backend question here is:
+ * asking on hover or on render turns a list into N IPC round trips, and
+ * this one is asked about every row of every list in the app.
+ */
+export function GetCredits(mbids: string[] | null): $CancellablePromise<{ [_ in string]?: $models.CreditPart[] | null } | null> {
+    return $Call.ByID(225964099, mbids);
+}
+
 /**
  * GetExploreShelves builds the page Explore shows before a query.
  * 

frontend/src/components/cover-grid/cover-grid.ts

@@ -52,7 +52,8 @@ import {
 } from '@utils/context-menu-controller.js';
 import type { ContextMenuHost } from '@utils/context-menu-controller.js';
 import { FavoritesController } from '@store/controllers/favorites-controller';
-import { artistLink, exploreLinkStyles } from '../../utils/explore-link';
+import { creditLink, exploreLinkStyles } from '../../utils/explore-link';
+import { creditStore } from '@store/credit-store';
 import {
     createAlbumArtDragImage,
     createDragImage,
@@ -425,8 +426,18 @@ export class CoverGrid
      * Lifecycle
      * ==================================================================== */
 
+    /** Unsubscribes the credit-arrival repaint. */
+    private creditsUnsub?: () => void;
+
     override connectedCallback() {
         super.connectedCallback();
+
+        this.creditsUnsub = creditStore.subscribe(() => {
+            this.requestUpdate();
+            // Two virtualizers when the grid is split; both draw rows.
+            this.renderRoot?.querySelectorAll('lit-virtualizer')
+                .forEach((v) => (v as unknown as { requestUpdate(): void }).requestUpdate());
+        });
         this.restoreSortPreferences();
         this.loadAlbums();
 
@@ -441,6 +452,8 @@ export class CoverGrid
 
     override disconnectedCallback() {
         super.disconnectedCallback();
+        this.creditsUnsub?.();
+        this.creditsUnsub = undefined;
 
         this.removeEventListener(
             'error',
@@ -1820,7 +1833,7 @@ export class CoverGrid
                         class="artist-name"
                         title="${album.ArtistName}"
                     >
-                        ${artistLink(album.ArtistName, album.ArtistMBID ?? '')}
+                        ${creditLink(creditStore.credits(album.MBID), album.ArtistName, album.ArtistMBID ?? '')}
                     </div>
                 </div>
             </div>

frontend/src/components/explore-album-details/explore-album-details.ts

@@ -20,7 +20,8 @@ type MBRelease = explore.MBRelease;
 type MBTrack = explore.MBTrack;
 import { exploreCache } from '../../store/explore-cache';
 import { libraryStore } from '../../store/library-store';
-import { artistLink, exploreLinkStyles } from '../../utils/explore-link';
+import { creditLink, exploreLinkStyles } from '../../utils/explore-link';
+import { creditStore } from '@store/credit-store';
 import { describeError } from '../../utils/describe-error';
 import { EventsOn } from '@runtime/runtime';
 import { Events } from '../../events';
@@ -704,8 +705,15 @@ export class ExploreAlbumDetails extends LitElement implements ContextMenuHost {
      * BrowseReleases fetch never signals readiness. */
     private releasesFallbackTimer?: number;
 
+    /** Unsubscribes the credit-arrival repaint. */
+    private creditsUnsub?: () => void;
+
     override connectedCallback() {
         super.connectedCallback();
+
+        this.creditsUnsub = creditStore.subscribe(() => {
+            this.requestUpdate();
+        });
         if (this.releaseGroupMBID || this.localAlbumId) {
             void this.loadAllData();
         }
@@ -760,6 +768,8 @@ export class ExploreAlbumDetails extends LitElement implements ContextMenuHost {
 
     override disconnectedCallback() {
         super.disconnectedCallback();
+        this.creditsUnsub?.();
+        this.creditsUnsub = undefined;
         this.downloadUnsub?.();
         this.downloadUnsub = null;
         this.unsubReleasesReady?.();
@@ -2850,7 +2860,11 @@ export class ExploreAlbumDetails extends LitElement implements ContextMenuHost {
         return html`
             ${artist
                 ? html`<div class="album-artist">
-                      ${artistLink(artist, artistMbid)}
+                      ${creditLink(
+                          creditStore.credits(this.releaseGroupMBID),
+                          artist,
+                          artistMbid,
+                      )}
                   </div>`
                 : nothing}
             ${metaParts.length > 0

frontend/src/components/explore-view/explore-view.ts

@@ -16,7 +16,8 @@ import { exploreCache, ARTIST_IMAGE_CACHE_LIMIT } from '../../store/explore-cach
 import { queueStore } from '../../store/queue-store';
 import { notificationStore } from '../../store/notification-store';
 import '../notifications/inline-notice';
-import { artistLink, trackLink, exploreLinkStyles } from '../../utils/explore-link';
+import { creditLink, trackLink, exploreLinkStyles } from '../../utils/explore-link';
+import { creditStore } from '@store/credit-store';
 import { describeError } from '../../utils/describe-error';
 import '@awesome.me/webawesome/dist/components/icon/icon.js';
 import '../library-status-indicator/library-status-indicator.js';
@@ -774,6 +775,14 @@ export class ExploreView extends ViewLifecycleMixin(LitElement) implements Conte
     }
 
     protected override onViewActivate(): void {
+        // A cached primary view, so this is torn down on the way out
+        // rather than on disconnect — which never fires here.
+        this.whileActive(
+            creditStore.subscribe(() => {
+                this.requestUpdate();
+            }),
+        );
+
         // Fetched on arrival rather than on connect: this is a cached
         // primary view, created and warmed at startup, so a fetch there
         // is three catalog queries every user pays for whether or not
@@ -2193,7 +2202,7 @@ export class ExploreView extends ViewLifecycleMixin(LitElement) implements Conte
                                 <div class="album-title" title="${rg.title}">
                                     ${rg.title}
                                 </div>
-                                <div class="album-artist">${artistLink(rg.artistCredit, rg.artistMbid ?? '')}</div>
+                                <div class="album-artist">${creditLink(creditStore.credits(rg.mbid), rg.artistCredit, rg.artistMbid ?? '')}</div>
                                 <div class="album-meta">
                                     <div class="album-meta-text">
                                         ${rg.primaryType
@@ -2255,7 +2264,7 @@ export class ExploreView extends ViewLifecycleMixin(LitElement) implements Conte
                                         ${trackLink(r.title, r.releaseName ?? '', r.releaseGroupMbid ?? '', r.mbid)}
                                     </div>
                                     <div class="track-artist">
-                                        ${artistLink(r.artistCredit, r.artistMbid ?? '')}
+                                        ${creditLink(creditStore.credits(r.mbid), r.artistCredit, r.artistMbid ?? '')}
                                     </div>
                                 </div>
                                 <div class="track-meta">

