Merge pull request 'fix: look tags up by artist name rather than by MusicBrainz id' (#11) from fix/tag-lookup-by-name into main

Reviewed-on: #11

.gitea/workflows/build-and-publish.yaml

@@ -45,7 +45,8 @@ jobs:
 
       # The suite runs inside the image, against the interpreter that ships,
       # rather than against whatever the runner happens to provide. A failing
-      # test fails the build. Layers are shared with the push build below.
+      # test fails the build. The runtime stage below is built from the same
+      # daemon afterwards, so its layers are already in cache.
       - name: Run the test suite inside the image
         run: docker build --target test -t music-curator:test .
 
@@ -124,9 +125,6 @@ jobs:
           echo "release=${release}" >> "$GITHUB_OUTPUT"
           echo "Computed bump=${bump}, release=${release}, base=${base}"
 
-      - name: Set up Buildx
-        uses: docker/setup-buildx-action@d7f5e7f509e45cec5c76c4d5afdd7de93d0b3df5 # v4
-
       - name: Log in to the Gitea container registry
         if: github.event_name != 'pull_request'
         uses: docker/login-action@650006c6eb7dba73a995cc03b0b2d7f5ca915bee # v4
@@ -135,21 +133,33 @@ jobs:
           username: ${{ github.repository_owner }}
           password: ${{ secrets.PACKAGES_TOKEN }}
 
-      - name: Build and push
+      # Plain `docker build` rather than buildx. buildx boots its own buildkit
-        uses: docker/build-push-action@f9f3042f7e2789586610d6e8b85c8f03e5195baf # v7
+      # in a container with a cache of its own, so it shared nothing with the
-        with:
+      # test build above and rebuilt the image from the base image up -- two
-          context: .
+      # full builds per run. It earns that cost when building for several
-          # Without this the last stage in the Dockerfile -- the test stage --
+      # platforms; this only ever targets the amd64 NAS, so it does not.
-          # would be what gets published.
+      #
-          target: runtime
+      # `--target runtime` is a strict prefix of the test stage, so every layer
-          # The NAS is the only host this runs on. Building arm64 as well would
+      # is already in the daemon's cache and this resolves in seconds.
-          # mean emulating it under QEMU for no consumer.
+      - name: Build the runtime image
-          platforms: linux/amd64
+        run: |
-          push: ${{ github.event_name != 'pull_request' }}
+          set -euo pipefail
-          tags: ${{ steps.version.outputs.tags }}
+          tags=()
-          labels: |
+          while IFS= read -r tag; do
-            org.opencontainers.image.source=${{ github.server_url }}/${{ github.repository }}
+            [ -n "$tag" ] && tags+=(-t "$tag")
-            org.opencontainers.image.revision=${{ github.sha }}
+          done <<< "${{ steps.version.outputs.tags }}"
+          docker build --target runtime \
+            --label "org.opencontainers.image.source=${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}" \
+            --label "org.opencontainers.image.revision=${GITHUB_SHA}" \
+            "${tags[@]}" .
+
+      - name: Push
+        if: github.event_name != 'pull_request'
+        run: |
+          set -euo pipefail
+          while IFS= read -r tag; do
+            [ -n "$tag" ] && docker push "$tag"
+          done <<< "${{ steps.version.outputs.tags }}"
 
       # Record the release: write the computed version into pyproject.toml, then
       # commit and tag it, so the packaging metadata always matches the release

README.md

@@ -7,9 +7,11 @@ which keeps an MP3 copy of a lossless library for an iPod. This one answers the
 question that mirror cannot: which of it is worth carrying, and which of it has
 not been played in years.
 
-**This is stage two.** It ingests the scrobble history, indexes the library from
+**This is stage three.** It ingests the scrobble history, indexes the library
-Lidarr, and matches one to the other. There are no playlists yet, and it writes
+from Lidarr, matches one to the other, and writes playlists into the mirror —
-nothing back — every Lidarr call is a `GET`. See "Where this is going" below.
+by listening history and by mood.
+Nothing is written back to Lidarr — every call there is a `GET`. See "Where this
+is going" below.
 
 ## What it does today
 
@@ -18,6 +20,7 @@ nothing back — every Lidarr call is a `GET`. See "Where this is going" below.
 - Indexes every artist, album and track Lidarr knows about, with file paths and
   the date each file landed.
 - Ties the two together and reports how well it managed.
+- Writes M3U playlists into the mirror, from the listening history.
 
