test: cover lyrics, transcode, and recommendation

DB-free unit tests for the three previously-untested features:
- lyrics: get-or-fetch caching, not_found TTL, force refetch, graceful miss;
  plus the LRC -> structured-lyrics serializer (timing, plain fallback, empty)
- transcode: path helpers, segment-name traversal guard, cache hit/miss, the
  unknown/not-downloaded 404 paths
- reco: ML path (order-preserving batched hydration) + metadata fallback for
  similar/radio, exclude handling, reason codes

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

.claude/settings.json

@@ -0,0 +1,24 @@
+{
+  "hooks": {
+    "PreToolUse": [
+      {
+        "matcher": "Bash",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "/Users/senko/.local/bin/graphify hook-guard search"
+          }
+        ]
+      },
+      {
+        "matcher": "Read|Glob",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "/Users/senko/.local/bin/graphify hook-guard read"
+          }
+        ]
+      }
+    ]
+  }
+}

.env.example

@@ -16,6 +16,9 @@ REDIS_URL=redis://localhost:6379/0
 JWT_SECRET=change-me-in-prod
 ACCESS_TOKEN_TTL_SECONDS=900
 REFRESH_TOKEN_TTL_SECONDS=2592000
+# Public self-service sign-up (POST /auth/register). Set to false to make
+# accounts admin-only. Registered users are never superusers.
+ALLOW_REGISTRATION=true
 
 # subsonic — key that encrypts per-user Subsonic app-passwords at rest.
 # GENERATE a strong secret for prod (`openssl rand -hex 32`); rotating it
@@ -27,8 +30,13 @@ MEDIA_PATH=/data/media
 TRANSCODE_CACHE_PATH=/data/transcode-cache
 MAX_PARALLEL_DOWNLOADS=2
 
+# sources — mounted folder the `local` source indexes (copies into MEDIA_PATH).
+# Unset → the local source is not registered. Mount read-only in compose.
+# LOCAL_MEDIA_IMPORT_PATH=/import
+
 # external services (all optional — backend degrades gracefully if unset)
 # ML_SERVICE_URL=http://ml:9000
 # ACOUSTID_API_KEY=
-MUSICBRAINZ_USER_AGENT=mcma-backend/0.1.0 ( https://github.com/your/repo )
+# Sent to MusicBrainz/AcoustID as part of the User-Agent (MCMA/<version> ( <email> )).
+# MUSICBRAINZ_OWNER_EMAIL=you@example.com
 # YOUTUBE_COOKIES_PATH=/data/cookies.txt

CLAUDE.md

@@ -74,3 +74,13 @@ These exist for future sync + ML and are easy to violate by accident:
 ## Python 3.14 note
 
 The project targets Python 3.14, which makes annotations lazy by default (PEP 649). `from __future__ import annotations` is therefore intentionally **absent** — do not add it back. All pins (pyproject `requires-python`, ruff `target-version`, mypy `python_version`, Dockerfile, `.python-version`) are on 3.14.
+
+## graphify
+
+This project has a knowledge graph at graphify-out/ with god nodes, community structure, and cross-file relationships.
+
+Rules:
+- For codebase questions, first run `graphify query "<question>"` when graphify-out/graph.json exists. Use `graphify path "<A>" "<B>"` for relationships and `graphify explain "<concept>"` for focused concepts. These return a scoped subgraph, usually much smaller than GRAPH_REPORT.md or raw grep output.
+- If graphify-out/wiki/index.md exists, use it for broad navigation instead of raw source browsing.
+- Read graphify-out/GRAPH_REPORT.md only for broad architecture review or when query/path/explain do not surface enough context.
+- After modifying code, run `graphify update .` to keep the graph current (AST-only, no API cost).

README.md

@@ -71,6 +71,26 @@ All settings come from environment variables (or `.env` in dev). See
 [`.env.example`](.env.example). External services (ML, AcoustID, MusicBrainz)
 are **optional** — the backend degrades gracefully when they are absent.
 
+## Sources & importing music
+
+Music enters the library through **source backends** (`app/infrastructure/sources`),
+selected via a registry. The first backend is **`local`** — it indexes a mounted
+folder, copying each audio file into managed storage and creating a track
+(`metadata_status=pending`; real metadata is filled later by enrichment).
+
+```bash
+# point the instance at an existing library (mount read-only in compose)
+LOCAL_MEDIA_IMPORT_PATH=/import
+
+GET  /api/v1/sources                 # list configured sources + availability
+POST /api/v1/sources/local/scan      # admin: enqueue an import (runs in the worker)
+GET  /api/v1/sources/local/health    # availability check
+```
+
+Scanning is a background job (arq worker) — the endpoint only enqueues it; the
+walk + file copies never run in the request cycle. Re-scans are idempotent
+(dedup on `(source, source_id)`, where `source_id` is the path within the root).
+
 ## Subsonic API (`/rest`)
 
 A Subsonic-compatible API is mounted at `/rest`, so standard clients (Symfonium,

alembic/versions/20260613_1300-track_enrichment_outcome.py

@@ -0,0 +1,39 @@
+"""tracks: enrichment outcome (error reason + completion time)
+
+Revision ID: 20260613_enrich_outcome
+Revises: 20260608_subsonic_pw
+Create Date: 2026-06-13 13:00:00.000000
+
+Adds ``tracks.metadata_error`` and ``tracks.enriched_at`` so a finished
+enrichment run records *why* it failed and *when* it completed. Lets the UI
+distinguish a still-pending/running track from one that is done or failed, and
+surface an actionable reason instead of a silent spinner (plan §6.2).
+"""
+
+from __future__ import annotations
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from alembic import op
+
+revision: str = "20260613_enrich_outcome"
+down_revision: str | None = "20260608_subsonic_pw"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "tracks",
+        sa.Column("metadata_error", sa.String(length=2048), nullable=True),
+    )
+    op.add_column(
+        "tracks",
+        sa.Column("enriched_at", sa.DateTime(timezone=True), nullable=True),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("tracks", "enriched_at")
+    op.drop_column("tracks", "metadata_error")

alembic/versions/20260614_1000-download_job_track_id.py

@@ -0,0 +1,47 @@
+"""download_jobs: link finished job to its imported track
+
+Revision ID: 20260614_dl_track_id
+Revises: 20260613_enrich_outcome
+Create Date: 2026-06-14 10:00:00.000000
+
+Adds ``download_jobs.track_id`` (nullable FK → ``tracks.id``) so a completed
+download can point at the library track it produced — the §A5 download manager
+links a "done" job to the track, and re-runs can tell a job already imported
+(plan §6.1).
+"""
+
+from __future__ import annotations
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from alembic import op
+
+revision: str = "20260614_dl_track_id"
+down_revision: str | None = "20260613_enrich_outcome"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "download_jobs",
+        sa.Column("track_id", sa.Uuid(), nullable=True),
+    )
+    op.create_foreign_key(
+        op.f("fk_download_jobs_track_id_tracks"),
+        "download_jobs",
+        "tracks",
+        ["track_id"],
+        ["id"],
+        ondelete="SET NULL",
+    )
+
+
+def downgrade() -> None:
+    op.drop_constraint(
+        op.f("fk_download_jobs_track_id_tracks"),
+        "download_jobs",
+        type_="foreignkey",
+    )
+    op.drop_column("download_jobs", "track_id")

alembic/versions/20260614_1125-dc126696f5a6_remote_placeholders_track_availability_.py

@@ -0,0 +1,65 @@
+"""remote placeholders: track availability, album/artist remote ids
+
+Revision ID: dc126696f5a6
+Revises: 20260614_dl_track_id
+Create Date: 2026-06-14 11:25:30.643588
+"""
+
+from __future__ import annotations
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = 'dc126696f5a6'
+down_revision: str | None = '20260614_dl_track_id'
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.add_column('albums', sa.Column('source', sa.String(length=32), nullable=True))
+    op.add_column('albums', sa.Column('source_id', sa.String(length=512), nullable=True))
+    op.create_unique_constraint('uq_albums_source_source_id', 'albums', ['source', 'source_id'])
+    op.add_column('artists', sa.Column('source', sa.String(length=32), nullable=True))
+    op.add_column('artists', sa.Column('source_id', sa.String(length=512), nullable=True))
+    op.create_unique_constraint('uq_artists_source_source_id', 'artists', ['source', 'source_id'])
+    op.add_column(
+        'tracks',
+        sa.Column('availability', sa.String(length=16), nullable=False, server_default='local'),
+    )
+    op.alter_column('tracks', 'availability', server_default=None)
+    op.alter_column('tracks', 'storage_uri',
+               existing_type=sa.VARCHAR(length=2048),
+               nullable=True)
+    op.alter_column('tracks', 'file_format',
+               existing_type=sa.VARCHAR(length=32),
+               nullable=True)
+    op.alter_column('tracks', 'file_size',
+               existing_type=sa.INTEGER(),
+               nullable=True)
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.alter_column('tracks', 'file_size',
+               existing_type=sa.INTEGER(),
+               nullable=False)
+    op.alter_column('tracks', 'file_format',
+               existing_type=sa.VARCHAR(length=32),
+               nullable=False)
+    op.alter_column('tracks', 'storage_uri',
+               existing_type=sa.VARCHAR(length=2048),
+               nullable=False)
+    op.drop_column('tracks', 'availability')
+    op.drop_constraint('uq_artists_source_source_id', 'artists', type_='unique')
+    op.drop_column('artists', 'source_id')
+    op.drop_column('artists', 'source')
+    op.drop_constraint('uq_albums_source_source_id', 'albums', type_='unique')
+    op.drop_column('albums', 'source_id')
+    op.drop_column('albums', 'source')
+    # ### end Alembic commands ###