frontend/src/components/now-playing-view/now-playing-view.ts

@@ -5,11 +5,12 @@ import '../audio-player/controls/player-controls';
 import '../audio-player/seekbar/seek-bar';
 import '../audio-player/volume-control/volume-control';
 import {
-    artistLink,
+    creditLink,
     albumLink,
     exploreLinkStyles,
 } from '@utils/explore-link';
 import { PlayerController } from '@store/controllers/player-controller';
+import { creditStore } from '@store/credit-store';
 import { FavoritesController } from '@store/controllers/favorites-controller';
 import { designTokens } from '../../styles/tokens.css';
 import { srOnly } from '../../styles/sr-only.css';
@@ -36,6 +37,22 @@ import { srOnly } from '../../styles/sr-only.css';
 @customElement('now-playing-view')
 export class NowPlayingView extends LitElement {
     private player = new PlayerController(this);
+
+    /** Unsubscribes the credit-arrival repaint. */
+    private creditsUnsub?: () => void;
+
+    override connectedCallback(): void {
+        super.connectedCallback();
+        // Credits arrive after the track does, so the name this view is
+        // already showing has to be re-rendered when they land.
+        this.creditsUnsub = creditStore.subscribe(() => this.requestUpdate());
+    }
+
+    override disconnectedCallback(): void {
+        super.disconnectedCallback();
+        this.creditsUnsub?.();
+        this.creditsUnsub = undefined;
+    }
     private favCtrl = new FavoritesController(this);
 
     static override styles = [designTokens, srOnly, exploreLinkStyles, css`
@@ -262,7 +279,11 @@ export class NowPlayingView extends LitElement {
                         ${track.title || track.fileName}
                     </h2>
                     <p class="artist">
-                        ${artistLink(track.artist, track.artistMbid)}
+                        ${creditLink(
+                            creditStore.credits(track.recordingMbid),
+                            track.artist,
+                            track.artistMbid,
+                        )}
                     </p>
                     ${track.album
                         ? html`<p class="album">

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

@@ -4,7 +4,7 @@ import '@awesome.me/webawesome/dist/components/icon/icon.js';
 import '@awesome.me/webawesome/dist/components/popup/popup.js';
 import type WaPopup from '@awesome.me/webawesome/dist/components/popup/popup.js';
 import {
-    artistLink,
+    creditLink,
     trackLink,
     exploreLinkStyles,
 } from '@utils/explore-link';
@@ -14,6 +14,7 @@ import {
     navigateToQueueSource,
 } from '@utils/queue-source-link';
 import { PlayerController } from '@store/controllers/player-controller';
+import { creditStore } from '@store/credit-store';
 import { QueueController } from '@store/controllers/queue-controller';
 import { FavoritesController } from '@store/controllers/favorites-controller';
 import { designTokens } from '../../styles/tokens.css';
@@ -325,6 +326,9 @@ export class NowPlaying extends LitElement {
     }
   `];
 
+    /** Unsubscribes the credit-arrival repaint. */
+    private creditsUnsub?: () => void;
+
     override connectedCallback() {
         super.connectedCallback();
         this.loadScrollMode();
@@ -341,10 +345,21 @@ export class NowPlaying extends LitElement {
             this.geometryDirty = true;
             this.requestUpdate();
         });
+
+        // A credit arriving changes the rendered text, and the marquee
+        // measures that text — so this is a geometry change, not just a
+        // repaint.  Saying so is what stops the bar scrolling to the
+        // old width.
+        this.creditsUnsub = creditStore.subscribe(() => {
+            this.geometryDirty = true;
+            this.requestUpdate();
+        });
     }
 
     override disconnectedCallback() {
         super.disconnectedCallback();
+        this.creditsUnsub?.();
+        this.creditsUnsub = undefined;
         // A drag interrupted by the bar going away still has to clean up.
         this.attachDragListeners(false);
         window.removeEventListener(SCROLL_CHANGE_EVENT, this.handleScrollModeEvent);
@@ -483,7 +498,7 @@ export class NowPlaying extends LitElement {
               @mouseleave=${this.handleArtistMouseLeave}
               @transitionend=${() => this.onScrollCycleEnd('artist')}
             >
-              <span class="scroll-content">${artistLink(track.artist, track.artistMbid) || 'Unknown Artist'}</span>
+              <span class="scroll-content">${creditLink(creditStore.credits(track.recordingMbid), track.artist, track.artistMbid) || 'Unknown Artist'}</span>
             </span>
             ${describeQueueSource(this.queue.source)
                 ? html`

frontend/src/components/playlist-details/playlist-details.ts

@@ -24,6 +24,7 @@ import type * as playlist from '@go/playlist/models.js';
 import { EventsOn } from '@runtime/runtime';
 import { Events } from '../../events';
 import { queueStore } from '@store/queue-store';
+import { creditStore } from '@store/credit-store';
 import { PlayerController } from '@store/controllers/player-controller';
 import { SearchController } from '@store/controllers/search-controller';
 import { SelectionController } from '@utils/selection-controller';
@@ -61,7 +62,8 @@ import '@components/duplicate-tracks-dialog/duplicate-tracks-dialog.js';
 import type { DuplicateTracksDialog } from '@components/duplicate-tracks-dialog/duplicate-tracks-dialog.js';
 import { formatMilliseconds } from '@utils/time';
 import {
-    artistLink,
+    creditLink,
+    creditText,
     albumLink,
     trackLink,
     exploreLinkStyles,
@@ -106,6 +108,9 @@ export class PlaylistDetails
      * rather than guessed.  Without the hint the flow layout's 100 px
      * default drives constant scroll-error correction, which reads as
      * the list jumping under the pointer. */
+    /** Unsubscribes the credit-arrival repaint. */
+    private creditsUnsub?: () => void;
+
     @query('lit-virtualizer')
     private virtualizer?: LitVirtualizer;
 
