feat(enrichment): record status/errors and trust high-confidence AcoustID

Two related gaps surfaced from "uploaded a track, nothing changed / no status":

- A track could stay stuck on `pending` forever (an unexpected worker error
  rolled back the run without recording anything), and `failed` carried no
  reason. Add `tracks.metadata_error` + `tracks.enriched_at` (migration), stamp
  the outcome in apply_enrichment, add TrackRepository.mark_enrichment_failed,
  wrap enrich_task to persist crashes as `failed` in a fresh session, and emit a
  human-readable no-match reason. Expose metadata_error/enriched_at in TrackOut.

- The tag-first merge let junk embedded tags (e.g. "Music Track"/"Sound_13958")
  override even a 0.99-confidence AcoustID match. Add acoustid_trust_score
  (default 0.85): above it the acoustic identity wins for title/artist/album/
  year, tags are fallback; below it, tag-first as before.

Add a license-free real-file fixture (Scarlet Fire / Otis McDonald) whose junk
tags AcoustID overrides, with an always-on tag-reader test plus fpcalc/AcoustID/
network-gated identity + full-pipeline tests (skip on host, run in the container).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

app/api/schemas/track.py

@@ -17,6 +17,8 @@ class TrackOut(BaseModel):
     file_format: str
     file_size: int
     metadata_status: str
+    metadata_error: str | None
+    enriched_at: dt.datetime | None
     source: str
     has_cover: bool
     created_at: dt.datetime

app/api/v1/tracks.py

@@ -42,6 +42,8 @@ async def _build_track_out(
             file_format=t.file_format,
             file_size=t.file_size,
             metadata_status=t.metadata_status,
+            metadata_error=t.metadata_error,
+            enriched_at=t.enriched_at,
             source=t.source,
             has_cover=bool(t.album_id and albums.get(t.album_id) and albums[t.album_id].cover_path),
             created_at=t.created_at,

app/application/metadata_service.py

@@ -58,6 +58,7 @@ class MetadataEnrichmentService:
         acoustid: AcoustIdClient,
         cover_extractor: CoverArtExtractor | None = None,
         cover_provider: CoverArtProvider | None = None,
+        acoustid_trust_score: float = 0.85,
     ) -> None:
         self._tracks = tracks
         self._artists = artists
@@ -68,6 +69,7 @@ class MetadataEnrichmentService:
         self._acoustid = acoustid
         self._cover_extractor = cover_extractor
         self._cover_provider = cover_provider
+        self._acoustid_trust_score = acoustid_trust_score
 
     async def enrich(self, track_id: uuid.UUID) -> EnrichmentResult:
         track = await self._tracks.get_by_id(track_id)
@@ -81,16 +83,31 @@ class MetadataEnrichmentService:
         tags = await self._read_local(track.storage_uri)
         match = await self._identify(track.storage_uri)
 
-        # Merge sources: prefer embedded tags, fall back to the AcoustID match.
-        # ``title`` is guaranteed non-None by the existing track title; the rest
-        # stay None when neither source has them.
+        # Merge order is tag-first by default — embedded tags fix the common
+        # well-tagged offline case. But a *high-confidence* AcoustID match is the
+        # more trustworthy identity (downloaded files routinely carry junk tags
+        # like "Music Track"/"Sound_12345"), so above the trust threshold the
+        # acoustic match wins for the identity fields and tags become fallback.
         tag_title = tags.title if tags else None
         tag_artist = tags.artist if tags else None
         tag_album = tags.album if tags else None
-        title = _opt_str(tag_title, match.title if match else None) or track.title
-        artist_name = _opt_str(tag_artist, match.artist if match else None)
-        album_title = _opt_str(tag_album, match.album if match else None)
-        year = _first_int(tags.year if tags else None, match.year if match else None)
+        match_title = match.title if match else None
+        match_artist = match.artist if match else None
+        match_album = match.album if match else None
+        match_year = match.year if match else None
+        tag_year = tags.year if tags else None
+        trust_match = match is not None and match.score >= self._acoustid_trust_score
+
+        if trust_match:
+            title = _opt_str(match_title, tag_title) or track.title
+            artist_name = _opt_str(match_artist, tag_artist)
+            album_title = _opt_str(match_album, tag_album)
+            year = _first_int(match_year, tag_year)
+        else:
+            title = _opt_str(tag_title, match_title) or track.title
+            artist_name = _opt_str(tag_artist, match_artist)
+            album_title = _opt_str(tag_album, match_album)
+            year = _first_int(tag_year, match_year)
         genre = tags.genre if tags else None
         track_number = tags.track_number if tags else None
         duration = _first_int(
@@ -114,6 +131,9 @@ class MetadataEnrichmentService:
 
         identified = bool(artist_name) or album_id is not None or mbid is not None
         status = "enriched" if identified else "failed"
+        # On a clean "no identity" outcome, record *why* so the UI shows a reason
+        # rather than a bare "failed". A successful run clears any prior error.
+        metadata_error = None if identified else self._no_match_reason()
 
         await self._tracks.apply_enrichment(
             track_id,
@@ -128,10 +148,20 @@ class MetadataEnrichmentService:
             acoustid_fingerprint=acoustid_id,
             musicbrainz_id=mbid,
             metadata_status=status,
+            metadata_error=metadata_error,
         )
         log.info("enrich_complete", track_id=str(track_id), status=status, mbid=mbid)
         return EnrichmentResult(track_id=track_id, status=status, matched_mbid=mbid)
 