alembic/versions/20260728_1000-user_settings.py

@@ -0,0 +1,58 @@
+"""user_settings: per-user preferences + scrobbling config
+
+Revision ID: 20260728_user_settings
+Revises: dc126696f5a6
+Create Date: 2026-07-28 10:00:00.000000
+
+Adds the ``user_settings`` table (1:1 with ``users``, PK = user_id): general
+preferences (theme, stream quality) plus scrobbling config. The scrobbler
+session key is stored Fernet-encrypted, never in plaintext.
+"""
+
+from __future__ import annotations
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from alembic import op
+
+revision: str = "20260728_user_settings"
+down_revision: str | None = "dc126696f5a6"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "user_settings",
+        sa.Column("user_id", sa.Uuid(), nullable=False),
+        sa.Column("theme", sa.String(length=16), nullable=False),
+        sa.Column("stream_quality", sa.String(length=16), nullable=False),
+        sa.Column("scrobble_enabled", sa.Boolean(), nullable=False),
+        sa.Column("scrobble_provider", sa.String(length=16), nullable=True),
+        sa.Column("scrobble_username", sa.String(length=255), nullable=True),
+        sa.Column("scrobble_session_key_enc", sa.String(length=512), nullable=True),
+        sa.Column(
+            "created_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+        sa.Column(
+            "updated_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+        sa.ForeignKeyConstraint(
+            ["user_id"],
+            ["users.id"],
+            name=op.f("fk_user_settings_user_id_users"),
+            ondelete="CASCADE",
+        ),
+        sa.PrimaryKeyConstraint("user_id", name=op.f("pk_user_settings")),
+    )
+
+
+def downgrade() -> None:
+    op.drop_table("user_settings")

alembic/versions/20260728_1100-sync_ingestion_columns.py

@@ -0,0 +1,57 @@
+"""sync: server-ingestion columns on the event logs
+
+Revision ID: 20260728_sync_synced_at
+Revises: 20260728_user_settings
+Create Date: 2026-07-28 11:00:00.000000
+
+Adds ``synced_at`` to ``likes`` and ``play_history`` — the server-side ingestion
+time used as the delta-sync ordering key (distinct from the event time
+``created_at``/``played_at``, which a sync push preserves from the client even
+when the event happened offline earlier). Existing rows are backfilled from
+their event time so a first sync after upgrade behaves sensibly.
+"""
+
+from __future__ import annotations
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from alembic import op
+
+revision: str = "20260728_sync_synced_at"
+down_revision: str | None = "20260728_user_settings"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "likes",
+        sa.Column(
+            "synced_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+    )
+    op.execute("UPDATE likes SET synced_at = created_at")
+    op.create_index(op.f("ix_likes_synced_at"), "likes", ["synced_at"])
+
+    op.add_column(
+        "play_history",
+        sa.Column(
+            "synced_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+    )
+    op.execute("UPDATE play_history SET synced_at = played_at")
+    op.create_index(op.f("ix_play_history_synced_at"), "play_history", ["synced_at"])
+
+
+def downgrade() -> None:
+    op.drop_index(op.f("ix_play_history_synced_at"), table_name="play_history")
+    op.drop_column("play_history", "synced_at")
+    op.drop_index(op.f("ix_likes_synced_at"), table_name="likes")
+    op.drop_column("likes", "synced_at")

app/api/covers.py

@@ -0,0 +1,57 @@
+"""Shared cover-art serving helper (presentation).
+
+Streams a stored cover image from the :class:`FileStorage` port. Used by the
+native ``/api/v1`` cover endpoints and the Subsonic ``getCoverArt`` adapter so
+the streaming/content-type logic lives in one place.
+"""
+
+import uuid
+
+from fastapi.responses import StreamingResponse
+
+from app.domain.entities.album import Album
+from app.domain.errors import NotFoundError, StorageError
+from app.domain.ports import AlbumRepository, FileStorage, TrackRepository
+
+_CONTENT_TYPE_BY_EXT: dict[str, str] = {
+    "jpg": "image/jpeg",
+    "jpeg": "image/jpeg",
+    "png": "image/png",
+    "webp": "image/webp",
+    "gif": "image/gif",
+}
+
+# Covers are immutable for a given album (a new cover means a new key), so let
+# clients cache aggressively.
+_CACHE_CONTROL = "public, max-age=86400"
+
+
+def _content_type_for(key: str) -> str:
+    ext = key.rsplit(".", 1)[-1].lower() if "." in key else ""
+    return _CONTENT_TYPE_BY_EXT.get(ext, "application/octet-stream")
+
+
+async def stream_cover(storage: FileStorage, cover_path: str) -> StreamingResponse:
+    """Stream a stored cover by its storage key. Raises ``NotFoundError`` if the
+    object is missing (a dangling ``cover_path`` reads as "no cover")."""
+    try:
+        stream, total = await storage.open_range(cover_path, 0, None)
+    except StorageError as exc:
+        raise NotFoundError("Cover not found.") from exc
+    return StreamingResponse(
+        stream,
+        media_type=_content_type_for(cover_path),
+        headers={"Content-Length": str(total), "Cache-Control": _CACHE_CONTROL},
+    )
+
+
+async def resolve_album_for_track(
+    track_repo: TrackRepository,
+    album_repo: AlbumRepository,
+    track_id: uuid.UUID,
+) -> Album | None:
+    """The album that owns a track (cover lives on the album), or ``None``."""
+    track = await track_repo.get_by_id(track_id)
+    if track is None or track.album_id is None:
+        return None
+    return await album_repo.get_by_id(track.album_id)

app/api/deps.py

@@ -6,19 +6,29 @@ bound to the request-scoped DB session; stateless adapters (hasher, token
 service) are process-cached.
 """
 
+import datetime as dt
 from collections.abc import AsyncIterator
 from functools import lru_cache
 from typing import Annotated
 
 from fastapi import Depends, Query
 from fastapi.security import HTTPAuthorizationCredentials, HTTPBearer
+from sqlalchemy import func, select
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.application.auth_service import AuthService
+from app.application.download_service import DownloadService
+from app.application.lyrics_service import LyricsService
+from app.application.metadata_service import MetadataEnrichmentService
+from app.application.recommendation_service import RecommendationService
+from app.application.remote_library_service import RemoteLibraryService
 from app.application.streaming_service import StreamingService
 from app.application.subsonic_auth_service import SubsonicAuthService
+from app.application.sync_service import SyncService
+from app.application.transcode_service import TranscodeService
 from app.application.upload_service import UploadService
 from app.application.user_service import UserService
+from app.application.user_settings_service import UserSettingsService
 from app.core.config import get_settings
 from app.core.security import Argon2PasswordHasher, JwtTokenService, SubsonicPasswordCipher
 from app.domain.entities import User
@@ -28,14 +38,24 @@ from app.infrastructure.db import get_sessionmaker
 from app.infrastructure.db.repositories import (
     SqlAlchemyAlbumRepository,
     SqlAlchemyArtistRepository,
+    SqlAlchemyDownloadJobRepository,
     SqlAlchemyHistoryRepository,
     SqlAlchemyLikeRepository,
+    SqlAlchemyLyricsRepository,
     SqlAlchemyPlaylistRepository,
     SqlAlchemyRefreshTokenRepository,
     SqlAlchemyTrackRepository,
     SqlAlchemyUserRepository,
+    SqlAlchemyUserSettingsRepository,
 )
+from app.infrastructure.metadata.acoustid import AcoustIdHttpClient
+from app.infrastructure.metadata.fingerprint import FpcalcFingerprinter
+from app.infrastructure.metadata.lrclib import LrclibHttpClient
+from app.infrastructure.metadata.tags import MutagenTagReader
+from app.infrastructure.ml.recommender import NullRecommender
+from app.infrastructure.sources.registry import SourceRegistry, build_source_registry
 from app.infrastructure.storage.provider import get_file_storage
+from app.workers.queue import enqueue_download, enqueue_enrich, enqueue_materialize
 
 
 async def get_session() -> AsyncIterator[AsyncSession]:
@@ -70,6 +90,14 @@ def get_subsonic_cipher() -> SubsonicCipher:
     return SubsonicPasswordCipher(get_settings().subsonic_secret_key.get_secret_value())
 
 
+@lru_cache
+def get_source_registry() -> SourceRegistry:
+    return build_source_registry(get_settings())
+
+
+SourceRegistryDep = Annotated[SourceRegistry, Depends(get_source_registry)]
+
+
 # -- request-scoped services ---------------------------------------------------
 def get_auth_service(session: SessionDep) -> AuthService:
     return AuthService(
@@ -95,9 +123,17 @@ def get_subsonic_auth_service(session: SessionDep) -> SubsonicAuthService:
     )
 
 
+def get_user_settings_service(session: SessionDep) -> UserSettingsService:
+    return UserSettingsService(
+        settings=SqlAlchemyUserSettingsRepository(session),
+        cipher=get_subsonic_cipher(),
+    )
+
+
 AuthServiceDep = Annotated[AuthService, Depends(get_auth_service)]
 UserServiceDep = Annotated[UserService, Depends(get_user_service)]
 SubsonicAuthServiceDep = Annotated[SubsonicAuthService, Depends(get_subsonic_auth_service)]
+UserSettingsServiceDep = Annotated[UserSettingsService, Depends(get_user_settings_service)]
 