 ## Matching
 
@@ -109,10 +112,117 @@ matcher. The line to watch is:
 unmatched by an artist the library holds: N pairs, M plays
 ```
 
-That is a track that was played, sitting beside a file it should have matched.
+That is a track that was played by an artist the library holds. It is then
-Those are the matcher's real misses, and every one is a candidate for being
+split three ways, because owning an artist is a weak proxy for owning a track
-wrongly called cold in stage four. The report lists the worst fifteen by play
+and a shared title is a weak proxy for a shared song:
-count so they can be eyeballed.
+
+- **the library's own title credits the scrobbled artist** — `Voodoo People
+  (Pendulum Remix)` against a play credited to Pendulum. Same song, filed
+  under the original artist. These are the genuine misses.
+- **the same title under an unrelated artist** — a collision, not a miss.
+  Across fifty thousand tracks these are constant: `Everyday` is Rusko and
+  also Def Leppard, `Kaleidoscope` is Delta Heavy and also Chappell Roan.
+  Matching on title alone would be far worse than missing them, which is why
+  there is no such tier.
+- **the title is nowhere in the library** — never bought.
+
+Only the first is worth chasing. Counting all three as matcher failures
+overstates the problem and would over-block the cull.
+
+## Playlists
+
+Written into `<mirror>/_playlists/` as extended M3U, rebuilt every pass. Six
+rules, capped at `--playlist-limit` tracks each:
+
+| Playlist | Rule |
+| -------------------- | --------------------------------------------------------- |
+| `heavy-rotation` | Most played over the last twelve months |
+| `all-time` | Most played ever |
+| `neglected` | Played heavily once, silent for twelve months |
+| `deep-cuts` | Never played, from albums whose other tracks you play constantly |
+| `unheard-favourites` | Never played, by the artists you play most |
+| `unheard` | Never played, anywhere in the library |
+
+Ninety days was the obvious window for "recent" and is the wrong one: on a real
+history it holds a few hundred plays spread thinly across a twenty-thousand
+track rotation, so nothing ranks meaningfully. Twelve months does.
+
+The two `unheard` playlists rotate **weekly**, not per pass. A pass runs every
+few hours, and a playlist that reorders itself each time is one that has to be
+re-imported each time — the Music app imports a snapshot of a file, it does not
+track it.
+
+### Moods
+
+A second set of playlists selects by **Last.fm's crowd tags** rather than by
+listening history. MusicBrainz genres arrive free with the Lidarr index and are
+no use for this: they are sparse and formal, and will not tell you a record is
+screamo or synthwave. People typing tags will.
+
+Tags are fetched once per artist — one `artist.getTopTags` call each — and
+refreshed every ninety days. `--tag-limit` spreads the first sweep over several
+passes.
+
+Artists are looked up **by name**, not by MusicBrainz id, despite Lidarr having
+an id for every one of them. Last.fm's mbid index is stale and partial: it
+answers "the artist you supplied could not be found" for Devo, Escape the Fate,
+Blasterjaxx and a few hundred others whose pages plainly exist and carry exactly
+the tags wanted. Its name index is the one its own site runs on. The id is kept
+only as a fallback, for a name Lidarr spells differently.
+
+An artist neither key resolves is recorded as fetched with no tags, so the next
+pass does not spend a request on it again. A genuine failure — a rate limit, a
+bad key — is *not* recorded, so that one is retried.
+
+The built-in moods are chosen for this library rather than as a general
+taxonomy:
+
+| Mood | Selected on |
+| ------------------ | ------------------------------------------------- |
+| `80s-synths` | synthpop, new wave, synthwave — released 1975-1992 |
+| `high-energy-rock` | hard rock, punk, pop punk, alternative |
+| `screamo` | screamo, post-hardcore, metalcore, emo |
+| `drum-and-bass` | drum and bass, liquid funk, neurofunk, jungle |
+| `dance` | house, big room, hardstyle, trance, dubstep |
+| `classic-rock` | classic rock, prog, psychedelic, blues rock |
+
+An artist qualifies when their tag weights inside a mood sum to at least 30 out
+of Last.fm's 0-100 scale. One low-weight tag is not a genre, it is somebody's
+stray opinion.
+
+`years` filters on the album's release date, which is what separates eighties
+synth records from everything else a synthpop tag drags in.
+
+`--vibes` replaces the whole set with a JSON file of the same shape, so a new
+mood does not need a new release:
+
+```json
+[{ "name": "shoegaze", "tags": ["shoegaze", "dream pop"], "min_score": 40 }]
+```
+
+Names are validated when the file is read, not when the file is written. A bad
+one would otherwise surface as a playlist created somewhere unintended.
+
+### Paths
+
+Lidarr knows where the lossless source is; the playlists have to point at the
+MP3s music-mirror made from it. The mapping strips a library root from Lidarr's
+track paths and re-roots them under the mirror, with the suffix changed.
+
+`--library-root` is derived from the common parent of the indexed artist folders
+when unset, so it agrees with Lidarr by construction rather than by being kept
+in step by hand. Override it if that guess is wrong.
+
+Entries are written **relative to the playlist file**, so one playlist works
+from the NAS, from a Mac over SMB, and from Linux, without rewriting.
+
+A track is only listed once its mirror file has been confirmed to exist. Lidarr
+holding the FLAC says nothing about whether the MP3 has been encoded yet. If a
+large number are missing, the run says so — that is what a wrong `--library-root`
+or `--mirror` looks like, since the paths then map to nothing at all.
+
+`_playlists/` survives music-mirror's prune: it only deletes `*.mp3`, and its
+empty-directory sweep skips a directory holding M3Us.
 
 ## How the ingest works
 
@@ -168,6 +278,11 @@ music-curator --report-only   # report on the store, fetch nothing
 | `--backfill-limit` | `MUSIC_CURATOR_BACKFILL_LIMIT` | `0` | Cap backfill requests per pass; 0 for no cap |
 | `--lidarr-url` | `MUSIC_CURATOR_LIDARR_URL` | unset | Lidarr base URL, e.g. `http://lidarr:8686` |
 | `--lidarr-api-key` | `MUSIC_CURATOR_LIDARR_API_KEY` | unset | Lidarr API key |
+| `--mirror` | `MUSIC_CURATOR_MIRROR` | unset | Root of the MP3 mirror; playlists go here |
+| `--library-root` | `MUSIC_CURATOR_LIBRARY_ROOT` | derived | Prefix to strip from Lidarr's paths |
+| `--playlist-limit` | `MUSIC_CURATOR_PLAYLIST_LIMIT` | `100` | Most tracks in any one playlist |
+| `--vibes` | `MUSIC_CURATOR_VIBES` | built-in | JSON file of mood definitions |
+| `--tag-limit` | `MUSIC_CURATOR_TAG_LIMIT` | `0` | Cap artist tag lookups per pass |
 | `--skip-index` | — | off | Match against the index already held |
 | `--report-only` | — | off | Report without fetching |
 
@@ -216,7 +331,8 @@ nix shell nixpkgs#python3Packages.pytest -c pytest
 | ------------------------------------------------ | ------------ |
 | Last.fm ingest and store | done |
 | Lidarr index and the scrobble-to-track matcher | done |
-| M3U playlists written into the mirror | next |
+| M3U playlists from the listening history | done |
+| Genre and mood playlists from Last.fm tags | done |
 | Cold-music report, unmonitoring what is not played | last |
 
 The cull will unmonitor cold albums in Lidarr and tag their artists. It will

compose.yaml

@@ -30,5 +30,11 @@ services:
       # Cap the backfill at this many requests per pass. Unlimited by default,
       # which finishes a long history in one go.
       # MUSIC_CURATOR_BACKFILL_LIMIT: "0"
+      # The MP3 mirror music-mirror maintains. Playlists are written into
+      # _playlists/ inside it; leave unset to skip them.
+      MUSIC_CURATOR_MIRROR: /mirror
+      # Most tracks in any one playlist.
+      # MUSIC_CURATOR_PLAYLIST_LIMIT: "100"
     volumes:
       - /mnt/tank/apps/music-curator:/data
+      - /mnt/tank/media/music-mp3:/mirror

music_curator.py

@@ -26,6 +26,7 @@ import re
 import signal
 import sqlite3
 import sys
+import tempfile
 import time
 import unicodedata
 import urllib.error
@@ -62,6 +63,28 @@ REQUEST_DELAY_SECONDS = 0.25
 BACKOFF_SECONDS = 2.0
 BACKOFF_CEILING_SECONDS = 60.0
 
+# What music-mirror names its output, and where playlists are written inside the