@@ -224,6 +229,14 @@ export class PlaylistDetails
 
     override connectedCallback() {
         super.connectedCallback();
+
+        // Credits arrive after the rows that asked for them, and a
+        // virtualizer repaints from its *own* properties — a host
+        // update alone leaves the rows exactly as they were.
+        this.creditsUnsub = creditStore.subscribe(() => {
+            this.requestUpdate();
+            this.virtualizer?.requestUpdate();
+        });
         this.loadTracks();
 
         this.tracksChangedCleanup = EventsOn(
@@ -248,6 +261,8 @@ export class PlaylistDetails
 
     override disconnectedCallback() {
         super.disconnectedCallback();
+        this.creditsUnsub?.();
+        this.creditsUnsub = undefined;
 
         if (this.tracksChangedCleanup) {
             this.tracksChangedCleanup();
@@ -1575,7 +1590,7 @@ export class PlaylistDetails
                                           : nothing}
                                   </div>
                                   <span class="cell col-title" title="${track.Title || track.FilePath}">${trackLink(track.Title, track.Album, track.ReleaseGroupMBID, track.RecordingMBID, undefined, track.Artist) || track.FilePath}</span>
-                                  <span class="cell col-artist" title="${track.Artist}">${artistLink(track.Artist, track.ArtistMBID)}</span>
+                                  <span class="cell col-artist" title="${creditText(creditStore.credits(track.RecordingMBID), track.Artist)}">${creditLink(creditStore.credits(track.RecordingMBID), track.Artist, track.ArtistMBID)}</span>
                                   <span class="cell col-album" title="${track.Album}">${albumLink(track.Album, track.ReleaseGroupMBID, undefined, track.Artist)}</span>
                                   <span class="cell col-duration">${formatMilliseconds(track.Duration)}</span>`}
                         </div>

frontend/src/components/queue-panel/queue-panel.ts

@@ -12,6 +12,7 @@ import '@awesome.me/webawesome/dist/components/popup/popup.js';
 import type WaPopup from '@awesome.me/webawesome/dist/components/popup/popup.js';
 import '@awesome.me/webawesome/dist/components/dropdown-item/dropdown-item.js';
 import { QueueController } from '@store/controllers/queue-controller';
+import { creditStore } from '@store/credit-store';
 import {
     describeQueueSource,
     isQueueSourceNavigable,
@@ -55,7 +56,7 @@ import { tracksByFilePath } from '@utils/track-index.js';
 import type { TrackDetails } from '@components/track-details/track-details.js';
 import type { CoverArtUrls } from '@components/track-details/track-details.js';
 import {
-    artistLink,
+    creditLink,
     trackLink,
     exploreLinkStyles,
 } from '@utils/explore-link';
@@ -108,6 +109,9 @@ export class QueuePanel
     @query('#playlist-submenu')
     private playlistSubmenuPopup!: WaPopup;
 
+    /** Unsubscribes the credit-arrival repaint. */
+    private creditsUnsub?: () => void;
+
     @query('lit-virtualizer')
     private virtualizer!: LitVirtualizer;
 
@@ -635,6 +639,14 @@ export class QueuePanel
 
     override connectedCallback() {
         super.connectedCallback();
+
+        // Credits arrive after the rows that asked for them, and a
+        // virtualizer repaints from its *own* properties — a host
+        // update alone leaves the rows exactly as they were.
+        this.creditsUnsub = creditStore.subscribe(() => {
+            this.requestUpdate();
+            this.virtualizer?.requestUpdate();
+        });
         this.style.setProperty(
             '--queue-width',
             `${this.panelWidth}px`,
@@ -671,6 +683,8 @@ export class QueuePanel
 
     override disconnectedCallback() {
         super.disconnectedCallback();
+        this.creditsUnsub?.();
+        this.creditsUnsub = undefined;
         document.removeEventListener(
             'mousemove',
             this.handleMouseMove,
@@ -1675,7 +1689,7 @@ export class QueuePanel
                         ${trackLink(title, track.album, track.releaseGroupMbid, track.recordingMbid, undefined, track.artist)}
                     </span>
                     <span class="track-artist" title=${artist}>
-                        ${artistLink(track.artist, track.artistMbid) || 'Unknown Artist'}
+                        ${creditLink(creditStore.credits(track.recordingMbid), track.artist, track.artistMbid) || 'Unknown Artist'}
                     </span>
                 </div>
                 <button

frontend/src/components/smart-playlist-details/smart-playlist-details.ts

@@ -15,6 +15,7 @@ import {
 import { EventsOn } from '@runtime/runtime';
 import { Events } from '../../events';
 import { queueStore } from '@store/queue-store';
+import { creditStore } from '@store/credit-store';
 import { PlayerController } from '@store/controllers/player-controller';
 import { SearchController } from '@store/controllers/search-controller';
 import { SelectionController } from '@utils/selection-controller';
@@ -51,7 +52,8 @@ import type { CoverArtUrls } from '@components/track-details/track-details.js';
 import { libraryStore } from '@store/library-store';
 import { formatMilliseconds } from '@utils/time';
 import {
-    artistLink,
+    creditLink,
+    creditText,
     albumLink,
     trackLink,
     exploreLinkStyles,
@@ -136,6 +138,9 @@ export class SmartPlaylistDetails
      * rather than guessed: without the hint the flow layout's 100 px
      * default drives constant scroll-error correction, which reads as
      * the list jumping under the pointer. */
+    /** Unsubscribes the credit-arrival repaint. */
+    private creditsUnsub?: () => void;
+
     @query('lit-virtualizer')
     private virtualizer?: LitVirtualizer;
 
@@ -609,6 +614,14 @@ export class SmartPlaylistDetails
     override connectedCallback() {
         super.connectedCallback();
 