 
 # -- file storage (process-cached) ---------------------------------------------
@@ -111,6 +147,7 @@ def get_upload_service(session: SessionDep, storage: FileStorageDep) -> UploadSe
         artists=SqlAlchemyArtistRepository(session),
         storage=storage,
         tmp_dir=settings.upload_tmp_dir,
+        enqueue_enrich=enqueue_enrich,
     )
 
 
@@ -121,8 +158,111 @@ def get_streaming_service(session: SessionDep, storage: FileStorageDep) -> Strea
     )
 
 
+def get_metadata_service(session: SessionDep, storage: FileStorageDep) -> MetadataEnrichmentService:
+    """Wires the §6.2 fingerprint/AcoustID adapters for read-only, inline use
+    (the metadata editor's "find matches" — §A7). The full pipeline (incl.
+    cover art) stays in the worker (`tasks/enrich_task.py`)."""
+    settings = get_settings()
+    api_key = settings.acoustid_api_key.get_secret_value() if settings.acoustid_api_key else None
+    acoustid = AcoustIdHttpClient(
+        api_key=api_key,
+        user_agent=settings.musicbrainz_user_agent,
+        api_url=settings.acoustid_api_url,
+    )
+    return MetadataEnrichmentService(
+        tracks=SqlAlchemyTrackRepository(session),
+        artists=SqlAlchemyArtistRepository(session),
+        albums=SqlAlchemyAlbumRepository(session),
+        storage=storage,
+        tag_reader=MutagenTagReader(),
+        fingerprinter=FpcalcFingerprinter(settings.fpcalc_path),
+        acoustid=acoustid,
+        acoustid_trust_score=settings.acoustid_trust_score,
+    )
+
+
+def get_transcode_service(session: SessionDep) -> TranscodeService:
+    """Request-side cache lookups for transcoded renditions (§6.6). Generation
+    itself runs in the ``transcode_track`` worker, never here."""
+    return TranscodeService(
+        tracks=SqlAlchemyTrackRepository(session),
+        cache_root=get_settings().transcode_cache_path,
+    )
+
+
+def get_recommendation_service(session: SessionDep) -> RecommendationService:
+    """Radio + similarity (§6.5). ML is optional and no service/contract exists
+    yet, so we wire ``NullRecommender`` — the service then uses its metadata
+    fallback. Swap in ``RemoteRecommender(ml_service_url)`` once ML lands."""
+    return RecommendationService(
+        recommender=NullRecommender(),
+        tracks=SqlAlchemyTrackRepository(session),
+        artists=SqlAlchemyArtistRepository(session),
+        likes=SqlAlchemyLikeRepository(session),
+    )
+
+
+def get_lyrics_service(session: SessionDep) -> LyricsService:
+    """Wires the LRCLIB lyrics provider + cache repo (plan §6.7). LRCLIB is
+    keyless, so this is always available; failures degrade to ``not_found``."""
+    settings = get_settings()
+    return LyricsService(
+        lyrics=SqlAlchemyLyricsRepository(session),
+        tracks=SqlAlchemyTrackRepository(session),
+        artists=SqlAlchemyArtistRepository(session),
+        albums=SqlAlchemyAlbumRepository(session),
+        provider=LrclibHttpClient(user_agent=settings.musicbrainz_user_agent),
+    )
+
+
+def get_download_service(session: SessionDep, storage: FileStorageDep) -> DownloadService:
+    return DownloadService(
+        jobs=SqlAlchemyDownloadJobRepository(session),
+        tracks=SqlAlchemyTrackRepository(session),
+        artists=SqlAlchemyArtistRepository(session),
+        storage=storage,
+        enqueue_download=enqueue_download,
+        enqueue_enrich=enqueue_enrich,
+    )
+
+
+def get_remote_library_service(session: SessionDep) -> RemoteLibraryService:
+    return RemoteLibraryService(
+        tracks=SqlAlchemyTrackRepository(session),
+        artists=SqlAlchemyArtistRepository(session),
+        jobs=SqlAlchemyDownloadJobRepository(session),
+        enqueue_materialize=enqueue_materialize,
+    )
+
+
 UploadServiceDep = Annotated[UploadService, Depends(get_upload_service)]
 StreamingServiceDep = Annotated[StreamingService, Depends(get_streaming_service)]
+MetadataServiceDep = Annotated[MetadataEnrichmentService, Depends(get_metadata_service)]
+LyricsServiceDep = Annotated[LyricsService, Depends(get_lyrics_service)]
+TranscodeServiceDep = Annotated[TranscodeService, Depends(get_transcode_service)]
+RecommendationServiceDep = Annotated[
+    RecommendationService, Depends(get_recommendation_service)
+]
+DownloadServiceDep = Annotated[DownloadService, Depends(get_download_service)]
+RemoteLibraryServiceDep = Annotated[RemoteLibraryService, Depends(get_remote_library_service)]
+
+
+async def _db_now(session: AsyncSession) -> dt.datetime:
+    """The database clock — the sync cursor watermark (avoids app/DB skew)."""
+    return (await session.execute(select(func.now()))).scalar_one()
+
+
+def get_sync_service(session: SessionDep) -> SyncService:
+    return SyncService(
+        likes=SqlAlchemyLikeRepository(session),
+        history=SqlAlchemyHistoryRepository(session),
+        playlists=SqlAlchemyPlaylistRepository(session),
+        tracks=SqlAlchemyTrackRepository(session),
+        now=lambda: _db_now(session),
+    )
+
+
+SyncServiceDep = Annotated[SyncService, Depends(get_sync_service)]
 
 
 # -- library repository deps ---------------------------------------------------

app/api/errors.py

@@ -18,6 +18,7 @@ from app.domain.errors import (
     DependencyUnavailableError,
     DomainError,
     NotFoundError,
+    NotSupportedError,
     PermissionDeniedError,
     RangeNotSatisfiableError,
     StorageError,
@@ -33,6 +34,7 @@ _STATUS_BY_ERROR: dict[type[DomainError], int] = {
     ValidationError: status.HTTP_422_UNPROCESSABLE_CONTENT,
     AuthenticationError: status.HTTP_401_UNAUTHORIZED,
     PermissionDeniedError: status.HTTP_403_FORBIDDEN,
+    NotSupportedError: status.HTTP_501_NOT_IMPLEMENTED,
     DependencyUnavailableError: status.HTTP_503_SERVICE_UNAVAILABLE,
     StorageError: status.HTTP_500_INTERNAL_SERVER_ERROR,
 }

app/api/rest/media.py

@@ -1,10 +1,11 @@
-"""Subsonic media endpoints: stream, download, cover art.
+"""Subsonic media endpoints: stream, download, cover art, lyrics.
 
 ``stream`` and ``download`` reuse :class:`StreamingService` (honouring HTTP
 Range) — they return raw bytes, not the Subsonic envelope. Transcoding params
 (``maxBitRate``/``format``) are accepted but ignored; the original file is served
-(no in-request ffmpeg — CLAUDE.md). ``getCoverArt`` returns a placeholder until
+(no in-request ffmpeg — CLAUDE.md). ``getCoverArt`` serves the album cover (a
-the cover pipeline lands (the ``/api/v1`` cover endpoints are still stubs).
+placeholder when there's none). ``getLyricsBySongId`` adapts the native
+``LyricsService`` into the OpenSubsonic structured-lyrics shape.
 """
 
 import base64
@@ -13,9 +14,22 @@ from typing import Annotated
 from fastapi import APIRouter, Header, Query
 from fastapi.responses import Response, StreamingResponse
 
-from app.api.deps import StreamingServiceDep, SubsonicUser, TrackRepoDep
+from app.api.covers import resolve_album_for_track, stream_cover
-from app.api.rest.ids import decode_track, parse
+from app.api.deps import (
-from app.domain.errors import NotFoundError
+    AlbumRepoDep,
+    ArtistRepoDep,
+    FileStorageDep,
+    LyricsServiceDep,
+    StreamingServiceDep,
+    SubsonicFormat,
+    SubsonicUser,
+    TrackRepoDep,
+)
+from app.api.rest.envelope import subsonic_response
+from app.api.rest.ids import IdKind, decode_track, parse
+from app.api.rest.serializers import structured_lyrics
+from app.domain.entities.album import Album
+from app.domain.errors import NotFoundError, StorageError
 
 router = APIRouter()
 