+# mirror. music-mirror's prune only deletes `*.mp3` and only removes directories
+# it finds empty, so a directory of M3Us survives it untouched.
+MIRROR_SUFFIX = ".mp3"
+PLAYLIST_DIRECTORY = "_playlists"
+
+# Tags are re-fetched this often. They move slowly, and the first pass over a
+# library already costs one request per artist.
+TAG_REFRESH_SECONDS = 90 * 86400
+
+# How much of an artist's tag weight has to fall inside a vibe before their
+# tracks are eligible for it. Weights are Last.fm's own 0-100 popularity, summed
+# across whichever of the vibe's tags the artist carries.
+VIBE_MIN_SCORE = 30
+
+# A vibe writes a file named after itself, so the name has to be a filename.
+SAFE_VIBE_NAME = re.compile(r"^[a-z0-9][a-z0-9-]*$")
+
+# Group-readable, for the same reason music-mirror sets it: the mirror is read
+# back by whatever serves it, and a playlist nobody can read is not a playlist.
+GROUP_READ = 0o040
+
 SCHEMA_VERSION = "2"
 
 # Versions this build upgrades in place. Everything added since version 1 is a
@@ -141,6 +164,11 @@ CREATE TABLE IF NOT EXISTS lidarr_track (
 );
 CREATE INDEX IF NOT EXISTS lidarr_track_recording ON lidarr_track (recording_mbid);
 CREATE INDEX IF NOT EXISTS lidarr_track_norm ON lidarr_track (norm_artist, norm_title);
+-- Separate from the composite above, which a lookup by title alone cannot use:
+-- its leading column is the artist. The report searches by title on its own, and
+-- without this it scans every track for every unmatched key -- forty-seven
+-- seconds on a library of eighty-four thousand.
+CREATE INDEX IF NOT EXISTS lidarr_track_title ON lidarr_track (norm_title);
 CREATE INDEX IF NOT EXISTS lidarr_track_album ON lidarr_track (album_id);
 
 -- One row per distinct thing listened to, with the verdict on whether it could
@@ -161,6 +189,24 @@ CREATE TABLE IF NOT EXISTS scrobble_key (
     PRIMARY KEY (artist, track)
 );
 CREATE INDEX IF NOT EXISTS scrobble_key_track ON scrobble_key (track_id);
+
+-- Last.fm's crowd tags for each library artist. MusicBrainz genres come free
+-- with the Lidarr index and are useless for this: they are sparse and formal,
+-- and will not tell you a record is screamo or synthwave. People typing tags
+-- will. Keyed on the normalised name rather than the Lidarr id so it survives
+-- an artist being removed and re-added there.
+CREATE TABLE IF NOT EXISTS artist_tag (
+    norm_artist TEXT    NOT NULL,
+    tag         TEXT    NOT NULL,
+    weight      INTEGER NOT NULL,
+    PRIMARY KEY (norm_artist, tag)
+);
+CREATE INDEX IF NOT EXISTS artist_tag_tag ON artist_tag (tag);
+
+CREATE TABLE IF NOT EXISTS artist_tag_fetched (
+    norm_artist TEXT PRIMARY KEY,
+    fetched_at  INTEGER NOT NULL
+);
 """
 
 
@@ -220,7 +266,16 @@ WHITESPACE = re.compile(r"\s+")
 
 
 class LastfmError(Exception):
-    """A Last.fm request that failed in a way retrying will not fix."""
+    """A Last.fm request that failed in a way retrying will not fix.
+
+    `code` is the service's own error number where the failure came from the
+    API rather than the transport. Callers need it to tell "this thing does not
+    exist", which is final, from "something went wrong", which is not.
+    """
+
+    def __init__(self, message, code=None):
+        super().__init__(message)
+        self.code = code
 
 
 class LidarrError(Exception):
@@ -401,7 +456,7 @@ class Lastfm:
                 return payload
             detail = f"error {code}: {payload.get('message', '')}".strip()
             if code not in RETRYABLE_ERRORS:
-                raise LastfmError(f"{method}: {detail}")
+                raise LastfmError(f"{method}: {detail}", code=code)
             self._retry_or_raise(method, attempt, detail, None)
 
         raise LastfmError(f"{method}: gave up after {self.attempts} attempts")
@@ -530,6 +585,24 @@ class Store:
                 tracks,
             )
 
+    def replace_tags(self, norm_artist, pairs, now):
+        """Replace one artist's tags, and record that they were fetched.
+
+        The fetch is recorded even when nothing came back, so an artist Last.fm
+        has never heard of is not asked about again on every pass.
+        """
+        with self.connection:
+            self.connection.execute("DELETE FROM artist_tag WHERE norm_artist = ?", (norm_artist,))
+            self.connection.executemany(
+                "INSERT OR IGNORE INTO artist_tag (norm_artist, tag, weight) VALUES (?, ?, ?)",
+                [(norm_artist, tag, weight) for tag, weight in pairs],
+            )
+            self.connection.execute(
+                "INSERT INTO artist_tag_fetched (norm_artist, fetched_at) VALUES (?, ?)"
+                " ON CONFLICT (norm_artist) DO UPDATE SET fetched_at = excluded.fetched_at",
+                (norm_artist, now),
+            )
+
     def rebuild_keys(self):
         """Collapse the scrobble history into one row per distinct track.
 
@@ -1041,6 +1114,450 @@ def format_time(uts):
     return datetime.fromtimestamp(uts, tz=timezone.utc).strftime("%Y-%m-%d %H:%M")
 
 
+# Playlists written into the mirror. Each is a rule over the listening history
+# and the library index; none of them look at the audio.
+#
+# The rotating ones are shuffled by week rather than by pass. A pass runs every
+# few hours, and a playlist that reorders itself every time is one that has to
+# be re-imported every time -- the Music app does not track a file, it imports a
+# snapshot of one.
+ROTATION_PERIOD_SECONDS = 7 * 86400
+SHUFFLE_MULTIPLIER = 2654435761
+SHUFFLE_MODULUS = 104729
+
+# Common table expressions the rules share. `played` is every library track that
+# has ever been matched to a scrobble, with its count and the last time it was
+# heard; `recent` is the same restricted to a window.
+PLAYED_CTE = """
+WITH played AS (
+    SELECT sk.track_id AS track_id, COUNT(*) AS plays, MAX(s.uts) AS last_uts
+      FROM scrobble s
+      JOIN scrobble_key sk ON sk.artist = s.artist AND sk.track = s.track
+     WHERE sk.track_id IS NOT NULL
+     GROUP BY sk.track_id
+)
+"""
+RECENT_CTE = """
+WITH recent AS (
+    SELECT sk.track_id AS track_id, COUNT(*) AS plays
+      FROM scrobble s
+      JOIN scrobble_key sk ON sk.artist = s.artist AND sk.track = s.track
+     WHERE sk.track_id IS NOT NULL AND s.uts >= :year_ago
+     GROUP BY sk.track_id
+)
+"""
+SELECT_TRACK = """
+SELECT t.id AS id, a.name AS artist, t.title AS title,
+       t.duration AS duration, t.path AS path
+  FROM lidarr_track t
+  JOIN lidarr_artist a ON a.id = t.artist_id
+"""
+PLAYABLE = " t.has_file = 1 AND t.path IS NOT NULL AND t.path <> '' "
+SHUFFLE = f" ((t.id * {SHUFFLE_MULTIPLIER}) + :week) % {SHUFFLE_MODULUS} "
+
+PLAYLISTS = (
+    (
+        "heavy-rotation",
+        "most played over the last twelve months",
+        RECENT_CTE + SELECT_TRACK + f"""
+          JOIN recent r ON r.track_id = t.id
+         WHERE {PLAYABLE}
+         ORDER BY r.plays DESC, a.name, t.title
+         LIMIT :limit
+        """,
+    ),
+    (
+        "all-time",
+        "most played ever",
+        PLAYED_CTE + SELECT_TRACK + f"""
+          JOIN played p ON p.track_id = t.id
+         WHERE {PLAYABLE}
+         ORDER BY p.plays DESC, a.name, t.title
+         LIMIT :limit
+        """,
+    ),
+    (
+        "neglected",
+        "played heavily once, silent for twelve months",
+        PLAYED_CTE + SELECT_TRACK + f"""
+          JOIN played p ON p.track_id = t.id
+         WHERE {PLAYABLE} AND p.last_uts < :year_ago
+         ORDER BY p.plays DESC, a.name, t.title
+         LIMIT :limit
+        """,
+    ),
+    (
+        "deep-cuts",
+        "never played, from albums whose other tracks you play constantly",
+        PLAYED_CTE + """,
+        album_plays AS (
+            SELECT t.album_id AS album_id, SUM(p.plays) AS plays
+              FROM played p JOIN lidarr_track t ON t.id = p.track_id
+             GROUP BY t.album_id
+        )
+        """ + SELECT_TRACK + f"""
+          JOIN album_plays ap ON ap.album_id = t.album_id
+          LEFT JOIN played p ON p.track_id = t.id
+         WHERE {PLAYABLE} AND p.track_id IS NULL
+         ORDER BY ap.plays DESC, t.id
+         LIMIT :limit
+        """,
+    ),
+    (
+        "unheard-favourites",
+        "never played, by the artists you play most; rotates weekly",
+        PLAYED_CTE + """,
+        artist_plays AS (
+            SELECT t.artist_id AS artist_id, SUM(p.plays) AS plays