+        // Credits arrive after the rows that asked for them, and a
+        // virtualizer repaints from its *own* properties — a host
+        // update alone leaves the rows exactly as they were.
+        this.creditsUnsub = creditStore.subscribe(() => {
+            this.requestUpdate();
+            this.virtualizer?.requestUpdate();
+        });
+
         if (this.autoEdit) {
             // Skip evaluation for new playlists — go straight to editor.
             this.autoEdit = false;
@@ -649,6 +662,8 @@ export class SmartPlaylistDetails
 
     override disconnectedCallback() {
         super.disconnectedCallback();
+        this.creditsUnsub?.();
+        this.creditsUnsub = undefined;
 
         if (this.playlistDeletedCleanup) {
             this.playlistDeletedCleanup();
@@ -1423,7 +1438,7 @@ export class SmartPlaylistDetails
                                           : nothing}
                                   </div>
                                   <span class="cell col-title" title="${track.Title || track.FilePath}">${trackLink(track.Title, track.Album, track.ReleaseGroupMBID, track.RecordingMBID, undefined, track.Artist) || track.FilePath}</span>
-                                  <span class="cell col-artist" title="${track.Artist}">${artistLink(track.Artist, track.ArtistMBID)}</span>
+                                  <span class="cell col-artist" title="${creditText(creditStore.credits(track.RecordingMBID), track.Artist)}">${creditLink(creditStore.credits(track.RecordingMBID), track.Artist, track.ArtistMBID)}</span>
                                   <span class="cell col-album" title="${track.Album}">${albumLink(track.Album, track.ReleaseGroupMBID, undefined, track.Artist)}</span>
                                   <span class="cell col-duration">${formatMilliseconds(track.Duration)}</span>`}
                         </div>

frontend/src/components/top-results-row/top-results-row.ts

@@ -9,7 +9,8 @@ import {
 } from '@go/explore/service.js';
 import '../library-status-indicator/library-status-indicator.js';
 import type { LibraryStatus } from '../library-status-indicator/library-status-indicator.js';
-import { artistLink, exploreLinkStyles } from '../../utils/explore-link';
+import { creditLink, exploreLinkStyles } from '../../utils/explore-link';
+import { creditStore } from '@store/credit-store';
 import { libraryStatusFor } from '../../utils/library-status';
 import { downloadStore } from '../../store/download-store';
 
@@ -61,14 +62,23 @@ export class TopResultsRow extends LitElement {
      * the property and never updates this element. One subscription for
      * the row, not one per card.
      */
+    /** Unsubscribes the credit-arrival repaint. */
+    private creditsUnsub?: () => void;
+
     override connectedCallback(): void {
         super.connectedCallback();
+
+        this.creditsUnsub = creditStore.subscribe(() => {
+            this.requestUpdate();
+        });
         this.unsubRequests = downloadStore.subscribe(() =>
             this.requestUpdate(),
         );
     }
 
     override disconnectedCallback(): void {
+        this.creditsUnsub?.();
+        this.creditsUnsub = undefined;
         this.unsubRequests?.();
         this.unsubRequests = undefined;
         super.disconnectedCallback();
@@ -327,7 +337,7 @@ export class TopResultsRow extends LitElement {
                         ${artistPart || metaPart
                             ? html`<span class="card-subtitle"
                                   >${artistPart
-                                      ? artistLink(artistPart, r.artistMbid ?? '')
+                                      ? creditLink(creditStore.credits(r.mbid), artistPart, r.artistMbid ?? '')
                                       : nothing}${artistPart && metaPart
                                       ? ' · '
                                       : ''}${metaPart}</span

frontend/src/components/track-list/track-list.ts

@@ -25,6 +25,7 @@ import type { SortOption } from '@components/page-header/page-header';
 import { TrackListController } from '@store/controllers/tracklist-controller';
 import { FavoritesController } from '@store/controllers/favorites-controller';
 import { queueStore } from '@store/queue-store';
+import { creditStore } from '@store/credit-store';
 import type { QueueSource } from '@store/queue-store';
 import { LibraryController } from '@store/controllers/library-controller';
 import {
@@ -40,6 +41,7 @@ import {
 } from './search-ranking';
 import {
     artistLink,
+    creditLink,
     albumLink,
     trackLink,
     exploreLinkStyles,
@@ -1339,6 +1341,17 @@ export class TrackList
             'shortcut:tracklist-delete',
             this.handleShortcutDelete,
         );
+
+        // Credits arrive after the rows that asked for them.  The
+        // virtualizer produces its rows from its *own* properties, so a
+        // host re-render alone repaints nothing — the same reason a
+        // selection change pushes requestUpdate() into it.
+        this.whileActive(
+            creditStore.subscribe(() => {
+                this.requestUpdate();
+                this.virtualizer?.requestUpdate();
+            }),
+        );
     }
 
     /**
@@ -2150,7 +2163,15 @@ export class TrackList
                 if (col.id === 'trackName') {
                     display = trackLink(track.TrackName, track.Album, track.ReleaseGroupMBID, track.RecordingMBID, display as any, track.ArtistName);
                 } else if (col.id === 'artistName') {
-                    display = artistLink(track.ArtistName, track.ArtistMBID, display as any);
+                    // A search term highlights the *flat* credit string,
+                    // and mapping those spans onto decomposed parts is a
+                    // different problem from rendering the credit.  While
+                    // filtering, the single link is the honest answer.
+                    creditStore.request(track.RecordingMBID);
+                    const parts = term ? undefined : creditStore.get(track.RecordingMBID);
+                    display = parts && parts.length > 1
+                        ? creditLink(parts, track.ArtistName, track.ArtistMBID)
+                        : artistLink(track.ArtistName, track.ArtistMBID, display as any);
                 } else if (col.id === 'album') {
                     display = albumLink(track.Album, track.ReleaseGroupMBID, display as any, track.ArtistName);
                 }