@@ -57,7 +71,7 @@ async def download(
     if track is None:
         raise NotFoundError("Song not found.")
     result = await service.open_stream(track_id, None)
-    filename = f"{track.title}.{track.file_format}"
+    filename = f"{track.title}.{track.file_format or 'bin'}"
     headers = {
         "Content-Length": str(result.content_length),
         "Content-Disposition": f'attachment; filename="{filename}"',
@@ -69,10 +83,57 @@ async def download(
 @router.api_route("/getCoverArt.view", methods=["GET", "POST"])
 async def get_cover_art(
     _user: SubsonicUser,
+    album_repo: AlbumRepoDep,
+    track_repo: TrackRepoDep,
+    storage: FileStorageDep,
     id: Annotated[str, Query()],
     size: Annotated[int | None, Query()] = None,
 ) -> Response:
-    # Validate the id shape so clients get a clean error on garbage, then serve a
+    # Cover ids reuse the entity id: ``al-<uuid>`` (album) or ``tr-<uuid>``
-    # placeholder. TODO: stream real covers once the cover pipeline exists.
+    # (track → its album). Unlike the native API, Subsonic clients expect an
-    parse(id)
+    # image either way, so a missing cover falls back to a placeholder rather
+    # than 404. ``size`` is accepted but ignored (we serve the stored image).
+    kind, value = parse(id)
+    album: Album | None
+    if kind is IdKind.ALBUM:
+        album = await album_repo.get_by_id(value)
+    elif kind is IdKind.TRACK:
+        album = await resolve_album_for_track(track_repo, album_repo, value)
+    else:
+        album = None
+
+    if album is not None and album.cover_path:
+        try:
+            return await stream_cover(storage, album.cover_path)
+        except (NotFoundError, StorageError):
+            pass
     return Response(content=_PLACEHOLDER_PNG, media_type="image/png")
+
+
+@router.api_route("/getLyricsBySongId", methods=["GET", "POST"])
+@router.api_route("/getLyricsBySongId.view", methods=["GET", "POST"])
+async def get_lyrics_by_song_id(
+    _user: SubsonicUser,
+    fmt: SubsonicFormat,
+    lyrics_service: LyricsServiceDep,
+    track_repo: TrackRepoDep,
+    artist_repo: ArtistRepoDep,
+    id: Annotated[str, Query()],
+) -> Response:
+    # OpenSubsonic structured lyrics over the native LyricsService (§6.7). A miss
+    # is a normal empty ``lyricsList`` — the service degrades to not_found rather
+    # than raising, so clients get 200 either way.
+    track_id = decode_track(id)
+    track = await track_repo.get_by_id(track_id)
+    if track is None:
+        raise NotFoundError("Song not found.")
+    lyrics = await lyrics_service.get_lyrics(track_id)
+    artist = await artist_repo.get_by_id(track.artist_id)
+    return subsonic_response(
+        structured_lyrics(
+            lyrics,
+            display_artist=artist.name if artist is not None else "",
+            display_title=track.title,
+        ),
+        fmt=fmt,
+    )

app/api/rest/serializers.py

@@ -6,10 +6,17 @@ JSON equivalents). No business logic — they only reshape and rename.
 """
 
 import datetime as dt
+import re
 from typing import Any
 
 from app.api.rest.ids import encode_album, encode_artist, encode_track
 from app.domain.entities import Album, Artist, Track
+from app.domain.entities.lyrics import Lyrics
+
+# One LRC timecode: ``[mm:ss.xx]`` / ``[mm:ss.xxx]`` (fraction optional). A line
+# may carry several (the same words repeat at multiple times); metadata tags like
+# ``[ar:..]`` don't match, so they're ignored.
+_LRC_TAG_RE = re.compile(r"\[(\d+):(\d{1,2})(?:[.:](\d{1,3}))?\]")
 
 # Suffix → MIME, for the ``contentType``/``suffix`` song attributes. A
 # presentation detail (mirrors StreamingService's content-type negotiation).
@@ -80,8 +87,8 @@ def song_dict(
         "albumId": encode_album(track.album_id) if track.album_id is not None else None,
         "artistId": encode_artist(track.artist_id),
         "coverArt": cover,
-        "size": track.file_size,
+        "size": track.file_size or 0,
-        "contentType": content_type_for(track.file_format),
+        "contentType": content_type_for(track.file_format or ""),
         "suffix": track.file_format,
         "duration": track.duration_seconds,
         "year": track.year,
@@ -90,3 +97,52 @@ def song_dict(
         "type": "music",
         "isVideo": False,
     }
+
+
+def _parse_lrc(synced: str) -> list[dict[str, Any]]:
+    """LRC text → OpenSubsonic ``line`` dicts (``start`` in ms, ``value`` text),
+    ordered by time. Lines with no timecode (blank lines, metadata tags) drop out;
+    a timecode carrying several stamps yields one line per stamp."""
+    lines: list[tuple[int, str]] = []
+    for raw in synced.splitlines():
+        stamps = list(_LRC_TAG_RE.finditer(raw))
+        if not stamps:
+            continue
+        text = _LRC_TAG_RE.sub("", raw).strip()
+        for m in stamps:
+            minutes, seconds = int(m.group(1)), int(m.group(2))
+            # LRC fractions are centiseconds (2 digits) or ms (3); pad to ms.
+            ms = int((m.group(3) or "0").ljust(3, "0")[:3])
+            lines.append(((minutes * 60 + seconds) * 1000 + ms, text))
+    lines.sort(key=lambda pair: pair[0])
+    return [{"start": start, "value": text} for start, text in lines]
+
+
+def structured_lyrics(
+    lyrics: Lyrics, *, display_artist: str, display_title: str
+) -> dict[str, Any]:
+    """OpenSubsonic ``getLyricsBySongId`` payload. Prefers synced (LRC) lines and
+    falls back to plain text; an empty ``lyricsList`` when the track has none."""
+    lines: list[dict[str, Any]] = []
+    synced = False
+    if lyrics.synced:
+        lines = _parse_lrc(lyrics.synced)
+        synced = bool(lines)
+    if not lines and lyrics.plain:
+        lines = [{"value": line} for line in lyrics.plain.splitlines()]
+    if not lines:
+        return {"lyricsList": {}}
+    return {
+        "lyricsList": {
+            "structuredLyrics": [
+                {
+                    "displayArtist": display_artist,
+                    "displayTitle": display_title,
+                    "lang": "xxx",
+                    "offset": 0,
+                    "synced": synced,
+                    "line": lines,
+                }
+            ]
+        }
+    }

app/api/schemas/admin.py

@@ -0,0 +1,39 @@
+"""Admin (instance-management) response schemas."""
+
+from pydantic import BaseModel
+
+from app.api.health import CheckStatus
+
+
+class ServicesStatusOut(BaseModel):
+    """Backing-dependency health for the admin dashboard (mirrors readiness)."""
+
+    database: CheckStatus
+    redis: CheckStatus
+    ml: CheckStatus
+
+
+class ReindexJob(BaseModel):
+    source: str
+    job_id: str
+
+
+class ReindexResponse(BaseModel):
+    """The scan jobs enqueued by a re-index, one per indexable source."""
+
+    jobs: list[ReindexJob]
+
+
+class AdminSettingsOut(BaseModel):
+    """Effective, non-secret instance configuration. Secrets and connection
+    strings are never exposed — only whether an optional integration is set up."""
+
+    environment: str
+    allow_registration: bool
+    storage_backend: str
+    media_path: str
+    youtube_enabled: bool
+    coverart_enabled: bool
+    ml_configured: bool
+    acoustid_configured: bool
+    local_import_configured: bool

app/api/schemas/album.py

@@ -13,4 +13,5 @@ class AlbumOut(BaseModel):
     artist_name: str
     year: int | None
     track_count: int
+    has_cover: bool
     created_at: dt.datetime

app/api/schemas/auth.py

@@ -10,6 +10,11 @@ class LoginRequest(BaseModel):
     password: str = Field(min_length=1)
 
 
+class RegisterRequest(BaseModel):
+    username: str = Field(min_length=1, max_length=64)
+    password: str = Field(min_length=8)
+
+
 class RefreshRequest(BaseModel):
     refresh_token: str
 

app/api/schemas/download.py

@@ -0,0 +1,59 @@
+"""Schemas for the download job endpoints (§A5 download manager)."""
+
+import datetime as dt
+import uuid
+
+from pydantic import BaseModel, Field
+
+from app.domain.entities.download import DownloadJob
+
+
+class DownloadCreate(BaseModel):
+    """Request to download an item discovered on a fetch source."""
+
+    source: str
+    source_id: str = Field(min_length=1)
+    # Optional free-text the result came from — stored for display only.
+    query: str | None = None
+
+
+class DownloadJobOut(BaseModel):
+    id: uuid.UUID
+    source: str
+    source_id: str | None
+    query: str | None
+    status: str
+    progress: float
+    error_message: str | None
+    retry_count: int
+    track_id: uuid.UUID | None
+    created_at: dt.datetime
+    updated_at: dt.datetime
+
+    @classmethod
+    def from_entity(cls, job: DownloadJob) -> DownloadJobOut:
+        return cls(
+            id=job.id,
+            source=job.source,
+            source_id=job.source_id,
+            query=job.query,
+            status=job.status,
+            progress=job.progress,
+            error_message=job.error_message,
+            retry_count=job.retry_count,
+            track_id=job.track_id,
+            created_at=job.created_at,
+            updated_at=job.updated_at,
+        )
+
+
+class DownloadCreateResponse(BaseModel):
+    """Result of requesting a download.
+
+    ``already_in_library`` → the item was already imported (``track_id`` set, no
+    job). Otherwise ``job`` describes the queued (or already in-flight) download.
+    """
+
+    already_in_library: bool
+    track_id: uuid.UUID | None
+    job: DownloadJobOut | None

app/api/schemas/external_search.py

@@ -0,0 +1,48 @@
+"""Schemas for searching external (fetch) sources — the §A4 discover screen."""
+
+import uuid
+
+from pydantic import BaseModel
+
+from app.domain.entities.track import Track
+from app.domain.sources import SearchResult
+
+
+class ExternalSearchResultOut(BaseModel):
+    source: str
+    source_id: str
+    title: str
+    artist: str | None
+    album: str | None
+    duration_seconds: int | None
+    thumbnail_url: str | None
+    # Remote browse (plan: Model C) — set when this hit is already saved in the
+    # library, so the UI can show "Play"/"Saved" instead of "Save to library".
+    in_library: bool
+    track_id: uuid.UUID | None
+    availability: str | None
+
+    @classmethod
+    def from_entity(
+        cls, r: SearchResult, *, existing: Track | None = None
+    ) -> ExternalSearchResultOut:
+        return cls(
+            source=r.source,
+            source_id=r.source_id,
+            title=r.title,
+            artist=r.artist,
+            album=r.album,
+            duration_seconds=r.duration_seconds,
+            thumbnail_url=r.thumbnail_url,
+            in_library=existing is not None,
+            track_id=existing.id if existing is not None else None,
+            availability=existing.availability if existing is not None else None,
+        )
+
+
+class ExternalSearchResponse(BaseModel):
+    """Flat list of hits across one or more searchable sources, plus the names of
+    sources that were unavailable (so the UI can show a soft warning)."""
+
+    results: list[ExternalSearchResultOut]
+    searched_sources: list[str]