+              FROM played p JOIN lidarr_track t ON t.id = p.track_id
+             GROUP BY t.artist_id
+        )
+        """ + SELECT_TRACK + f"""
+          LEFT JOIN played p ON p.track_id = t.id
+         WHERE {PLAYABLE} AND p.track_id IS NULL
+           AND t.artist_id IN (SELECT artist_id FROM artist_plays
+                                ORDER BY plays DESC LIMIT 50)
+         ORDER BY {SHUFFLE}
+         LIMIT :limit
+        """,
+    ),
+    (
+        "unheard",
+        "never played, anywhere in the library; rotates weekly",
+        PLAYED_CTE + SELECT_TRACK + f"""
+          LEFT JOIN played p ON p.track_id = t.id
+         WHERE {PLAYABLE} AND p.track_id IS NULL
+         ORDER BY {SHUFFLE}
+         LIMIT :limit
+        """,
+    ),
+)
+
+
+# Moods, as sets of Last.fm tags. Chosen for this library -- drum and bass,
+# punk and its descendants, big-room dance, classic rock -- rather than as a
+# general taxonomy. `--vibes` replaces the lot with a JSON file of the same
+# shape, so a new one does not need a new release.
+#
+# `years` filters on the album's release date, which is what separates eighties
+# synth records from everything a synthpop tag would otherwise drag in.
+DEFAULT_VIBES = (
+    {
+        "name": "80s-synths",
+        "tags": [
+            "synthpop", "synth pop", "synth-pop", "new wave", "synthwave",
+            "new romantic", "electropop", "80s", "1980s",
+        ],
+        "years": [1975, 1992],
+    },
+    {
+        "name": "high-energy-rock",
+        "tags": [
+            "hard rock", "punk rock", "pop punk", "punk", "alternative rock",
+            "rock", "garage rock", "skate punk",
+        ],
+    },
+    {
+        "name": "screamo",
+        "tags": [
+            "screamo", "post-hardcore", "metalcore", "emo", "hardcore",
+            "melodic hardcore", "emocore",
+        ],
+    },
+    {
+        "name": "drum-and-bass",
+        "tags": [
+            "drum and bass", "drum n bass", "dnb", "liquid funk", "neurofunk",
+            "jungle", "breakbeat",
+        ],
+    },
+    {
+        "name": "dance",
+        "tags": [
+            "electro house", "house", "big room", "electronic dance music",
+            "edm", "hardstyle", "trance", "dubstep", "electro",
+        ],
+    },
+    {
+        "name": "classic-rock",
+        "tags": [
+            "classic rock", "progressive rock", "psychedelic rock",
+            "blues rock", "70s", "60s",
+        ],
+    },
+)
+
+
+def load_vibes(path):
+    """Return the vibe definitions, from a file when one is given.
+
+    Validated up front rather than at write time: a bad name would otherwise
+    surface as a file created somewhere unintended, which is a poor way to find
+    out about a typo.
+    """
+    if not path:
+        return DEFAULT_VIBES
+    try:
+        loaded = json.loads(Path(path).read_text(encoding="utf-8"))
+    except (OSError, json.JSONDecodeError) as error:
+        raise ValueError(f"cannot read vibes from {path}: {error}") from error
+
+    if not isinstance(loaded, list):
+        raise ValueError(f"{path}: expected a list of vibes")
+    for vibe in loaded:
+        name = vibe.get("name") if isinstance(vibe, dict) else None
+        if not name or not SAFE_VIBE_NAME.fullmatch(str(name)):
+            raise ValueError(f"{path}: {name!r} is not a usable vibe name (a-z, 0-9, -)")
+        if not vibe.get("tags"):
+            raise ValueError(f"{path}: vibe {name!r} lists no tags")
+    return tuple(loaded)
+
+
+def parse_tags(payload):
+    """Return (tag, weight) pairs from an artist.getTopTags response.
+
+    The documented sample carries only a name and a URL per tag; a live response
+    also carries a 0-100 `count`. Rather than depend on which, the count is used
+    when present and the documented ordering -- by popularity -- stands in for it
+    when it is not.
+    """
+    block = payload.get("toptags") or {}
+    pairs = []
+    for position, entry in enumerate(as_list(block.get("tag"))):
+        tag = (entry.get("name") or "").strip().casefold()
+        if not tag:
+            continue
+        weight = int(entry.get("count") or 0) or max(1, 100 - position * 5)
+        pairs.append((tag, weight))
+    return pairs
+
+
+def fetch_tags(client, name, mbid):
+    """Return an artist's tags, asking by name first.
+
+    By name, not by MusicBrainz id, despite Lidarr having an id for everything.
+    Last.fm's mbid index is stale and partial -- it answers "the artist you
+    supplied could not be found" for Devo, Escape the Fate and a few hundred
+    others whose pages plainly exist -- while the name index is the one its own
+    site runs on. The id is kept only as a fallback for a name Lidarr spells
+    differently.
+
+    Returns an empty list when the artist is genuinely unknown, which the caller
+    records so it is not asked again.
+    """
+    attempts = [{"artist": name}] if name else []
+    if mbid:
+        attempts.append({"mbid": mbid})
+
+    for query in attempts:
+        try:
+            return parse_tags(client.call("artist.getTopTags", {**query, "autocorrect": 1}))
+        except LastfmError as error:
+            # Error 6 here means "no such artist", which the next key may still
+            # answer. Anything else is a real failure and belongs to the caller.
+            if error.code != 6:
+                raise
+    return []
+
+
+def sync_tags(client, store, now, limit=0):
+    """Fetch crowd tags for library artists that have none, or stale ones.
+
+    One request per artist, once, and then only for artists newly added. An
+    artist Last.fm has never heard of is recorded as fetched with no tags, so it
+    is not asked about again every pass.
+    """
+    stale = store.connection.execute(
+        "SELECT a.norm_name AS norm_name, a.name AS name, a.mbid AS mbid"
+        "  FROM lidarr_artist a"
+        "  LEFT JOIN artist_tag_fetched f ON f.norm_artist = a.norm_name"
+        " WHERE a.norm_name <> ''"
+        "   AND (f.fetched_at IS NULL OR f.fetched_at < :cutoff)"
+        " ORDER BY a.name",
+        {"cutoff": now - TAG_REFRESH_SECONDS},
+    ).fetchall()
+    if not stale:
+        return 0
+
+    if limit > 0 and len(stale) > limit:
+        logger.info("tagging %d of %d artists this pass; the rest follow next", limit, len(stale))
+        stale = stale[:limit]
+    else:
+        logger.info("fetching tags for %d artists", len(stale))
+
+    resolved = 0
+    unknown = 0
+    for artist in stale:
+        try:
+            pairs = fetch_tags(client, artist["name"], artist["mbid"])
+        except LastfmError as error:
+            # Something went wrong rather than the artist not existing. Left
+            # unrecorded on purpose, so the next pass tries again.
+            logger.warning("no tags for %s: %s", artist["name"], error)
+            continue
+        store.replace_tags(artist["norm_name"], pairs, now)
+        resolved += 1
+        if not pairs:
+            unknown += 1
+
+    logger.info("tagged %d artists (%d with nothing to say about them)", resolved, unknown)
+    return resolved
+
+
+def build_vibe_playlists(store, vibes, mirror_root, library_root, limit, now):
+    """Write one playlist per mood. Returns how many tracks were listed."""
+    if not store.scalar("SELECT COUNT(*) FROM artist_tag"):
+        return 0
+
+    directory = Path(mirror_root) / PLAYLIST_DIRECTORY
+    week = now // ROTATION_PERIOD_SECONDS
+    total = 0
+
+    for vibe in vibes:
+        tags = [str(tag).strip().casefold() for tag in vibe["tags"]]
+        placeholders = ",".join("?" * len(tags))
+        years = vibe.get("years")
+        parameters = [*tags, vibe.get("min_score", VIBE_MIN_SCORE)]
+        year_clause = ""
+        if years:
+            year_clause = (
+                " AND CAST(substr(al.release_date, 1, 4) AS INTEGER) BETWEEN ? AND ?"
+            )
+            parameters += [int(years[0]), int(years[1])]
+        parameters += [week, limit]
+
+        sql = f"""
+            WITH vibe AS (
+                SELECT norm_artist, SUM(weight) AS score
+                  FROM artist_tag
+                 WHERE tag IN ({placeholders})
+                 GROUP BY norm_artist
+                HAVING SUM(weight) >= ?
+            )
+            SELECT t.id AS id, a.name AS artist, t.title AS title,
+                   t.duration AS duration, t.path AS path
+              FROM lidarr_track t