+    def _no_match_reason(self) -> str:
+        """Explain a ``failed`` (no-identity) run in terms a user can act on:
+        which optional identification step was unavailable, if any."""
+        if not self._fingerprinter.is_available():
+            return "No metadata match: audio fingerprinting (fpcalc) is unavailable."
+        if not self._acoustid.is_available():
+            return "No metadata match: AcoustID lookup is unavailable (no API key)."
+        return "No metadata match found in tags or AcoustID."
+
     async def _read_local(self, storage_uri: str) -> AudioTags | None:
         try:
             async with self._storage.as_local_path(storage_uri) as path:

app/core/config.py

@@ -90,6 +90,10 @@ class Settings(BaseSettings):
     ml_service_url: str | None = None
     acoustid_api_key: SecretStr | None = None
     acoustid_api_url: str = "https://api.acoustid.org/v2/lookup"
+    # Above this AcoustID match score, trust the acoustic identification over
+    # embedded file tags (which are frequently junk on downloaded files —
+    # e.g. "Music Track" / "Sound_12345"). Below it, keep the tag-first merge.
+    acoustid_trust_score: float = 0.85
     # MusicBrainz/AcoustID require a meaningful User-Agent identifying the
     # application and a way to contact its maintainer (see
     # https://musicbrainz.org/doc/XML_Web_Service/Rate_Limiting). Self-hosted

app/domain/entities/track.py

@@ -28,5 +28,7 @@ class Track:
     genre: str | None
     year: int | None
     metadata_status: str
+    metadata_error: str | None
+    enriched_at: dt.datetime | None
     created_at: dt.datetime
     updated_at: dt.datetime

app/domain/ports.py

@@ -172,11 +172,18 @@ class TrackRepository(Protocol):
         acoustid_fingerprint: str | None,
         musicbrainz_id: str | None,
         metadata_status: str,
+        metadata_error: str | None = None,
     ) -> Track:
         """Persist auto-enrichment results. Nullable fields are filled only when
         a non-``None`` value is supplied (re-enrich never erases prior data);
-        ``title``/``artist_id``/``metadata_status`` are always written. Callers
-        must not invoke this for ``metadata_status == 'manual'`` tracks."""
+        ``title``/``artist_id``/``metadata_status`` are always written, and the
+        run's outcome (``metadata_error`` + completion time) is always stamped.
+        Callers must not invoke this for ``metadata_status == 'manual'`` tracks."""
+        ...
+
+    async def mark_enrichment_failed(self, track_id: uuid.UUID, *, error: str) -> None:
+        """Record that an enrichment run crashed unexpectedly: set ``failed`` +
+        the error reason. A no-op for ``manual`` or missing tracks."""
         ...
 
 

app/infrastructure/db/models/track.py

@@ -6,9 +6,10 @@
 imports/downloads stay idempotent (plan §4, §6.1).
 """
 
+import datetime as dt
 import uuid
 
-from sqlalchemy import ForeignKey, Integer, String, UniqueConstraint
+from sqlalchemy import DateTime, ForeignKey, Integer, String, UniqueConstraint
 from sqlalchemy.orm import Mapped, mapped_column
 
 from app.infrastructure.db.base import Base
@@ -63,6 +64,16 @@ class TrackModel(UUIDPrimaryKeyMixin, TimestampMixin, Base):
         nullable=False,
         default=MetadataStatus.PENDING.value,
     )
+    # Human-readable reason the last enrichment run set ``failed`` (no match, or
+    # an unexpected worker error). ``None`` once a run succeeds. Surfaced in the
+    # UI so a stuck/failed track is diagnosable, not silent.
+    metadata_error: Mapped[str | None] = mapped_column(String(2048), nullable=True)
+    # When the last enrichment run finished (success or failure). ``None`` while
+    # still ``pending`` — lets the UI distinguish "queued/running" from "done".
+    enriched_at: Mapped[dt.datetime | None] = mapped_column(
+        DateTime(timezone=True),
+        nullable=True,
+    )
 
     added_by: Mapped[uuid.UUID | None] = mapped_column(
         ForeignKey("users.id", ondelete="SET NULL"),

app/infrastructure/db/repositories/like_repository.py

@@ -39,6 +39,8 @@ def _track_to_entity(row: TrackModel) -> Track:
         genre=row.genre,
         year=row.year,
         metadata_status=row.metadata_status,
+        metadata_error=row.metadata_error,
+        enriched_at=row.enriched_at,
         created_at=row.created_at,
         updated_at=row.updated_at,
     )

app/infrastructure/db/repositories/playlist_repository.py

@@ -38,6 +38,8 @@ def _track_to_entity(row: TrackModel) -> Track:
         genre=row.genre,
         year=row.year,
         metadata_status=row.metadata_status,
+        metadata_error=row.metadata_error,
+        enriched_at=row.enriched_at,
         created_at=row.created_at,
         updated_at=row.updated_at,
     )

app/infrastructure/db/repositories/track_repository.py

@@ -1,5 +1,6 @@
 """Track repository — adapter over ``AsyncSession``."""
 