app/api/schemas/lyrics.py

@@ -0,0 +1,33 @@
+"""Lyrics response schema (§6.7 / Now Playing lyrics panel).
+
+Returns the raw LRC (``synced``) and/or ``plain`` text; the client parses LRC
+timestamps for synced highlighting. A miss is a normal 200 with
+``status="not_found"`` and null text — not an error — so the panel can render a
+"no lyrics" state.
+"""
+
+import uuid
+
+from pydantic import BaseModel
+
+from app.domain.entities.lyrics import Lyrics
+
+
+class LyricsOut(BaseModel):
+    track_id: uuid.UUID
+    status: str
+    source: str | None
+    synced: str | None
+    plain: str | None
+    synced_available: bool
+
+    @classmethod
+    def from_entity(cls, lyrics: Lyrics) -> LyricsOut:
+        return cls(
+            track_id=lyrics.track_id,
+            status=lyrics.status,
+            source=lyrics.source,
+            synced=lyrics.synced,
+            plain=lyrics.plain,
+            synced_available=lyrics.synced is not None,
+        )

app/api/schemas/playlist.py

@@ -29,3 +29,7 @@ class PlaylistUpdate(BaseModel):
 class PlaylistAddTrack(BaseModel):
     track_id: uuid.UUID
     position: float | None = None
+
+
+class PlaylistReorder(BaseModel):
+    track_ids: list[uuid.UUID]

app/api/schemas/radio.py

@@ -0,0 +1,43 @@
+"""Radio + similarity response schemas (§6.5)."""
+
+import uuid
+
+from pydantic import BaseModel, Field
+
+from app.api.schemas.artist import ArtistOut
+from app.api.schemas.track import TrackOut
+
+
+class RadioRequest(BaseModel):
+    """Start or continue a radio. ``seed_track_id`` seeds from a track;
+    ``from_likes`` seeds from the caller's likes. ``exclude_ids`` are already-
+    queued tracks to skip (the client drives the infinite feed). ``exploration``
+    biases familiar↔new."""
+
+    seed_track_id: uuid.UUID | None = None
+    from_likes: bool = False
+    exploration: float = Field(default=0.25, ge=0.0, le=1.0)
+    count: int = Field(default=20, ge=1, le=50)
+    exclude_ids: list[uuid.UUID] = Field(default_factory=list)
+
+
+class RadioTrackOut(BaseModel):
+    track: TrackOut
+    # Short code the client localizes: ml | similar | from_likes | discover.
+    reason: str
+
+
+class RadioResponse(BaseModel):
+    # Where the picks came from: "ml" or "metadata" (fallback).
+    source: str
+    tracks: list[RadioTrackOut]
+
+
+class SimilarTracksOut(BaseModel):
+    source: str
+    tracks: list[TrackOut]
+
+
+class SimilarArtistsOut(BaseModel):
+    source: str
+    artists: list[ArtistOut]

app/api/schemas/settings.py

@@ -0,0 +1,49 @@
+"""User-settings request/response schemas.
+
+Enums are enforced at the API boundary (Pydantic ``Literal`` → 422 on bad
+input), so the service can trust the values it receives.
+"""
+
+from typing import Literal
+
+from pydantic import BaseModel, model_validator
+
+Theme = Literal["system", "light", "dark"]
+# Playback quality preference. ``original`` = no transcode; the lower tiers are
+# consumed by the (upcoming) transcoding pipeline.
+StreamQuality = Literal["original", "high", "medium", "low"]
+ScrobbleProvider = Literal["lastfm", "listenbrainz"]
+
+
+class SettingsOut(BaseModel):
+    theme: Theme
+    stream_quality: StreamQuality
+
+
+class SettingsUpdate(BaseModel):
+    """Partial update — omitted fields keep their current value."""
+
+    theme: Theme | None = None
+    stream_quality: StreamQuality | None = None
+
+
+class ScrobblingOut(BaseModel):
+    enabled: bool
+    provider: ScrobbleProvider | None
+    username: str | None
+    # Whether a session key is stored. The key itself is never returned.
+    configured: bool
+
+
+class ScrobblingUpdate(BaseModel):
+    enabled: bool = False
+    provider: ScrobbleProvider | None = None
+    username: str | None = None
+    # Write-only scrobbler session key / user token. Omit to keep the stored one.
+    session_key: str | None = None
+
+    @model_validator(mode="after")
+    def _provider_required_when_enabled(self) -> ScrobblingUpdate:
+        if self.enabled and self.provider is None:
+            raise ValueError("provider is required when scrobbling is enabled")
+        return self

app/api/schemas/source.py

@@ -0,0 +1,29 @@
+"""Schemas for the source endpoints."""
+
+from pydantic import BaseModel
+
+from app.domain.sources import SourceInfo
+
+
+class SourceInfoOut(BaseModel):
+    name: str
+    label: str
+    kind: str
+    available: bool
+
+    @classmethod
+    def from_entity(cls, info: SourceInfo) -> SourceInfoOut:
+        return cls(name=info.name, label=info.label, kind=info.kind, available=info.available)
+
+
+class ScanResponse(BaseModel):
+    """Result of enqueuing a source scan."""
+
+    source: str
+    job_id: str
+    status: str = "queued"
+
+
+class SourceHealthOut(BaseModel):
+    name: str
+    available: bool

app/api/schemas/storage.py

@@ -0,0 +1,61 @@
+"""Storage / library statistics response schemas (§A6)."""
+
+import datetime as dt
+
+from pydantic import BaseModel
+
+from app.api.schemas.track import TrackOut
+
+
+class DiskUsageOut(BaseModel):
+    total: int
+    used: int
+    free: int
+
+
+class FormatBreakdownOut(BaseModel):
+    file_format: str
+    track_count: int
+    total_size: int
+
+
+class GenreCountOut(BaseModel):
+    genre: str
+    track_count: int
+
+
+class StorageStatsOut(BaseModel):
+    """Everything the Storage screen needs in a single call."""
+
+    # library catalogue
+    total_tracks: int
+    total_artists: int
+    total_albums: int
+    total_size: int
+    total_duration_seconds: int
+    largest_track_size: int
+    earliest_added: dt.datetime | None
+    latest_added: dt.datetime | None
+
+    # breakdowns
+    by_format: list[FormatBreakdownOut]
+    by_metadata_status: dict[str, int]
+    by_source: dict[str, int]
+    top_genres: list[GenreCountOut]
+
+    # backing volume (``None`` for object-store backends)
+    disk: DiskUsageOut | None
+
+
+class DuplicateGroupOut(BaseModel):
+    """Tracks sharing one acoustic fingerprint — candidates for de-duplication."""
+
+    fingerprint: str
+    tracks: list[TrackOut]
+
+
+class CleanupEnqueuedOut(BaseModel):
+    """Acknowledgement that a cleanup job was queued (it runs in the worker)."""
+
+    status: str
+    job_id: str

app/api/schemas/sync.py

@@ -0,0 +1,82 @@
+"""Offline-first sync schemas (delta pull + idempotent push).
+
+The client keeps an opaque ``cursor`` (a server-clock timestamp). It pulls
+everything changed in the half-open window ``(since, cursor]`` and pushes the
+append-only events it accumulated offline. Events carry a client-generated
+``id`` so a replay is idempotent.
+"""
+
+import datetime as dt
+import uuid
+from typing import Literal
+
+from pydantic import BaseModel, Field
+
+from app.api.schemas.track import TrackOut
+
+LikeValue = Literal["like", "dislike", "neutral"]
+
+
+# -- pull (server -> client) --------------------------------------------------
+class LikeEventOut(BaseModel):
+    id: uuid.UUID
+    track_id: uuid.UUID
+    value: str
+    created_at: dt.datetime
+
+
+class PlayEventOut(BaseModel):
+    id: uuid.UUID
+    track_id: uuid.UUID
+    played_at: dt.datetime
+    play_duration_seconds: int | None
+    completed: bool
+
+
+class PlaylistSyncOut(BaseModel):
+    id: uuid.UUID
+    name: str
+    description: str | None
+    version: int
+    updated_at: dt.datetime
+    track_ids: list[uuid.UUID]
+
+
+class SyncChangesOut(BaseModel):
+    """Everything that changed for the caller since their last cursor. Feed
+    ``cursor`` back as ``?since=`` on the next pull."""
+
+    cursor: dt.datetime
+    likes: list[LikeEventOut]
+    plays: list[PlayEventOut]
+    playlists: list[PlaylistSyncOut]
+    tracks: list[TrackOut]
+
+
+# -- push (client -> server) --------------------------------------------------
+class LikeEventIn(BaseModel):
+    id: uuid.UUID
+    track_id: uuid.UUID
+    value: LikeValue
+    created_at: dt.datetime
+
+
+class PlayEventIn(BaseModel):
+    id: uuid.UUID
+    track_id: uuid.UUID
+    played_at: dt.datetime
+    play_duration_seconds: int | None = None
+    completed: bool = False
+
+
+class SyncPushIn(BaseModel):
+    likes: list[LikeEventIn] = Field(default_factory=list)
+    plays: list[PlayEventIn] = Field(default_factory=list)
+
+
+class SyncPushOut(BaseModel):
+    """How many events were newly stored (a replay reports 0) + a fresh cursor."""
+
+    cursor: dt.datetime
+    accepted_likes: int
+    accepted_plays: int

app/api/schemas/track.py

@@ -3,7 +3,9 @@
 import datetime as dt
 import uuid
 
-from pydantic import BaseModel
+from pydantic import BaseModel, Field
+
+from app.api.schemas.download import DownloadJobOut
 
 
 class TrackOut(BaseModel):
@@ -14,10 +16,17 @@ class TrackOut(BaseModel):
     album_id: uuid.UUID | None
     album_title: str | None
     duration_seconds: int | None
-    file_format: str
+    file_format: str | None
-    file_size: int
+    file_size: int | None
+    genre: str | None
+    year: int | None
+    track_number: int | None
     metadata_status: str
+    metadata_error: str | None
+    enriched_at: dt.datetime | None
+    availability: str
     source: str
+    has_cover: bool
     created_at: dt.datetime
 