+              JOIN lidarr_artist a ON a.id = t.artist_id
+              JOIN vibe v ON v.norm_artist = a.norm_name
+              LEFT JOIN lidarr_album al ON al.id = t.album_id
+             WHERE {PLAYABLE}{year_clause}
+             ORDER BY ((t.id * {SHUFFLE_MULTIPLIER}) + ?) % {SHUFFLE_MODULUS}
+             LIMIT ?
+        """
+
+        entries = []
+        for row in store.connection.execute(sql, parameters):
+            mirror = mirror_path_for(row["path"], library_root, mirror_root)
+            if mirror is None or not mirror.is_file():
+                continue
+            entries.append({**dict(row), "mirror": mirror})
+
+        write_playlist(directory / f"{vibe['name']}.m3u", entries)
+        total += len(entries)
+        logger.info("playlist %-20s %4d tracks -- by tag", vibe["name"], len(entries))
+
+    return total
+
+
+def library_root_of(store):
+    """Return the directory Lidarr's artist folders sit under.
+
+    Derived rather than configured, because it has to agree with what Lidarr
+    reports and no one wants to keep a second copy of that in step by hand.
+    """
+    paths = [
+        row["path"]
+        for row in store.connection.execute(
+            "SELECT path FROM lidarr_artist WHERE path IS NOT NULL AND path <> ''"
+        )
+    ]
+    if not paths:
+        return None
+    if len(paths) == 1:
+        return str(Path(paths[0]).parent)
+    try:
+        return os.path.commonpath(paths)
+    except ValueError:
+        return None
+
+
+def mirror_path_for(source, library_root, mirror_root):
+    """Return where music-mirror would have put the MP3 for a source file."""
+    try:
+        relative = Path(source).relative_to(library_root)
+    except ValueError:
+        return None
+    return (Path(mirror_root) / relative).with_suffix(MIRROR_SUFFIX)
+
+
+def write_playlist(path, entries):
+    """Write one extended M3U, atomically.
+
+    Paths are relative to the playlist file, so the same playlist works from the
+    NAS, from a Mac over SMB and from Linux without rewriting.
+    """
+    lines = ["#EXTM3U"]
+    for entry in entries:
+        seconds = round((entry["duration"] or 0) / 1000)
+        lines.append(f"#EXTINF:{seconds},{entry['artist']} - {entry['title']}")
+        lines.append(os.path.relpath(entry["mirror"], path.parent))
+
+    path.parent.mkdir(parents=True, exist_ok=True)
+    handle, temporary = tempfile.mkstemp(dir=path.parent, suffix=".m3u.part")
+    os.close(handle)
+    temporary = Path(temporary)
+    try:
+        temporary.write_text("\n".join(lines) + "\n", encoding="utf-8")
+        # The mirror is read back by something else; see music-mirror, which had
+        # to learn this the hard way.
+        mode = temporary.stat().st_mode
+        if not mode & GROUP_READ:
+            temporary.chmod(mode | GROUP_READ)
+        os.replace(temporary, path)
+    finally:
+        temporary.unlink(missing_ok=True)
+
+
+def build_playlists(store, mirror_root, library_root, limit, now):
+    """Write every playlist into the mirror. Returns how many tracks were listed.
+
+    A track is only listed once its mirror file has been confirmed to exist. The
+    index knows what Lidarr holds, which is the lossless source; whether the MP3
+    beside it has been encoded yet is music-mirror's business and is checked
+    rather than assumed.
+    """
+    directory = Path(mirror_root) / PLAYLIST_DIRECTORY
+    parameters = {
+        "limit": limit,
+        "year_ago": now - 365 * 86400,
+        "week": now // ROTATION_PERIOD_SECONDS,
+    }
+
+    total = 0
+    missing = 0
+    for name, description, sql in PLAYLISTS:
+        entries = []
+        for row in store.connection.execute(sql, parameters):
+            mirror = mirror_path_for(row["path"], library_root, mirror_root)
+            if mirror is None or not mirror.is_file():
+                missing += 1
+                continue
+            entries.append({**dict(row), "mirror": mirror})
+        write_playlist(directory / f"{name}.m3u", entries)
+        total += len(entries)
+        logger.info("playlist %-20s %4d tracks -- %s", name, len(entries), description)
+
+    if missing:
+        logger.warning(
+            "%d selected tracks had no file in the mirror and were left out."
+            " A handful means music-mirror has not encoded them yet; a large"
+            " number means --library-root or --mirror is pointing at the wrong"
+            " place, since the paths are then being mapped to nothing.",
+            missing,
+        )
+    return total
+
+
 def coverage_report(store):
     """Log how much of the listening history could be tied to the library.
 
@@ -1119,12 +1636,76 @@ def coverage_report(store):
         "   AND EXISTS (SELECT 1 FROM lidarr_artist a WHERE a.norm_name = k.norm_artist)"
     ).fetchone()
     logger.info(
-        "unmatched by an artist the library holds: %d pairs, %d plays -- these are the"
+        "unmatched by an artist the library holds: %d pairs, %d plays",
-        " matcher's misses, not music you do not own",
         suspect["pairs"],
         suspect["plays"],
     )
 
+    # Owning an artist is a weak proxy for owning a track, so that figure alone
+    # overstates the matcher's failings. Split it -- but not on the title alone.
+    # Across fifty thousand tracks, titles collide constantly: "Everyday" is
+    # Rusko and also Def Leppard, "Kaleidoscope" is Delta Heavy and also
+    # Chappell Roan. Matching those would be worse than missing them.
+    #
+    # The signal for a genuine attribution miss is that the library's own title
+    # credits the artist the scrobble is filed under -- "Voodoo People (Pendulum
+    # Remix)" against a play credited to Pendulum. The normalised title has that
+    # suffix stripped, which is exactly what let them meet, so the raw one has to
+    # be searched for the name.
+    attribution = store.connection.execute(
+        "SELECT COUNT(*) AS pairs, COALESCE(SUM(plays), 0) AS plays FROM scrobble_key k"
+        " WHERE k.track_id IS NULL"
+        "   AND EXISTS (SELECT 1 FROM lidarr_artist a WHERE a.norm_name = k.norm_artist)"
+        "   AND EXISTS (SELECT 1 FROM lidarr_track t"
+        "                WHERE t.norm_title = k.norm_track"
+        "                  AND instr(lower(t.title), lower(k.artist)) > 0)"
+    ).fetchone()
+    collision = store.connection.execute(
+        "SELECT COUNT(*) AS pairs, COALESCE(SUM(plays), 0) AS plays FROM scrobble_key k"
+        " WHERE k.track_id IS NULL"
+        "   AND EXISTS (SELECT 1 FROM lidarr_artist a WHERE a.norm_name = k.norm_artist)"
+        "   AND EXISTS (SELECT 1 FROM lidarr_track t WHERE t.norm_title = k.norm_track)"
+    ).fetchone()
+    logger.info(
+        "  the library's title credits the scrobbled artist: %d pairs, %d plays"
+        " -- remixes and guest spots, and the genuine misses",
+        attribution["pairs"],
+        attribution["plays"],
+    )
+    logger.info(
+        "  same title under an unrelated artist: %d pairs, %d plays"
+        " -- title collisions, not misses; matching these would be a mistake",
+        collision["pairs"] - attribution["pairs"],
+        collision["plays"] - attribution["plays"],
+    )
+    logger.info(
+        "  the rest, %d pairs, %d plays: you own the artist but not the track",
+        suspect["pairs"] - collision["pairs"],
+        suspect["plays"] - collision["plays"],
+    )
+
+    mismatched = store.connection.execute(
+        "SELECT k.artist, k.track, k.plays,"
+        "       (SELECT t.title FROM lidarr_track t"
+        "         WHERE t.norm_title = k.norm_track"
+        "           AND instr(lower(t.title), lower(k.artist)) > 0 LIMIT 1) AS library_title"
+        "  FROM scrobble_key k"
+        " WHERE k.track_id IS NULL"
+        "   AND EXISTS (SELECT 1 FROM lidarr_artist a WHERE a.norm_name = k.norm_artist)"
+        "   AND EXISTS (SELECT 1 FROM lidarr_track t"
+        "                WHERE t.norm_title = k.norm_track"
+        "                  AND instr(lower(t.title), lower(k.artist)) > 0)"
+        " ORDER BY k.plays DESC, k.artist LIMIT 10"
+    ).fetchall()
+    for position, row in enumerate(mismatched, start=1):
+        logger.info(
+            "  attribution %2d: %-45s %4d plays, library has %r",
+            position,
+            f"{row['artist']} - {row['track']}"[:45],
+            row["plays"],