frontend/src/store/credit-store.ts

@@ -0,0 +1,226 @@
+/**
+ * Multi-artist credits, keyed by recording MBID.
+ *
+ * A credit is ordered parts and the credit *string* is derived from
+ * them.  This store holds the parts for entities that have more than
+ * one credited artist; everything else renders the single link it
+ * always did.
+ *
+ * Three things about it are load-bearing.
+ *
+ * **Absence is an answer, and it is cached as one.**  The backend
+ * returns nothing for a single-artist credit, which is the common case
+ * by a wide margin — measured on a real library, 13% of tracks are
+ * multi-artist.  Caching only the hits would re-request the other 87%
+ * on every render, forever, which is the same shape as the bug that
+ * made `explore-album-details` ask the backend on hover.  A miss is
+ * stored as an empty array: *asked*, not *answered*.
+ *
+ * **The lookup is batched, and coalesced across callers.**  Every row
+ * of every tracklist asks this question, and one IPC round trip per row
+ * is how a 5,000-row list becomes unusable.  A virtualized list cannot
+ * hand over "the whole list" either — 50,000 rows is 100 queries for
+ * the ~30 on screen.  So `request()` is per-row and cheap: it collects
+ * into a pending set and flushes once on the next frame, which turns a
+ * screenful of rows into exactly one call.  `ensure()` remains for a
+ * caller that genuinely has a bounded list in hand.
+ *
+ * **It is bounded.**  A cache that grows with use is a leak with a
+ * schedule; a browsing afternoon touches far more credits than a
+ * screenful.  The cap is entries rather than bytes because a credit is
+ * a handful of short strings, unlike the art caches next door.
+ */
+
+import { GetCredits } from '@go/explore/service.js';
+import type { CreditPart } from '../utils/explore-link';
+import { LRUMap } from '../utils/lru-map';
+import { compact } from '../utils/binding';
+import { registerCacheProbe } from '../utils/cache-stats';
+
+/**
+ * Entries retained.  A credit is ~4 short strings, so this is well
+ * under a megabyte — sized to comfortably exceed any single list the
+ * app renders, because a cap below the visible count evicts rows that
+ * are still on screen and the re-render fetches them straight back.
+ */
+export const CREDIT_CACHE_LIMIT = 20_000;
+
+/** An empty parts array is the negative marker: asked, no decomposition. */
+type CachedParts = readonly CreditPart[];
+
+class CreditStore {
+    private cache = new LRUMap<string, CachedParts>(CREDIT_CACHE_LIMIT);
+
+    /** MBIDs with a request in flight, so a re-render does not refetch. */
+    private inFlight = new Set<string>();
+
+    private listeners = new Set<() => void>();
+
+    /** Collected by request(), flushed as one batch on the next frame. */
+    private pending = new Set<string>();
+
+    private flushHandle: number | null = null;
+
+    constructor() {
+        registerCacheProbe('credits', () => ({
+            entries: this.cache.size,
+            chars: this.retainedChars(),
+            limit: CREDIT_CACHE_LIMIT,
+        }));
+    }
+
+    /**
+     * The strings actually retained, counted rather than estimated —
+     * a bound that is only checkable against a guess is not checkable.
+     */
+    private retainedChars(): number {
+        let total = 0;
+
+        for (const parts of this.cache.values()) {
+            for (const part of parts) {
+                total +=
+                    part.creditedName.length +
+                    part.joinPhrase.length +
+                    part.artistMbid.length;
+            }
+        }
+
+        return total;
+    }
+
+    /**
+     * Subscribe to "some credits arrived".
+     *
+     * Deliberately not per-MBID: a list fetches its rows in one call and
+     * re-renders once, so a fine-grained signal would buy nothing and
+     * cost a listener per row.
+     */
+    subscribe(fn: () => void): () => void {
+        this.listeners.add(fn);
+
+        return () => this.listeners.delete(fn);
+    }
+
+    /**
+     * The parts for one entity, or undefined when it has not been asked
+     * about yet.
+     *
+     * An entity with a single-artist credit returns an empty array, and
+     * `creditLink` treats fewer than two parts as the fallback — so a
+     * caller does not have to distinguish "not asked" from "one artist"
+     * to render correctly, only to decide whether to ask.
+     */
+    get(mbid: string | undefined): readonly CreditPart[] | undefined {
+        if (!mbid) return undefined;
+
+        return this.cache.get(mbid);
+    }
+
+    /**
+     * Ask about one entity, joining whatever batch is forming.
+     *
+     * Safe to call from a render: it is a set insert and a scheduled
+     * flush, and an entity already cached or in flight is dropped.  The
+     * loop it looks like it might cause does not happen — after a flush
+     * every requested MBID is cached, so the re-render's requests are
+     * all dropped and nothing notifies again.
+     */
+    request(mbid: string | undefined): void {
+        if (!mbid) return;
+        if (this.cache.has(mbid)) return;
+        if (this.inFlight.has(mbid)) return;
+        if (this.pending.has(mbid)) return;
+
+        this.pending.add(mbid);
+
+        if (this.flushHandle !== null) return;
+
+        // A frame, not a microtask: the point is to collect every row a
+        // virtualizer renders in this pass, and those happen across the
+        // whole update, not within one microtask checkpoint.
+        this.flushHandle = requestAnimationFrame(() => {
+            this.flushHandle = null;
+
+            const batch = [...this.pending];
+
+            this.pending.clear();
+
+            void this.ensure(batch);
+        });
+    }
+
+    /**
+     * Ask and read in one call, for use inside a template.
+     *
+     * A getter with a side effect, deliberately: the alternative is
+     * every call site writing `request(x)` beside `get(x)` and one of
+     * them eventually forgetting, which renders a permanently
+     * single-artist credit that looks exactly like an entity with one
+     * artist.  Making the request the same act as the read is what
+     * stops the two drifting apart.
+     */
+    credits(mbid: string | undefined): readonly CreditPart[] | undefined {
+        this.request(mbid);
+
+        return this.get(mbid);
+    }
+
+    /**
+     * Fetch the credits for a list, skipping anything already known or
+     * already being fetched.
+     *
+     * `has` rather than `get` for the membership test: probing must not
+     * mark an entry recently-used, or scrolling past a row would keep
+     * it alive ahead of one actually being rendered.
+     */
+    async ensure(mbids: readonly (string | undefined)[]): Promise<void> {
+        const wanted = new Set<string>();
+
+        for (const mbid of mbids) {
+            if (!mbid) continue;
+            if (this.cache.has(mbid)) continue;
+            if (this.inFlight.has(mbid)) continue;
+
+            wanted.add(mbid);
+        }
+
+        if (wanted.size === 0) return;
+
+        const batch = [...wanted];
+
+        for (const mbid of batch) this.inFlight.add(mbid);
+
+        try {
+            const found = compact(await GetCredits(batch));
+
+            for (const mbid of batch) {
+                // Every MBID asked for gets an entry, present or not:
+                // the absent ones are the answer "one artist", and not
+                // recording that is what would re-ask forever.
+                this.cache.set(mbid, found[mbid] ?? []);
+            }
+
+            this.notify();
+        } catch (err) {
+            // A credit is an enrichment: without it every name renders
+            // as the single link it did before, which is a worse answer
+            // rather than a broken one.  Nothing user-facing is worth
+            // interrupting for, so this stays in the console.
+            console.error('Failed to load artist credits', err);
+        } finally {
+            for (const mbid of batch) this.inFlight.delete(mbid);
+        }
+    }
+
+    /** Drop everything. The tags on disk changed, so credits may have. */
+    invalidate(): void {
+        this.cache = new LRUMap<string, CachedParts>(CREDIT_CACHE_LIMIT);
+        this.notify();
+    }
+
+    private notify(): void {
+        for (const fn of this.listeners) fn();
+    }
+}
+
+export const creditStore = new CreditStore();