 
@@ -25,3 +34,54 @@ class TrackUpdate(BaseModel):
     title: str | None = None
     genre: str | None = None
     year: int | None = None
+
+
+class MetadataMatch(BaseModel):
+    """One AcoustID candidate for the metadata editor's match picker (§A7)."""
+
+    acoustid: str
+    score: float
+    recording_mbid: str | None
+    release_group_mbid: str | None
+    title: str | None
+    artist: str | None
+    album: str | None
+    year: int | None
+
+
+class MetadataMatchesOut(BaseModel):
+    items: list[MetadataMatch]
+
+
+class MetadataApply(BaseModel):
+    """Manual edits / accepted match applied via ``PUT /tracks/{id}/metadata``.
+
+    Sets ``metadata_status = manual`` (never overwritten by auto-enrichment)."""
+
+    title: str | None = None
+    artist_name: str | None = None
+    album_title: str | None = None
+    year: int | None = None
+    genre: str | None = None
+    track_number: int | None = None
+
+
+class RemoteTrackSave(BaseModel):
+    """Save a remote browse hit (§A4 discover) as a library placeholder —
+    ``availability="remote"``, no audio until first play (plan: Model C)."""
+
+    source: str
+    source_id: str = Field(min_length=1)
+    title: str
+    artist: str | None = None
+
+
+class MaterializeResponse(BaseModel):
+    """Result of requesting that a placeholder track's audio be fetched.
+
+    ``job`` is ``None`` when the track is already ``local`` — nothing to wait
+    for, the caller can stream immediately. Otherwise it's the (new or
+    already in-flight) job; poll ``GET /downloads/{job.id}`` until ``done``."""
+
+    track: TrackOut
+    job: DownloadJobOut | None

app/api/schemas/transcode.py

@@ -0,0 +1,11 @@
+"""Transcode/optimize response schemas (§6.6)."""
+
+from pydantic import BaseModel
+
+
+class OptimizeEnqueuedOut(BaseModel):
+    """Acknowledgement that a transcode job was queued (it runs in the worker)."""
+
+    status: str
+    job_id: str
+    quality: str

app/api/v1/__init__.py

@@ -2,6 +2,7 @@
 
 from fastapi import APIRouter
 
+from app.api.health import router as health_router
 from app.api.v1.admin import router as admin_router
 from app.api.v1.albums import router as albums_router
 from app.api.v1.artists import router as artists_router
@@ -22,6 +23,9 @@ from app.api.v1.user_settings import router as user_settings_router
 from app.api.v1.users import router as users_router
 
 api_v1_router = APIRouter(prefix="/api/v1")
+# Also expose health under /api/v1 (root /health stays in main.py for compose/nginx
+# probes); the webui pings ${apiBase}/health and apiBase is /api/v1.
+api_v1_router.include_router(health_router)
 api_v1_router.include_router(auth_router)
 api_v1_router.include_router(users_router)
 api_v1_router.include_router(tracks_router)

app/api/v1/admin.py

@@ -5,11 +5,18 @@ sign-up (plan §6.4).
 """
 
 import uuid
-from typing import Any
 
 from fastapi import APIRouter, Query, status
 
-from app.api.deps import SubsonicAuthServiceDep, SuperUser, UserServiceDep
+from app.api.deps import SourceRegistryDep, SubsonicAuthServiceDep, SuperUser, UserServiceDep
+from app.api.health import _check_db, _check_ml, _check_redis
+from app.api.schemas.admin import (
+    AdminSettingsOut,
+    ReindexJob,
+    ReindexResponse,
+    ServicesStatusOut,
+)
+from app.api.schemas.source import SourceInfoOut
 from app.api.schemas.subsonic import SubsonicPasswordResponse
 from app.api.schemas.user import (
     CreateUserRequest,
@@ -17,6 +24,9 @@ from app.api.schemas.user import (
     UpdateUserRequest,
     UserResponse,
 )
+from app.core.config import get_settings
+from app.domain.errors import DependencyUnavailableError, NotSupportedError
+from app.workers.queue import enqueue
 
 router = APIRouter(prefix="/admin", tags=["admin"])
 
@@ -91,24 +101,74 @@ async def rotate_user_subsonic_password(
 
 
 @router.get("/services")
-async def list_services(_admin: SuperUser) -> Any: ...
+async def list_services(_admin: SuperUser) -> ServicesStatusOut:
+    """Backing-dependency health for the admin dashboard — same probes as the
+    readiness endpoint (DB + Redis required, ML optional)."""
+    database = await _check_db()
+    redis = await _check_redis()
+    ml = await _check_ml()
+    return ServicesStatusOut(database=database, redis=redis, ml=ml)
 
 
 @router.get("/sources")
-async def list_admin_sources(_admin: SuperUser) -> Any: ...
+async def list_admin_sources(
-
+    _admin: SuperUser, registry: SourceRegistryDep
-
+) -> list[SourceInfoOut]:
-@router.patch("/sources/{source}")
+    """Configured sources and their live availability (same view as
-async def update_admin_source(source: str, _admin: SuperUser) -> Any: ...
+    ``/sources``, admin-scoped)."""
+    return [SourceInfoOut.from_entity(info) for info in registry.infos()]
 
 
 @router.post("/reindex")
-async def trigger_reindex(_admin: SuperUser) -> Any: ...
+async def trigger_reindex(admin: SuperUser, registry: SourceRegistryDep) -> ReindexResponse:
+    """Enqueue a full re-scan of every indexable source. The walk + file copies
+    run in the worker (never the request cycle); re-scans are idempotent."""
+    indexables = registry.indexables()
+    if not indexables:
+        raise DependencyUnavailableError("No indexable source is configured.")
+    jobs: list[ReindexJob] = []
+    for backend in indexables:
+        job_id = await enqueue("scan_local_folder", source=backend.name, added_by=str(admin.id))
+        jobs.append(ReindexJob(source=backend.name, job_id=job_id))
+    return ReindexResponse(jobs=jobs)
 
 
 @router.get("/settings")
-async def get_admin_settings(_admin: SuperUser) -> Any: ...
+async def get_admin_settings(_admin: SuperUser) -> AdminSettingsOut:
+    """Effective, non-secret instance configuration. Reflects the environment the
+    process booted with; secrets/connection strings are never returned — only
+    whether each optional integration is configured."""
+    settings = get_settings()
+    return AdminSettingsOut(
+        environment=settings.environment,
+        allow_registration=settings.allow_registration,
+        storage_backend=settings.storage_backend,
+        media_path=str(settings.media_path),
+        youtube_enabled=settings.youtube_enabled,
+        coverart_enabled=settings.coverart_enabled,
+        ml_configured=settings.ml_service_url is not None,
+        acoustid_configured=settings.acoustid_api_key is not None,
+        local_import_configured=settings.local_media_import_path is not None,
+    )
+
+
+# -- runtime config mutation (intentionally unsupported) ----------------------
+# The instance is env-configured (CLAUDE.md: nothing hardcoded, all from env) and
+# get_settings() is a cached singleton, so config is not mutable at runtime.
+# These endpoints answer 501 with a clear reason rather than silently no-op'ing;
+# a persistent override layer that shadows env would be a deliberate future
+# departure. Read the effective config via GET /admin/settings.
+@router.patch("/sources/{source}")
+async def update_admin_source(source: str, _admin: SuperUser) -> None:
+    raise NotSupportedError(
+        "Sources are configured via environment variables (e.g. YOUTUBE_ENABLED, "
+        "LOCAL_MEDIA_IMPORT_PATH); runtime changes are not supported."
+    )
 
 
 @router.patch("/settings")
-async def update_admin_settings(_admin: SuperUser) -> Any: ...
+async def update_admin_settings(_admin: SuperUser) -> None:
+    raise NotSupportedError(
+        "Instance settings are managed via environment configuration; "
+        "runtime changes are not supported."
+    )

app/api/v1/albums.py

@@ -1,11 +1,19 @@
 """Album endpoints."""
 
 import uuid
-from typing import Any
 
 from fastapi import APIRouter, Query
+from fastapi.responses import StreamingResponse
 