+            row["library_title"],
+        )
+
     with_files = store.scalar("SELECT COUNT(*) FROM lidarr_track WHERE has_file = 1")
     played = store.scalar(
         "SELECT COUNT(DISTINCT k.track_id) FROM scrobble_key k"
@@ -1193,7 +1774,10 @@ def report(store, now):
         logger.info("  top track 90d %2d: %-50s %d", position, label[:50], row["plays"])
 
 
-def run_once(client, store, user, now, backfill_limit, lidarr=None):
+def run_once(
+    client, store, user, now, backfill_limit, lidarr=None, mirror=None,
+    library_root=None, playlist_limit=100, vibes=DEFAULT_VIBES, tag_limit=0,
+):
     """Run a single pass. Returns the number of scrobbles added."""
     started = time.monotonic()
     added = catch_up(client, store, user, now)
@@ -1206,6 +1790,17 @@ def run_once(client, store, user, now, backfill_limit, lidarr=None):
     # arrived, and they need a verdict too.
     if store.scalar("SELECT COUNT(*) FROM lidarr_track"):
         match_library(store)
+        if mirror is not None:
+            root = library_root or library_root_of(store)
+            if root is None:
+                logger.warning(
+                    "cannot work out where Lidarr's music lives, so no playlists were"
+                    " written; set --library-root"
+                )
+            else:
+                build_playlists(store, mirror, root, playlist_limit, now)
+                sync_tags(client, store, now, limit=tag_limit)
+                build_vibe_playlists(store, vibes, mirror, root, playlist_limit, now)
 
     logger.info(
         "pass complete in %.1fs: %d scrobbles added, %d loved",
@@ -1278,6 +1873,37 @@ def build_parser():
         default=os.getenv("MUSIC_CURATOR_LIDARR_API_KEY"),
         help="Lidarr API key (env MUSIC_CURATOR_LIDARR_API_KEY)",
     )
+    parser.add_argument(
+        "--mirror",
+        default=os.getenv("MUSIC_CURATOR_MIRROR"),
+        help="root of the MP3 mirror; playlists are written into it"
+        " (env MUSIC_CURATOR_MIRROR)",
+    )
+    parser.add_argument(
+        "--library-root",
+        default=os.getenv("MUSIC_CURATOR_LIBRARY_ROOT"),
+        help="prefix to strip from Lidarr's track paths when mapping them into the"
+        " mirror; derived from the indexed artist folders when unset"
+        " (env MUSIC_CURATOR_LIBRARY_ROOT)",
+    )
+    parser.add_argument(
+        "--playlist-limit",
+        type=int,
+        default=int(os.getenv("MUSIC_CURATOR_PLAYLIST_LIMIT", "100")),
+        help="most tracks to put in any one playlist (env MUSIC_CURATOR_PLAYLIST_LIMIT)",
+    )
+    parser.add_argument(
+        "--vibes",
+        default=os.getenv("MUSIC_CURATOR_VIBES"),
+        help="JSON file of mood definitions, replacing the built-in set"
+        " (env MUSIC_CURATOR_VIBES)",
+    )
+    parser.add_argument(
+        "--tag-limit",
+        type=int,
+        default=int(os.getenv("MUSIC_CURATOR_TAG_LIMIT", "0")),
+        help="cap artist tag lookups per pass; 0 for no cap (env MUSIC_CURATOR_TAG_LIMIT)",
+    )
     parser.add_argument(
         "--skip-index",
         action="store_true",
@@ -1302,6 +1928,7 @@ def main(argv=None, clock=time.time):
 
     try:
         interval = parse_interval(args.interval) if args.interval else None
+        vibes = load_vibes(args.vibes)
     except ValueError as error:
         logger.error("%s", error)
         return 2
@@ -1342,7 +1969,19 @@ def main(argv=None, clock=time.time):
         while True:
             now = int(clock())
             try:
-                run_once(client, store, args.user, now, args.backfill_limit, lidarr)
+                run_once(
+                    client,
+                    store,
+                    args.user,
+                    now,
+                    args.backfill_limit,
+                    lidarr,
+                    args.mirror,
+                    args.library_root,
+                    args.playlist_limit,
+                    vibes,
+                    args.tag_limit,
+                )
             except (LastfmError, LidarrError) as error:
                 logger.error("%s", error)
                 if interval is None:

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "music-curator"
-version = "0.2.4"
+version = "0.5.0"
 description = "Ingest a Last.fm listening history and curate a music library from it"
 readme = "README.md"
 requires-python = ">=3.11"

tests/conftest.py

@@ -56,7 +56,9 @@ class FakeLastfm:
     is prepended to the first page with no `date`.
     """
 
-    def __init__(self, tracks=(), loved=(), nowplaying=None, outcomes=()):
+    def __init__(self, tracks=(), loved=(), nowplaying=None, outcomes=(), tags=None):
+        # Keyed by mbid or by artist name, whichever the caller asked with.
+        self.tags = tags or {}
         self.tracks = sorted(tracks, key=lambda track: int(track["date"]["uts"]), reverse=True)
         self.loved = list(loved)
         self.nowplaying = nowplaying
@@ -82,6 +84,15 @@ class FakeLastfm:
             return json.dumps(self._recent(query))
         if method == "user.getlovedtracks":
             return json.dumps(self._loved(query))
+        if method == "artist.gettoptags":
+            key = query.get("mbid") or query.get("artist", "")
+            if key not in self.tags:
+                # What the real service says for a key it cannot resolve, which
+                # for mbids is a great many artists whose pages plainly exist.
+                return json.dumps(
+                    {"error": 6, "message": "The artist you supplied could not be found"}
+                )
+            return json.dumps({"toptags": {"tag": self.tags[key]}})
         raise AssertionError(f"unexpected method {method}")
 
     def _recent(self, query):

tests/test_music_curator.py

@@ -1,5 +1,7 @@
 import json
+import stat
 import urllib.error
+from pathlib import Path
 
 import pytest
 from conftest import FakeLastfm, FakeLidarr, make_loved, make_tracks
@@ -55,9 +57,23 @@ LIBRARY = [
     {
         "name": "The Prodigy",
         "albums": [
-            {"title": "The Fat of the Land", "tracks": [{"title": "Breathe (Remastered)"}]}
+            {
+                "title": "The Fat of the Land",
+                "tracks": [
+                    {"title": "Breathe (Remastered)"},
+                    # Credits its remixer in the title, which is the only signal
+                    # separating a real attribution miss from a title collision.
+                    {"title": "Voodoo People (Pendulum Remix)"},
+                ],
+            }
         ],
     },
+    # Held by the library in its own right, which is what puts its scrobbles
+    # inside the "artist the library holds" filter at all.
+    {
+        "name": "Pendulum",
+        "albums": [{"title": "Immersion", "tracks": [{"title": "Watercolour"}]}],
+    },
 ]
 
 
@@ -393,7 +409,7 @@ def test_keys_carry_the_play_count_and_the_span(tmp_path):
 def test_re_indexing_drops_what_lidarr_no_longer_has(tmp_path):
     """The index is Lidarr's mirror, not an accumulation of everything ever seen."""
     store = indexed(tmp_path, [scrobble_of("AC/DC", "Hells Bells")])
-    assert store.scalar("SELECT COUNT(*) FROM lidarr_artist") == 3
+    assert store.scalar("SELECT COUNT(*) FROM lidarr_artist") == 4
 
     music_curator.index_library(
         music_curator.Lidarr("http://lidarr", "key", transport=FakeLidarr(LIBRARY[:1])), store
@@ -512,7 +528,7 @@ def test_albums_come_from_the_unfiltered_endpoint(tmp_path):
 
     album_calls = [query for path, query in api.calls if path == "album"]
     assert album_calls == [{}]
-    assert store.scalar("SELECT COUNT(*) FROM lidarr_album") == 3
+    assert store.scalar("SELECT COUNT(*) FROM lidarr_album") == 4
 
 
 def test_a_bad_album_falls_back_to_asking_per_artist(tmp_path):
@@ -526,7 +542,7 @@ def test_a_bad_album_falls_back_to_asking_per_artist(tmp_path):
 
     assert store.scalar("SELECT COUNT(*) FROM lidarr_album WHERE artist_id = 2") == 0
     # The other two artists keep their albums.
-    assert store.scalar("SELECT COUNT(*) FROM lidarr_album") == 2
+    assert store.scalar("SELECT COUNT(*) FROM lidarr_album") == 3
     assert store.get_state("index_albums_skipped") == "1"
 
 
@@ -549,10 +565,10 @@ def test_an_artist_lidarr_cannot_serve_does_not_kill_the_index(tmp_path):
 
     music_curator.index_library(music_curator.Lidarr("http://lidarr", "key", transport=api), store)
 