frontend/src/utils/explore-link.ts

@@ -294,3 +294,81 @@ async function openAlbum(
 
     navigate(target, detail);
 }
+
+/**
+ * One credited artist within a multi-artist credit.
+ *
+ * Mirrors `artist_credit_part` / `file_artists`: the name **as
+ * credited** (which is not the artist's own name — MusicBrainz credits
+ * "Snoop Dogg" on a track by the artist called "Snoop Doggy Dogg"), the
+ * MBID to navigate to, and the literal connector that follows this
+ * part.
+ */
+export interface CreditPart {
+    /** The name as credited. Display uses this. */
+    creditedName: string;
+    /** The artist's MusicBrainz ID. Navigation uses this. */
+    artistMbid: string;
+    /** The connector following this part: " feat. ", " & ", ", ", "". */
+    joinPhrase: string;
+}
+
+/**
+ * Render a credit as links, one per credited artist, with the join
+ * phrases as plain text between them.
+ *
+ * Join phrases are **assembly instructions, not disassembly
+ * instructions**.  This concatenates parts; it never searches for a
+ * name inside a credit string.  That distinction is the whole point:
+ * the stored credit text may have come from a file's tags while the
+ * parts come from the catalog, and measured on a real library those
+ * disagree for about one in three multi-artist credits ("Skrillex
+ * feat. Swae Lee" tagged against "Skrillex & Swae Lee" upstream).  A
+ * search would miss, or match the wrong span.  Building from parts,
+ * the link boundaries are known by construction.
+ *
+ * Falls back to `artistLink(fallbackName, fallbackMbid)` — today's
+ * behaviour exactly — when there are no parts.  That is the common
+ * case and not a degraded one: a single-artist credit *is* one link,
+ * and a file with no recording MBID or no catalog row has nothing to
+ * decompose.  Do not try to split the fallback string; there is
+ * genuinely no information in it to split on.
+ *
+ * @param parts - The credit's parts in position order, if known.
+ * @param fallbackName - The credit as a single string.
+ * @param fallbackMbid - The primary artist's MBID.
+ */
+export function creditLink(
+    parts: readonly CreditPart[] | undefined,
+    fallbackName: string,
+    fallbackMbid: string,
+): TemplateResult | string {
+    // One part is one link, so it is the fallback rather than a special
+    // case — and a zero-part credit reaching here would otherwise
+    // render as nothing at all, which is worse than the single-artist
+    // answer it replaced.
+    if (!parts || parts.length < 2) {
+        return artistLink(fallbackName, fallbackMbid);
+    }
+
+    return html`${parts.map(
+        (part) =>
+            html`${artistLink(part.creditedName, part.artistMbid)}${part.joinPhrase}`,
+    )}`;
+}
+
+/**
+ * The plain-text form of a credit, for `title=` attributes and any
+ * other place that needs a string rather than a template.
+ *
+ * Rendered from the same parts by the same concatenation, so the
+ * tooltip cannot disagree with the links beneath it.
+ */
+export function creditText(
+    parts: readonly CreditPart[] | undefined,
+    fallbackName: string,
+): string {
+    if (!parts || parts.length < 2) return fallbackName;
+
+    return parts.map((p) => p.creditedName + p.joinPhrase).join('');
+}

frontend/test/utils/credit-link.test.ts

@@ -0,0 +1,121 @@
+/**
+ * A track credited to more than one artist has one navigable artist in
+ * this app and the rest are punctuation. `creditLink` is the fix: it
+ * renders a credit as one link per credited artist with the join
+ * phrases as plain text between them.
+ *
+ * The rule these tests exist to pin is that join phrases are
+ * **assembly** instructions, not disassembly instructions — the credit
+ * is built from its parts, never found by searching a name inside a
+ * credit string. Measured on a real library, the stored credit text and
+ * the catalog's parts disagree for about one in three multi-artist
+ * credits, so a search would miss or match the wrong span.
+ */
+import { describe, expect, it } from 'vitest';
+import { html, render } from 'lit';
+
+import { creditLink, creditText, type CreditPart } from '@utils/explore-link';
+
+const TUPAC = '11111111-1111-4111-8111-111111111111';
+const SNOOP = '22222222-2222-4222-8222-222222222222';
+
+const parts: CreditPart[] = [
+    { creditedName: '2Pac', artistMbid: TUPAC, joinPhrase: ' feat. ' },
+    { creditedName: 'Snoop Dogg', artistMbid: SNOOP, joinPhrase: '' },
+];
+
+function renderToEl(value: unknown): HTMLElement {
+    const host = document.createElement('div');
+    render(html`${value}`, host);
+
+    return host;
+}
+
+describe('creditLink', () => {
+    it('renders one link per credited artist', () => {
+        const el = renderToEl(creditLink(parts, '2Pac feat. Snoop Dogg', TUPAC));
+        const links = el.querySelectorAll('a.explore-link');
+
+        expect(links).toHaveLength(2);
+        expect(links[0]?.textContent).toBe('2Pac');
+        expect(links[1]?.textContent).toBe('Snoop Dogg');
+    });
+
+    it('puts the join phrase between the links as plain text', () => {
+        const el = renderToEl(creditLink(parts, '2Pac feat. Snoop Dogg', TUPAC));
+
+        // The whole credit reads correctly...
+        expect(el.textContent?.replace(/\s+/g, ' ').trim()).toBe(
+            '2Pac feat. Snoop Dogg',
+        );
+
+        // ...and " feat. " is not inside either link, which is the
+        // difference between a credit and a link with punctuation in it.
+        for (const link of el.querySelectorAll('a.explore-link')) {
+            expect(link.textContent).not.toMatch(/feat/);
+        }
+    });
+
+    it('falls back to a single link when there are no parts', () => {
+        const el = renderToEl(creditLink(undefined, 'Alina Baraz & Galimatias', TUPAC));
+        const links = el.querySelectorAll('a.explore-link');
+
+        expect(links).toHaveLength(1);
+        expect(links[0]?.textContent).toBe('Alina Baraz & Galimatias');
+    });
+
+    it('does not split the fallback string on its separators', () => {
+        // "&" and "with" appear inside real artist names — "Simon &
+        // Garfunkel" is one artist — so a credit with no parts is one
+        // link, always. This is the whole reason primaryArtist() does
+        // not split on them either.
+        const el = renderToEl(creditLink(undefined, 'Simon & Garfunkel', TUPAC));
+
+        expect(el.querySelectorAll('a.explore-link')).toHaveLength(1);
+    });
+
+    it('treats a one-part credit as the single-link case', () => {
+        // A zero- or one-part credit reaching the multi-artist branch
+        // would render as nothing, or as a link with a dangling join
+        // phrase after it.
+        const one: CreditPart[] = [
+            { creditedName: 'Solo', artistMbid: TUPAC, joinPhrase: '' },
+        ];
+        const el = renderToEl(creditLink(one, 'Solo', TUPAC));
+
+        expect(el.querySelectorAll('a.explore-link')).toHaveLength(1);
+        expect(el.textContent?.trim()).toBe('Solo');
+    });
+
+    it('renders the credited name, not the artist name', () => {
+        // MusicBrainz credits "Snoop Dogg" on a track by the artist
+        // called "Snoop Doggy Dogg". Display follows the credit;
+        // navigation follows the MBID.
+        const el = renderToEl(creditLink(parts, 'anything', TUPAC));
+
+        expect(el.textContent).toContain('Snoop Dogg');
+        expect(el.textContent).not.toContain('Snoop Doggy Dogg');
+    });
+});
+
+describe('creditText', () => {
+    it('reassembles the credit as a string', () => {
+        expect(creditText(parts, 'ignored')).toBe('2Pac feat. Snoop Dogg');
+    });
+
+    it('is the fallback string when there are no parts', () => {
+        expect(creditText(undefined, 'Alina Baraz & Galimatias')).toBe(
+            'Alina Baraz & Galimatias',
+        );
+    });
+
+    it('agrees with what creditLink renders', () => {
+        // The tooltip and the links come from the same parts by the same
+        // concatenation, so they cannot disagree.
+        const el = renderToEl(creditLink(parts, 'ignored', TUPAC));
+
+        expect(el.textContent?.replace(/\s+/g, ' ').trim()).toBe(
+            creditText(parts, 'ignored'),
+        );
+    });
+});