-from app.api.deps import AlbumRepoDep, ArtistRepoDep, CurrentUser, TrackRepoDep
+from app.api.covers import stream_cover
+from app.api.deps import (
+    AlbumRepoDep,
+    ArtistRepoDep,
+    CurrentUser,
+    FileStorageDep,
+    StreamUser,
+    TrackRepoDep,
+)
 from app.api.schemas.album import AlbumOut
 from app.api.schemas.pagination import PagedResponse
 from app.api.schemas.track import TrackOut
@@ -30,6 +38,7 @@ async def _build_album_out(
             artist_name=artists[a.artist_id].name if a.artist_id in artists else "Unknown Artist",
             year=a.year,
             track_count=track_counts.get(a.id, 0),
+            has_cover=bool(a.cover_path),
             created_at=a.created_at,
         )
         for a in albums
@@ -109,4 +118,14 @@ async def get_album_tracks(
 
 
 @router.get("/{album_id}/cover")
-async def get_album_cover(album_id: uuid.UUID, _: CurrentUser) -> Any: ...
+async def get_album_cover(
+    album_id: uuid.UUID,
+    album_repo: AlbumRepoDep,
+    storage: FileStorageDep,
+    _: StreamUser,
+) -> StreamingResponse:
+    # ``<img>`` can't send a bearer header → StreamUser accepts ``?token=``.
+    album = await album_repo.get_by_id(album_id)
+    if album is None or not album.cover_path:
+        raise NotFoundError("Cover not found.")
+    return await stream_cover(storage, album.cover_path)

app/api/v1/artists.py

@@ -1,14 +1,20 @@
 """Artist endpoints."""
 
 import uuid
-from typing import Any
 
 from fastapi import APIRouter, Query
 
-from app.api.deps import AlbumRepoDep, ArtistRepoDep, CurrentUser, TrackRepoDep
+from app.api.deps import (
+    AlbumRepoDep,
+    ArtistRepoDep,
+    CurrentUser,
+    RecommendationServiceDep,
+    TrackRepoDep,
+)
 from app.api.schemas.album import AlbumOut
 from app.api.schemas.artist import ArtistOut
 from app.api.schemas.pagination import PagedResponse
+from app.api.schemas.radio import SimilarArtistsOut
 from app.api.schemas.track import TrackOut
 from app.api.v1.albums import _build_album_out
 from app.api.v1.tracks import _build_track_out
@@ -124,4 +130,24 @@ async def get_artist_tracks(
 
 
 @router.get("/{artist_id}/similar")
-async def get_similar_artists(artist_id: uuid.UUID, _: CurrentUser) -> Any: ...
+async def get_similar_artists(
+    artist_id: uuid.UUID,
+    service: RecommendationServiceDep,
+    artist_repo: ArtistRepoDep,
+    _: CurrentUser,
+    limit: int = Query(20, ge=1, le=100),
+) -> SimilarArtistsOut:
+    """Artists similar to this one (§6.5). ML when configured, else a shared-
+    genre metadata heuristic."""
+    source, artists = await service.similar_artists(artist_id, limit=limit)
+    items = [
+        ArtistOut(
+            id=a.id,
+            name=a.name,
+            album_count=await artist_repo.album_count(a.id),
+            track_count=await artist_repo.track_count(a.id),
+            created_at=a.created_at,
+        )
+        for a in artists
+    ]
+    return SimilarArtistsOut(source=source, artists=items)

app/api/v1/auth.py

@@ -2,9 +2,16 @@
 
 from fastapi import APIRouter, status
 
-from app.api.deps import AuthServiceDep, CurrentUser
+from app.api.deps import AuthServiceDep, CurrentUser, UserServiceDep
-from app.api.schemas.auth import LoginRequest, RefreshRequest, TokenResponse
+from app.api.schemas.auth import (
+    LoginRequest,
+    RefreshRequest,
+    RegisterRequest,
+    TokenResponse,
+)
 from app.api.schemas.user import UserResponse
+from app.core.config import get_settings
+from app.domain.errors import PermissionDeniedError
 from app.domain.tokens import TokenPair
 
 router = APIRouter(prefix="/auth", tags=["auth"])
@@ -23,6 +30,22 @@ async def login(body: LoginRequest, auth: AuthServiceDep) -> TokenResponse:
     return _to_token_response(pair)
 
 
+@router.post("/register", response_model=TokenResponse, status_code=status.HTTP_201_CREATED)
+async def register(
+    body: RegisterRequest, users: UserServiceDep, auth: AuthServiceDep
+) -> TokenResponse:
+    """Public self-service sign-up (gated by ``ALLOW_REGISTRATION``).
+
+    Registered accounts are always regular users — superusers are created
+    admin-only. On success the new account is logged straight in.
+    """
+    if not get_settings().allow_registration:
+        raise PermissionDeniedError("Registration is disabled on this instance.")
+    await users.create_user(username=body.username, password=body.password, is_superuser=False)
+    pair = await auth.login(body.username, body.password)
+    return _to_token_response(pair)
+
+
 @router.post("/refresh", response_model=TokenResponse)
 async def refresh(body: RefreshRequest, auth: AuthServiceDep) -> TokenResponse:
     pair = await auth.refresh(body.refresh_token)

app/api/v1/downloads.py

@@ -1,36 +1,78 @@
-"""Download job endpoints. Heavy work is dispatched to arq workers."""
+"""Download job endpoints (§A5). Heavy work is dispatched to arq workers — these
+handlers only create/inspect/cancel/retry job records."""
 
 import uuid
-from typing import Any
 
-from fastapi import APIRouter
+from fastapi import APIRouter, Query, Response
+
+from app.api.deps import CurrentUser, DownloadServiceDep
+from app.api.schemas.download import DownloadCreate, DownloadCreateResponse, DownloadJobOut
+from app.api.schemas.pagination import PagedResponse
 
 router = APIRouter(prefix="/downloads", tags=["downloads"])
 
 
 @router.get("")
-async def list_downloads() -> Any: ...
+async def list_downloads(
+    service: DownloadServiceDep,
+    user: CurrentUser,
+    status: str | None = Query(default=None),
+    mine: bool = Query(default=False),
+    limit: int = Query(50, ge=1, le=200),
+    offset: int = Query(0, ge=0),
+) -> PagedResponse[DownloadJobOut]:
+    jobs, total = await service.list(
+        requested_by=user.id if mine else None,
+        status=status,
+        limit=limit,
+        offset=offset,
+    )
+    return PagedResponse(
+        items=[DownloadJobOut.from_entity(j) for j in jobs],
+        total=total,
+        limit=limit,
+        offset=offset,
+    )
 
 
-@router.post("")
+@router.post("", status_code=202)
-async def create_download() -> Any: ...
+async def create_download(
+    body: DownloadCreate,
+    service: DownloadServiceDep,
+    user: CurrentUser,
+) -> DownloadCreateResponse:
+    result = await service.request(
+        source=body.source,
+        source_id=body.source_id,
+        query=body.query,
+        requested_by=user.id,
+    )
+    return DownloadCreateResponse(
+        already_in_library=result.already_in_library,
+        track_id=result.track_id,
+        job=DownloadJobOut.from_entity(result.job) if result.job is not None else None,
+    )
 
 
 @router.get("/{job_id}")
-async def get_download(job_id: uuid.UUID) -> Any: ...
+async def get_download(
+    job_id: uuid.UUID, service: DownloadServiceDep, _: CurrentUser
+) -> DownloadJobOut:
+    job = await service.get(job_id)
+    return DownloadJobOut.from_entity(job)
 
 
-@router.delete("/{job_id}")
+@router.delete("/{job_id}", status_code=204)
-async def cancel_download(job_id: uuid.UUID) -> Any: ...
+async def cancel_download(
+    job_id: uuid.UUID, service: DownloadServiceDep, _: CurrentUser
+) -> Response:
+    await service.cancel(job_id)
+    return Response(status_code=204)
 
 
 @router.post("/{job_id}/retry")
-async def retry_download(job_id: uuid.UUID) -> Any: ...
+async def retry_download(
-
+    job_id: uuid.UUID, service: DownloadServiceDep, _: CurrentUser
-
+) -> DownloadJobOut:
-@router.post("/pause")
+    job = await service.retry(job_id)
-async def pause_downloads() -> Any: ...
+    return DownloadJobOut.from_entity(job)
-
-
-@router.post("/resume")
-async def resume_downloads() -> Any: ...

app/api/v1/playlists.py

@@ -1,23 +1,32 @@
 """Playlist endpoints."""
 
 import uuid
-from typing import Any
 
 from fastapi import APIRouter, Query, Response
+from fastapi.responses import StreamingResponse
 
+from app.api.covers import stream_cover
 from app.api.deps import (
     AlbumRepoDep,
     ArtistRepoDep,
     CurrentUser,
+    FileStorageDep,
     PlaylistRepoDep,
+    StreamUser,
     TrackRepoDep,
 )
 from app.api.schemas.pagination import PagedResponse
-from app.api.schemas.playlist import PlaylistAddTrack, PlaylistCreate, PlaylistOut, PlaylistUpdate
+from app.api.schemas.playlist import (
+    PlaylistAddTrack,
+    PlaylistCreate,
+    PlaylistOut,
+    PlaylistReorder,
+    PlaylistUpdate,
+)
 from app.api.schemas.track import TrackOut
 from app.api.v1.tracks import _build_track_out
 from app.domain.entities.playlist import Playlist
-from app.domain.errors import NotFoundError, PermissionDeniedError
+from app.domain.errors import NotFoundError, PermissionDeniedError, ValidationError
 from app.infrastructure.db.repositories.playlist_repository import SqlAlchemyPlaylistRepository
 
 router = APIRouter(prefix="/playlists", tags=["playlists"])
@@ -182,8 +191,43 @@ async def remove_playlist_track(
 