+import datetime as dt
 import uuid
 
 from sqlalchemy import func, select
@@ -26,6 +27,8 @@ def _to_entity(row: TrackModel) -> Track:
         genre=row.genre,
         year=row.year,
         metadata_status=row.metadata_status,
+        metadata_error=row.metadata_error,
+        enriched_at=row.enriched_at,
         created_at=row.created_at,
         updated_at=row.updated_at,
     )
@@ -189,6 +192,7 @@ class SqlAlchemyTrackRepository:
         acoustid_fingerprint: str | None,
         musicbrainz_id: str | None,
         metadata_status: str,
+        metadata_error: str | None = None,
     ) -> Track:
         row = await self._session.get(TrackModel, track_id)
         if row is None:
@@ -197,6 +201,10 @@ class SqlAlchemyTrackRepository:
         row.title = title
         row.artist_id = artist_id
         row.metadata_status = metadata_status
+        # A finished run always stamps outcome: clear/set the reason and mark the
+        # completion time so the UI can tell "still pending" from "done/failed".
+        row.metadata_error = metadata_error
+        row.enriched_at = dt.datetime.now(dt.UTC)
         # Nullable extras: fill gaps only — never erase data a prior run found.
         if album_id is not None:
             row.album_id = album_id
@@ -217,3 +225,16 @@ class SqlAlchemyTrackRepository:
         await self._session.flush()
         await self._session.refresh(row)
         return _to_entity(row)
+
+    async def mark_enrichment_failed(self, track_id: uuid.UUID, *, error: str) -> None:
+        """Record that an enrichment run crashed (unexpected exception). Runs in
+        its own session so the failure is persisted even though the run's own
+        transaction rolled back. Never overwrites ``manual`` (a no-op then), and
+        a missing track is a clean no-op."""
+        row = await self._session.get(TrackModel, track_id)
+        if row is None or row.metadata_status == "manual":
+            return
+        row.metadata_status = "failed"
+        row.metadata_error = error
+        row.enriched_at = dt.datetime.now(dt.UTC)
+        await self._session.flush()

app/workers/tasks/enrich_task.py

@@ -42,19 +42,32 @@ async def enrich_track(_ctx: dict[str, Any], *, track_id: str) -> dict[str, Any]
         base_url=settings.coverart_base_url,
     )
 
-    async with session_scope() as session:
-        service = MetadataEnrichmentService(
-            tracks=SqlAlchemyTrackRepository(session),
-            artists=SqlAlchemyArtistRepository(session),
-            albums=SqlAlchemyAlbumRepository(session),
-            storage=get_file_storage(),
-            tag_reader=MutagenTagReader(),
-            fingerprinter=FpcalcFingerprinter(settings.fpcalc_path),
-            acoustid=acoustid,
-            cover_extractor=MutagenCoverExtractor(),
-            cover_provider=cover_provider,
-        )
-        result = await service.enrich(uuid.UUID(track_id))
+    tid = uuid.UUID(track_id)
+    try:
+        async with session_scope() as session:
+            service = MetadataEnrichmentService(
+                tracks=SqlAlchemyTrackRepository(session),
+                artists=SqlAlchemyArtistRepository(session),
+                albums=SqlAlchemyAlbumRepository(session),
+                storage=get_file_storage(),
+                tag_reader=MutagenTagReader(),
+                fingerprinter=FpcalcFingerprinter(settings.fpcalc_path),
+                acoustid=acoustid,
+                cover_extractor=MutagenCoverExtractor(),
+                cover_provider=cover_provider,
+                acoustid_trust_score=settings.acoustid_trust_score,
+            )
+            result = await service.enrich(tid)
+    except Exception as exc:
+        # The run's own transaction rolled back, leaving the track stuck at
+        # ``pending``. Record the failure in a fresh session so the UI shows a
+        # ``failed`` status with a reason instead of a silent, endless spinner.
+        log.exception("enrich_failed", track_id=track_id)
+        async with session_scope() as session:
+            await SqlAlchemyTrackRepository(session).mark_enrichment_failed(
+                tid, error=f"Enrichment crashed: {type(exc).__name__}: {exc}"
+            )
+        return {"track_id": track_id, "status": "failed", "mbid": None}
 
     return {
         "track_id": str(result.track_id),