-    assert store.scalar("SELECT COUNT(*) FROM lidarr_artist") == 3
+    assert store.scalar("SELECT COUNT(*) FROM lidarr_artist") == 4
     assert store.scalar("SELECT COUNT(*) FROM lidarr_track WHERE artist_id = 2") == 0
     # The other two artists are indexed in full.
-    assert store.scalar("SELECT COUNT(*) FROM lidarr_track") == 3
+    assert store.scalar("SELECT COUNT(*) FROM lidarr_track") == 5
     assert store.get_state("index_skipped") == "1"
 
 
@@ -732,3 +748,453 @@ def test_a_bracketed_guest_credit_matches_the_bare_tag(tmp_path):
     method, track_id = verdict(store, "Yellowcard", "Here I Am Alive (feat. Someone)")
     assert method == "name"
     assert track_id is not None
+
+
+def test_misses_are_split_by_whether_the_library_holds_the_title(tmp_path):
+    """Owning an artist is a weak proxy for owning a track. Counting both as
+    matcher failures overstates the problem and would over-block the cull."""
+    store = indexed(
+        tmp_path,
+        [
+            # The library holds "Hells Bells", but under AC/DC, not Yellowcard:
+            # an attribution disagreement, and a real miss.
+            scrobble_of("Yellowcard", "Hells Bells"),
+            # Yellowcard is in the library; this track is not, under any artist.
+            scrobble_of("Yellowcard", "A Single She Never Bought"),
+        ],
+    )
+
+    def count(extra):
+        return store.connection.execute(
+            "SELECT COUNT(*) FROM scrobble_key k WHERE k.track_id IS NULL"
+            "   AND EXISTS (SELECT 1 FROM lidarr_artist a WHERE a.norm_name = k.norm_artist)"
+            f"  {extra}"
+        ).fetchone()[0]
+
+    assert count("") == 2
+    assert count("AND EXISTS (SELECT 1 FROM lidarr_track t WHERE t.norm_title = k.norm_track)") == 1
+
+
+def test_the_report_survives_the_attribution_split(tmp_path):
+    store = indexed(tmp_path, [scrobble_of("Yellowcard", "Hells Bells")])
+
+    music_curator.report(store, NOW)
+
+
+def attribution_pairs(store):
+    """Unmatched pairs where the library's own title credits the scrobbled artist."""
+    return [
+        row["artist"]
+        for row in store.connection.execute(
+            "SELECT k.artist FROM scrobble_key k WHERE k.track_id IS NULL"
+            "   AND EXISTS (SELECT 1 FROM lidarr_artist a WHERE a.norm_name = k.norm_artist)"
+            "   AND EXISTS (SELECT 1 FROM lidarr_track t"
+            "                WHERE t.norm_title = k.norm_track"
+            "                  AND instr(lower(t.title), lower(k.artist)) > 0)"
+        )
+    ]
+
+
+def test_a_shared_title_is_not_an_attribution_miss(tmp_path):
+    """Across fifty thousand tracks, titles collide constantly: "Everyday" is
+    Rusko and also Def Leppard. Matching those would be worse than missing."""
+    store = indexed(tmp_path, [scrobble_of("Yellowcard", "Hells Bells")])
+
+    # The library holds "Hells Bells", by AC/DC, and its title says nothing
+    # about Yellowcard. A collision, not a miss.
+    assert verdict(store, "Yellowcard", "Hells Bells") == ("none", None)
+    assert attribution_pairs(store) == []
+
+
+def test_a_remix_credited_in_the_library_title_is_an_attribution_miss(tmp_path):
+    """The library has "Voodoo People (Pendulum Remix)" under The Prodigy; the
+    scrobble credits Pendulum. Same song, different filing."""
+    store = indexed(tmp_path, [scrobble_of("Pendulum", "Voodoo People")])
+
+    assert attribution_pairs(store) == ["Pendulum"]
+
+
+def test_the_title_index_is_used_for_the_report_lookup(tmp_path):
+    """Without it the report scans every track for every unmatched key: forty-
+    seven seconds on a real library."""
+    store = indexed(tmp_path, [])
+
+    plan = "\n".join(
+        row[-1]
+        for row in store.connection.execute(
+            "EXPLAIN QUERY PLAN SELECT 1 FROM scrobble_key k WHERE k.track_id IS NULL"
+            " AND EXISTS (SELECT 1 FROM lidarr_track t WHERE t.norm_title = k.norm_track)"
+        )
+    )
+    assert "lidarr_track_title" in plan, plan
+
+
+def playlist_library(tmp_path):
+    """A library on disk, with mirror MP3s beside the Lidarr source paths."""
+    source = tmp_path / "music"
+    mirror = tmp_path / "mirror"
+    library = [
+        {
+            "name": "Played Band",
+            "albums": [
+                {
+                    "title": "Known",
+                    "tracks": [{"title": "Hit"}, {"title": "Album Track"}],
+                }
+            ],
+        },
+        {
+            "name": "Silent Band",
+            "albums": [{"title": "Unknown", "tracks": [{"title": "Never Heard"}]}],
+        },
+    ]
+    api = FakeLidarr(library)
+    # FakeLidarr invents /music/<artist>/<album>/<title>.flac; put the mirror
+    # MP3s at the paths music-mirror would have produced from those.
+    for handle in api.files:
+        relative = Path(handle["path"]).relative_to("/music")
+        handle["path"] = str(source / relative)
+        mp3 = (mirror / relative).with_suffix(".mp3")
+        mp3.parent.mkdir(parents=True, exist_ok=True)
+        mp3.write_bytes(b"not really an mp3")
+    for artist in api.artists:
+        artist["path"] = str(source / Path(artist["path"]).name)
+    return api, source, mirror
+
+
+def test_playlists_are_written_into_the_mirror(tmp_path):
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    ingest(store, FakeLastfm([scrobble_of("Played Band", "Hit")] * 1))
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    music_curator.match_library(store)
+
+    music_curator.build_playlists(store, mirror, str(source), 100, NOW)
+
+    written = sorted(p.name for p in (mirror / "_playlists").glob("*.m3u"))
+    assert written == [
+        "all-time.m3u",
+        "deep-cuts.m3u",
+        "heavy-rotation.m3u",
+        "neglected.m3u",
+        "unheard-favourites.m3u",
+        "unheard.m3u",
+    ]
+
+    played = (mirror / "_playlists" / "all-time.m3u").read_text().splitlines()
+    assert played[0] == "#EXTM3U"
+    assert played[1].startswith("#EXTINF:")
+    assert "Played Band - Hit" in played[1]
+    # Relative to the playlist file, so the same file works from any mount.
+    assert played[2] == "../Played Band/Known/Hit.mp3"
+    assert (mirror / "_playlists" / "all-time.m3u").parent.joinpath(played[2]).resolve().is_file()
+
+
+def test_an_unplayed_track_lands_in_the_unheard_playlist(tmp_path):
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    ingest(store, FakeLastfm([scrobble_of("Played Band", "Hit")]))
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    music_curator.match_library(store)
+
+    music_curator.build_playlists(store, mirror, str(source), 100, NOW)
+
+    unheard = (mirror / "_playlists" / "unheard.m3u").read_text()
+    assert "Never Heard" in unheard
+    assert "Album Track" in unheard
+    # The one thing that was played must not be in it.
+    assert "Played Band - Hit\n" not in unheard
+
+
+def test_a_track_with_no_mirror_file_is_left_out(tmp_path):
+    """The index knows what Lidarr holds; whether music-mirror has encoded the
+    MP3 yet is a different question, and is checked rather than assumed."""
+    api, source, mirror = playlist_library(tmp_path)
+    for mp3 in mirror.rglob("*.mp3"):
+        mp3.unlink()
+    store = store_at(tmp_path)
+    ingest(store, FakeLastfm([scrobble_of("Played Band", "Hit")]))
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    music_curator.match_library(store)
+
+    total = music_curator.build_playlists(store, mirror, str(source), 100, NOW)
+
+    assert total == 0
+    assert (mirror / "_playlists" / "all-time.m3u").read_text() == "#EXTM3U\n"
+
+
+def test_the_library_root_is_derived_from_the_artist_folders(tmp_path):
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+
+    assert music_curator.library_root_of(store) == str(source)
+
+
+def test_the_playlist_limit_is_honoured(tmp_path):
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    music_curator.match_library(store)
+
+    music_curator.build_playlists(store, mirror, str(source), 1, NOW)
+
+    unheard = (mirror / "_playlists" / "unheard.m3u").read_text().splitlines()
+    assert len([line for line in unheard if line.startswith("#EXTINF")]) == 1
+
+
+def test_the_rotation_moves_weekly_not_every_pass(tmp_path):
+    """A playlist that reorders on every pass is one that has to be re-imported