lefthook.yml

@@ -26,7 +26,11 @@ pre-commit:
         go generate ./...
         if [ -n "$(git diff --name-only)" ]; then
           echo "Generated code is out of date. Run 'make generate' and stage the changes."
-          git diff --stat
+          # --no-pager, or this blocks forever on `less` waiting for a
+          # keypress that a hook run without a tty will never get: the
+          # commit hangs at exactly the moment it is trying to tell you
+          # why it failed.
+          git --no-pager diff --stat
           exit 1
         fi
 

scripts/index-cache-snapshot.sh

@@ -0,0 +1,89 @@
+#!/bin/sh
+# Snapshot the index build's database, which is the only copy of it.
+#
+# `/srv/yellowjacket/index-cache` is the `YJ_HOME` the index job keeps
+# between runs (`.gitea/workflows/index-artifact.yml` mounts it at
+# `/cache`).  Its catalog is *derived*, not downloaded: the only way to
+# rebuild it is to re-stream the MetaBrainz dumps, which is hours at a
+# rate that is someone else's to decide.  On 2026-08-17 a schema repair
+# dropped it and cost exactly that.
+#
+# So it gets a snapshot, and this is the script a cron on that host runs.
+# It is deliberately not part of the workflow: a backup that only exists
+# while the thing it protects is being modified is not a backup.
+#
+# Usage (on the Gitea host):
+#
+#   scripts/index-cache-snapshot.sh [SOURCE_HOME] [DEST_DIR] [KEEP]
+#
+#   SOURCE_HOME  default /srv/yellowjacket/index-cache
+#   DEST_DIR     default /srv/yellowjacket/index-snapshots
+#   KEEP         how many to retain, default 2
+#
+# Suggested cron — daily, and nowhere near the Monday 04:00 build:
+#
+#   30 5 * * * /path/to/index-cache-snapshot.sh >> /var/log/yj-index-snapshot.log 2>&1
+#
+# Three things about it are load-bearing.
+#
+# **`VACUUM INTO`, not `cp`.** The database may be open, and a byte copy
+# of a live SQLite file is a corrupt file with a plausible size.
+# `VACUUM INTO` takes a read lock, writes a consistent compacted copy,
+# and is safe while the index job is running — it costs the snapshot's
+# own write, not the source's availability.
+#
+# **The staging directory is not copied.** `/cache/data/explore-staging`
+# is a resumable checkpoint of work in flight; it is large, it changes
+# constantly, and a build resumes without it. What cannot be re-derived
+# cheaply is the finished catalog, which is in the database.
+#
+# **A snapshot that is not verified is a belief.** Each one is opened
+# and asked for its catalog row count before the old ones are rotated
+# out, so a run that produced an unreadable file leaves the previous
+# good snapshot in place and fails loudly.
+set -eu
+
+SOURCE_HOME="${1:-/srv/yellowjacket/index-cache}"
+DEST_DIR="${2:-/srv/yellowjacket/index-snapshots}"
+KEEP="${3:-2}"
+
+DB="$SOURCE_HOME/data/yj.db"
+STAMP=$(date +%Y%m%d-%H%M%S)
+OUT="$DEST_DIR/yj-index-$STAMP.db"
+
+die() { echo "index-snapshot: $*" >&2; exit 1; }
+
+command -v sqlite3 >/dev/null 2>&1 || die "sqlite3 is not installed"
+[ -f "$DB" ] || die "no database at $DB (is SOURCE_HOME right?)"
+
+mkdir -p "$DEST_DIR"
+
+# Headroom: the copy is at most the size of the source, usually less
+# (VACUUM compacts).  Refusing here beats a half-written snapshot.
+need_kb=$(du -k "$DB" | cut -f1)
+free_kb=$(df -Pk "$DEST_DIR" | awk 'NR == 2 { print $4 }')
+[ "$free_kb" -gt "$need_kb" ] || die "not enough space in $DEST_DIR (need ~${need_kb}K, have ${free_kb}K)"
+
+# A failed snapshot must leave nothing behind. `VACUUM INTO` refuses an
+# existing file, so a partial one from a disk-full write would block
+# every later run -- and worse, rotation counts files by name, so it
+# would eventually be kept *instead of* a good one.
+cleanup() { [ -n "${KEPT:-}" ] || rm -f "$OUT"; }
+trap cleanup EXIT
+
+echo "index-snapshot: $DB -> $OUT"
+sqlite3 "$DB" "VACUUM INTO '$OUT'" || die "VACUUM INTO failed"
+
+rows=$(sqlite3 "$OUT" "SELECT count(*) FROM explore_index" 2>/dev/null) \
+	|| die "snapshot is unreadable: keeping the previous ones"
+[ "${rows:-0}" -gt 0 ] || die "snapshot has an empty catalog: keeping the previous ones"
+
+KEPT=1
+
+echo "index-snapshot: ok, $rows catalog rows, $(du -h "$OUT" | cut -f1)"
+
+# Rotate only after the new one has been verified.
+ls -1t "$DEST_DIR"/yj-index-*.db 2>/dev/null | tail -n +"$((KEEP + 1))" | while read -r old; do
+	echo "index-snapshot: removing $old"
+	rm -f "$old"
+done

scripts/skill-check.sh

@@ -1,17 +1,48 @@
 #!/usr/bin/env bash
 #
-# Every command in .pi/ is a `make` target on purpose: the Makefile is
-# the source of truth for *how* to invoke something, and the skill only
-# decides *which* and *in what order*.  This check keeps that honest —
-# a renamed or deleted target turns into a failing commit rather than
-# into an agent confidently running a command that no longer exists.
+# Every command in the agent-facing docs is a `make` target on purpose:
+# the Makefile is the source of truth for *how* to invoke something, and
+# the docs only decide *which* and *in what order*.  This check keeps
+# that honest — a renamed or deleted target turns into a failing commit
+# rather than into an agent confidently running a command that no longer
+# exists.
 #
-# It extracts every `make <target>` mentioned under .pi/ and asserts the
-# target exists.  Usage:  scripts/skill-check.sh
+# It checks two things.  Usage:  scripts/skill-check.sh
+#
+# **Every `make <target>` named in an agent-facing doc exists.**  The
+# scanned set is `.pi/` *and* CLAUDE.md, which is the half that was
+# missing: CLAUDE.md names 27 targets and nothing verified one of them,
+# so the file the agents trust most was the file least checked.
+#
+# **AGENTS.md is a symlink to CLAUDE.md.**  This repo is worked on by
+# two agent harnesses that read different files by convention — Claude
+# Code reads CLAUDE.md, others read AGENTS.md — and two harnesses
+# reading two descriptions of one project is how they come to hold
+# different beliefs about it.  A symlink makes that impossible by
+# construction; a *copy* would pass every other check in this repo while
+# silently drifting, which is exactly the failure being prevented, so
+# the symlink itself is asserted rather than its contents compared.
 set -euo pipefail
 
 cd "$(dirname "$0")/.."
 
+# The symlink half runs even without .pi/, since it is not about .pi/.
+if [ -e AGENTS.md ] || [ -L AGENTS.md ]; then
+	if [ ! -L AGENTS.md ]; then
+		echo "skill-check: AGENTS.md is a regular file, not a symlink to CLAUDE.md." >&2
+		echo "  Two harnesses would read two descriptions of one project." >&2
+		echo "  Fix:  rm AGENTS.md && ln -s CLAUDE.md AGENTS.md" >&2
+		exit 1
+	fi
+
+	target="$(readlink AGENTS.md)"
+
+	if [ "$target" != "CLAUDE.md" ]; then
+		echo "skill-check: AGENTS.md points at '$target', expected CLAUDE.md." >&2
+		exit 1
+	fi
+fi
+
 [ -d .pi ] || exit 0
 
 # `make -pq` prints the database including every rule, without running
@@ -21,11 +52,37 @@ targets="$({ make -pqRr 2>/dev/null || true; } |
 	awk '/^[a-zA-Z0-9][^$#\/\t=]*:([^=]|$)/ {sub(/:.*/, "", $0); print}' |
 	sort -u)"
 
-# A mention counts only when it is code: backticked (`make ui-test`) or
-# the first thing on a line, as in a fenced block.  Bare prose is not
-# scanned, because English says things like "a renamed make target".
-mentioned="$(grep -rhoE '(`|^)make [a-z][a-z0-9-]*' .pi --include='*.md' |
-	sed 's/^`//' | awk '{print $2}' | sort -u)"
+# A mention counts only when it is code: backticked (`make ui-test`)
+# anywhere, or at the start of a line **inside a fenced block**.  Bare
+# prose is not scanned, because English says things like "a renamed make
+# target".
+#
+# The fence is why this is awk rather than one grep.  Line-start alone is
+# not evidence of code in a file that is mostly hard-wrapped prose: the
+# sentence "Two green branches do not / make a green merge" wrapped onto
+# a line beginning `make a`, and the check duly failed on a target called
+# `a`.  Inside a fence it is code; outside one it is a sentence that
+# happened to break there, and a check that fails on reflow gets
+# disabled rather than fixed.
+#
+# AGENTS.md is deliberately not in this list: it is a symlink to
+# CLAUDE.md, asserted above, so scanning it would report every failure
+# twice under two names.
+mentioned="$({ find .pi -name '*.md' 2>/dev/null; echo CLAUDE.md; } |
+	xargs awk '
+		FNR == 1 { fence = 0 }
+		/^```/   { fence = !fence; next }
+		{
+			rest = $0
+			while (match(rest, /`make [a-z][a-z0-9-]*/)) {
+				print substr(rest, RSTART + 6, RLENGTH - 6)
+				rest = substr(rest, RSTART + RLENGTH)
+			}
+			if (fence && match($0, /^make [a-z][a-z0-9-]*/)) {
+				print substr($0, 6, RLENGTH - 5)
+			}
+		}
+	' | sort -u)"
 
 missing=""
 
@@ -36,10 +93,10 @@ for t in $mentioned; do
 done
 
 if [ -n "$missing" ]; then
-	echo "skill-check: .pi/ documents make targets that do not exist:" >&2
+	echo "skill-check: the agent docs name make targets that do not exist:" >&2
 	for t in $missing; do
 		echo "  make $t" >&2
-		grep -rln "make $t" .pi --include='*.md' | sed 's/^/      /' >&2
+		grep -rln "make $t" .pi CLAUDE.md --include='*.md' | sed 's/^/      /' >&2
 	done
 	echo "Fix the docs, or restore the target." >&2
 	exit 1