+    on every pass; the Music app imports a snapshot, it does not track a file."""
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    music_curator.match_library(store)
+
+    def unheard_at(when):
+        music_curator.build_playlists(store, mirror, str(source), 100, when)
+        return (mirror / "_playlists" / "unheard.m3u").read_text()
+
+    same_week = unheard_at(NOW), unheard_at(NOW + 3600)
+    assert same_week[0] == same_week[1]
+
+
+def test_playlists_are_group_readable(tmp_path):
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    music_curator.match_library(store)
+
+    music_curator.build_playlists(store, mirror, str(source), 100, NOW)
+
+    for playlist in (mirror / "_playlists").glob("*.m3u"):
+        assert playlist.stat().st_mode & stat.S_IRGRP, playlist
+
+
+def test_tag_weights_fall_back_to_rank_when_no_count_is_sent():
+    """The documented sample carries only a name and a URL; a live response also
+    carries a count. Neither may be relied on alone."""
+    with_count = music_curator.parse_tags(
+        {"toptags": {"tag": [{"name": "Screamo", "count": 100}, {"name": "emo", "count": 40}]}}
+    )
+    assert with_count == [("screamo", 100), ("emo", 40)]
+
+    without = music_curator.parse_tags(
+        {"toptags": {"tag": [{"name": "screamo"}, {"name": "emo"}]}}
+    )
+    assert [tag for tag, _ in without] == ["screamo", "emo"]
+    assert without[0][1] > without[1][1]
+
+
+def test_a_lone_tag_is_not_a_list():
+    assert music_curator.parse_tags({"toptags": {"tag": {"name": "dnb", "count": 90}}}) == [
+        ("dnb", 90)
+    ]
+
+
+def test_an_artist_lastfm_cannot_answer_for_is_not_asked_again(tmp_path):
+    """Recording the fetch even when it returns nothing is what stops a pass
+    spending a request per unknown artist, forever."""
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    lastfm = FakeLastfm(tags={})
+
+    assert music_curator.sync_tags(client_for(lastfm), store, NOW) == 2
+    before = len(lastfm.calls)
+
+    assert music_curator.sync_tags(client_for(lastfm), store, NOW) == 0
+    assert len(lastfm.calls) == before
+
+
+def test_tags_are_refetched_once_they_go_stale(tmp_path):
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    lastfm = FakeLastfm(tags={})
+    music_curator.sync_tags(client_for(lastfm), store, NOW)
+
+    later = NOW + music_curator.TAG_REFRESH_SECONDS + 1
+    assert music_curator.sync_tags(client_for(lastfm), store, later) == 2
+
+
+def tagged_store(tmp_path, tags):
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    music_curator.sync_tags(client_for(FakeLastfm(tags=tags)), store, NOW)
+    return store, source, mirror
+
+
+def test_a_vibe_selects_by_tag(tmp_path):
+    store, source, mirror = tagged_store(
+        tmp_path,
+        {
+            "Played Band": [{"name": "screamo", "count": 100}],
+            "Silent Band": [{"name": "classic rock", "count": 100}],
+        },
+    )
+    vibes = [{"name": "screamo", "tags": ["screamo"]}]
+
+    music_curator.build_vibe_playlists(store, vibes, mirror, str(source), 100, NOW)
+
+    written = (mirror / "_playlists" / "screamo.m3u").read_text()
+    assert "Played Band" in written
+    assert "Silent Band" not in written
+
+
+def test_a_weakly_tagged_artist_is_below_the_threshold(tmp_path):
+    """A single low-weight tag is not a genre, it is somebody's stray opinion."""
+    store, source, mirror = tagged_store(
+        tmp_path, {"Played Band": [{"name": "screamo", "count": 3}]}
+    )
+    vibes = [{"name": "screamo", "tags": ["screamo"]}]
+
+    music_curator.build_vibe_playlists(store, vibes, mirror, str(source), 100, NOW)
+
+    assert (mirror / "_playlists" / "screamo.m3u").read_text() == "#EXTM3U\n"
+
+
+def test_a_vibe_can_be_restricted_by_release_year(tmp_path):
+    """What separates eighties synth records from everything else a synthpop
+    tag drags in."""
+    store, source, mirror = tagged_store(
+        tmp_path, {"Played Band": [{"name": "synthpop", "count": 100}]}
+    )
+    inside = [{"name": "eighties", "tags": ["synthpop"], "years": [1975, 1992]}]
+    outside = [{"name": "nineties", "tags": ["synthpop"], "years": [1993, 1999]}]
+
+    # FakeLidarr dates every album 2019, so neither window should catch it.
+    music_curator.build_vibe_playlists(store, inside, mirror, str(source), 100, NOW)
+    music_curator.build_vibe_playlists(store, outside, mirror, str(source), 100, NOW)
+    assert (mirror / "_playlists" / "eighties.m3u").read_text() == "#EXTM3U\n"
+    assert (mirror / "_playlists" / "nineties.m3u").read_text() == "#EXTM3U\n"
+
+    modern = [{"name": "modern", "tags": ["synthpop"], "years": [2000, 2030]}]
+    music_curator.build_vibe_playlists(store, modern, mirror, str(source), 100, NOW)
+    assert "Played Band" in (mirror / "_playlists" / "modern.m3u").read_text()
+
+
+def test_the_built_in_vibes_are_all_usable_filenames():
+    for vibe in music_curator.DEFAULT_VIBES:
+        assert music_curator.SAFE_VIBE_NAME.fullmatch(vibe["name"]), vibe["name"]
+        assert vibe["tags"]
+
+
+def test_a_vibes_file_replaces_the_built_in_set(tmp_path):
+    path = tmp_path / "vibes.json"
+    path.write_text(json.dumps([{"name": "mine", "tags": ["shoegaze"]}]))
+
+    assert music_curator.load_vibes(str(path)) == ({"name": "mine", "tags": ["shoegaze"]},)
+    assert music_curator.load_vibes(None) is music_curator.DEFAULT_VIBES
+
+
+@pytest.mark.parametrize(
+    "content",
+    [
+        '{"not": "a list"}',
+        '[{"tags": ["x"]}]',
+        '[{"name": "../escape", "tags": ["x"]}]',
+        '[{"name": "ok"}]',
+        "not json at all",
+    ],
+)
+def test_a_bad_vibes_file_is_refused_up_front(tmp_path, content):
+    """A bad name would otherwise surface as a file written somewhere
+    unintended, which is a poor way to learn about a typo."""
+    path = tmp_path / "vibes.json"
+    path.write_text(content)
+
+    with pytest.raises(ValueError):
+        music_curator.load_vibes(str(path))
+
+
+def test_tags_are_looked_up_by_name_not_by_mbid(tmp_path):
+    """Last.fm's mbid index is stale: it cannot find Devo or Escape the Fate by
+    one, though their pages plainly exist. Its name index can."""
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    lastfm = FakeLastfm(tags={"Played Band": [{"name": "screamo", "count": 90}]})
+
+    music_curator.sync_tags(client_for(lastfm), store, NOW)
+
+    asked = [call for call in lastfm.calls if call.get("method") == "artist.getTopTags"]
+    # The name is always tried first; the mbid only appears as a fallback for
+    # the artist that the name could not resolve.
+    assert "artist" in asked[0]
+    assert [call for call in asked if call.get("artist") == "Played Band"]
+    assert not [call for call in asked if call.get("mbid") == "artist-mbid-1"]
+    assert store.scalar("SELECT COUNT(*) FROM artist_tag WHERE tag = 'screamo'") == 1
+
+
+def test_the_mbid_is_tried_when_the_name_is_not_found(tmp_path):
+    """Kept only for a name Lidarr spells differently to Last.fm."""
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    lastfm = FakeLastfm(tags={"artist-mbid-1": [{"name": "dnb", "count": 80}]})
+
+    music_curator.sync_tags(client_for(lastfm), store, NOW)
+
+    assert store.scalar("SELECT COUNT(*) FROM artist_tag WHERE tag = 'dnb'") == 1
+    asked = [call for call in lastfm.calls if call.get("method") == "artist.getTopTags"]
+    assert any("mbid" in call for call in asked)
+
+
+def test_an_artist_neither_key_resolves_is_recorded_and_not_retried(tmp_path):
+    """Otherwise every pass spends a request on it again, for ever."""
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    lastfm = FakeLastfm(tags={})
+
+    assert music_curator.sync_tags(client_for(lastfm), store, NOW) == 2
+    spent = len(lastfm.calls)
+
+    assert music_curator.sync_tags(client_for(lastfm), store, NOW) == 0
+    assert len(lastfm.calls) == spent
+
+
+def test_a_real_failure_is_not_recorded_so_the_next_pass_retries(tmp_path):
+    """A rate limit is not the same as an artist not existing."""
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    lastfm = FakeLastfm(
+        tags={"Played Band": [], "Silent Band": []},
+        outcomes=[{"error": 10, "message": "Invalid API key"}],
+    )
+
+    music_curator.sync_tags(client_for(lastfm), store, NOW)
+
+    # One artist failed hard and must still be pending.
+    assert store.scalar("SELECT COUNT(*) FROM artist_tag_fetched") == 1