 
 @router.put("/{playlist_id}/tracks/reorder")
-async def reorder_playlist_tracks(playlist_id: uuid.UUID, _: CurrentUser) -> Any: ...
+async def reorder_playlist_tracks(
+    playlist_id: uuid.UUID,
+    body: PlaylistReorder,
+    playlist_repo: PlaylistRepoDep,
+    user: CurrentUser,
+) -> PlaylistOut:
+    playlist = await playlist_repo.get_by_id(playlist_id)
+    if playlist is None:
+        raise NotFoundError(f"Playlist {playlist_id} not found.")
+    if playlist.owner_id != user.id:
+        raise PermissionDeniedError("You don't own this playlist.")
+
+    total = await playlist_repo.get_track_total(playlist_id)
+    current_tracks = (
+        await playlist_repo.get_tracks(playlist_id, limit=total, offset=0) if total else []
+    )
+    current_ids = {t.id for t in current_tracks}
+    given_ids = body.track_ids
+    if len(given_ids) != len(set(given_ids)) or set(given_ids) != current_ids:
+        raise ValidationError("track_ids must be a permutation of the playlist's current tracks.")
+
+    await playlist_repo.reorder_tracks(playlist_id, given_ids)
+    updated = await playlist_repo.get_by_id(playlist_id)
+    assert updated is not None
+    items = await _build_playlist_out([updated], playlist_repo)
+    return items[0]
 
 
 @router.get("/{playlist_id}/cover")
-async def get_playlist_cover(playlist_id: uuid.UUID, _: CurrentUser) -> Any: ...
+async def get_playlist_cover(
+    playlist_id: uuid.UUID,
+    playlist_repo: PlaylistRepoDep,
+    storage: FileStorageDep,
+    _: StreamUser,
+) -> StreamingResponse:
+    # ``<img>`` can't send a bearer header → StreamUser accepts ``?token=``.
+    cover_path = await playlist_repo.get_cover_path(playlist_id)
+    if not cover_path:
+        raise NotFoundError("Cover not found.")
+    return await stream_cover(storage, cover_path)

app/api/v1/radio.py

@@ -1,15 +1,83 @@
-"""Radio / continuous-mix endpoints. Degrades gracefully when ML service is down."""
+"""Radio / continuous-mix endpoints (§6.5).
 
-from typing import Any
+Stateless: the client passes the seed + already-queued ids and pulls more as the
+queue drains (offline-first infinite feed). Degrades gracefully when no ML
+service is configured — the recommendation service falls back to metadata.
+"""
 
 from fastapi import APIRouter
 
+from app.api.deps import (
+    AlbumRepoDep,
+    ArtistRepoDep,
+    CurrentUser,
+    RecommendationServiceDep,
+)
+from app.api.schemas.radio import RadioRequest, RadioResponse, RadioTrackOut
+from app.api.v1.tracks import _build_track_out
+from app.application.recommendation_service import RadioPick
+
 router = APIRouter(prefix="/radio", tags=["radio"])
 
 
+async def _to_response(
+    source: str,
+    picks: list[RadioPick],
+    artist_repo: ArtistRepoDep,
+    album_repo: AlbumRepoDep,
+) -> RadioResponse:
+    tracks = [p.track for p in picks]
+    artist_ids = list({t.artist_id for t in tracks})
+    album_ids = list({t.album_id for t in tracks if t.album_id is not None})
+    artists = {a.id: a for a in await artist_repo.get_many(artist_ids)}
+    albums = {a.id: a for a in await album_repo.get_many(album_ids)}
+    outs = await _build_track_out(tracks, artists, albums)
+    return RadioResponse(
+        source=source,
+        tracks=[
+            RadioTrackOut(track=out, reason=pick.reason)
+            for out, pick in zip(outs, picks, strict=True)
+        ],
+    )
+
+
+async def _run_radio(
+    body: RadioRequest,
+    user: CurrentUser,
+    service: RecommendationServiceDep,
+    artist_repo: ArtistRepoDep,
+    album_repo: AlbumRepoDep,
+) -> RadioResponse:
+    source, picks = await service.radio(
+        user_id=user.id,
+        seed_track_id=body.seed_track_id,
+        from_likes=body.from_likes,
+        exploration=body.exploration,
+        limit=body.count,
+        exclude_ids=body.exclude_ids,
+    )
+    return await _to_response(source, picks, artist_repo, album_repo)
+
+
 @router.post("")
-async def start_radio() -> Any: ...
+async def start_radio(
+    body: RadioRequest,
+    user: CurrentUser,
+    service: RecommendationServiceDep,
+    artist_repo: ArtistRepoDep,
+    album_repo: AlbumRepoDep,
+) -> RadioResponse:
+    """Start a radio from a seed track or the caller's likes."""
+    return await _run_radio(body, user, service, artist_repo, album_repo)
 
 
 @router.post("/next")
-async def next_radio_track() -> Any: ...
+async def next_radio_track(
+    body: RadioRequest,
+    user: CurrentUser,
+    service: RecommendationServiceDep,
+    artist_repo: ArtistRepoDep,
+    album_repo: AlbumRepoDep,
+) -> RadioResponse:
+    """Fetch more tracks as the radio queue drains (pass ``exclude_ids``)."""
+    return await _run_radio(body, user, service, artist_repo, album_repo)

app/api/v1/search.py

@@ -1,12 +1,11 @@
 """Search endpoints: global and library-scoped."""
 
-from typing import Any
-
 from fastapi import APIRouter, Query
 
-from app.api.deps import AlbumRepoDep, ArtistRepoDep, CurrentUser, TrackRepoDep
+from app.api.deps import AlbumRepoDep, ArtistRepoDep, CurrentUser, SourceRegistryDep, TrackRepoDep
 from app.api.schemas.album import AlbumOut
 from app.api.schemas.artist import ArtistOut
+from app.api.schemas.external_search import ExternalSearchResponse, ExternalSearchResultOut
 from app.api.schemas.search import LibrarySearchResponse
 from app.api.schemas.track import TrackOut
 from app.api.v1.albums import _build_album_out
@@ -16,7 +15,31 @@ router = APIRouter(prefix="/search", tags=["search"])
 
 
 @router.get("")
-async def search(_: CurrentUser) -> Any: ...
+async def search(
+    _: CurrentUser,
+    registry: SourceRegistryDep,
+    track_repo: TrackRepoDep,
+    q: str = Query(min_length=1),
+    limit: int = Query(20, ge=1, le=50),
+) -> ExternalSearchResponse:
+    """Search every available fetch source and merge the hits (§A4 discover).
+
+    A source that is down contributes nothing rather than failing the whole
+    request (graceful degradation); only available sources are reported as
+    searched. Each hit is checked against the library by ``(source,
+    source_id)`` so the UI can show "Saved"/"Play" instead of "Save to
+    library" without a separate round-trip (remote browse, plan: Model C)."""
+    results: list[ExternalSearchResultOut] = []
+    searched: list[str] = []
+    for backend in registry.searchables():
+        if not backend.is_available():
+            continue
+        searched.append(backend.name)
+        hits = await backend.search(q, limit=limit)
+        for h in hits:
+            existing = await track_repo.get_by_source(h.source, h.source_id)
+            results.append(ExternalSearchResultOut.from_entity(h, existing=existing))
+    return ExternalSearchResponse(results=results, searched_sources=searched)
 
 
 @router.get("/library")

app/api/v1/sources.py

@@ -1,19 +1,57 @@
-"""External source endpoints (yt-dlp etc.)."""
+"""External source endpoints: enumerate sources, search, and trigger imports.
 
-from typing import Any
+Listing/health/search are read-only (any authenticated user). Scanning a source
+is an admin action and runs in a worker — the endpoint only enqueues it.
+"""
 
-from fastapi import APIRouter
+from fastapi import APIRouter, Query
+
+from app.api.deps import CurrentUser, SourceRegistryDep, SuperUser, TrackRepoDep
+from app.api.schemas.external_search import ExternalSearchResponse, ExternalSearchResultOut
+from app.api.schemas.source import ScanResponse, SourceHealthOut, SourceInfoOut
+from app.domain.errors import DependencyUnavailableError
+from app.workers.queue import enqueue
 
 router = APIRouter(prefix="/sources", tags=["sources"])
 
 
 @router.get("")
-async def list_sources() -> Any: ...
+async def list_sources(_: CurrentUser, registry: SourceRegistryDep) -> list[SourceInfoOut]:
+    return [SourceInfoOut.from_entity(info) for info in registry.infos()]
 
 
-@router.get("/{source}/search")
+@router.post("/{source}/scan")
-async def search_source(source: str) -> Any: ...
+async def scan_source(source: str, user: SuperUser, registry: SourceRegistryDep) -> ScanResponse:
+    backend = registry.indexable(source)  # 404 if unknown, 422 if not indexable
+    if not backend.is_available():
+        raise DependencyUnavailableError(f"Source {source!r} is not available.")
+    job_id = await enqueue("scan_local_folder", source=source, added_by=str(user.id))
+    return ScanResponse(source=source, job_id=job_id)
 
 
 @router.get("/{source}/health")
-async def source_health(source: str) -> Any: ...
+async def source_health(
+    source: str, _: CurrentUser, registry: SourceRegistryDep
+) -> SourceHealthOut:
+    backend = registry.get(source)  # 404 if unknown
+    return SourceHealthOut(name=backend.name, available=backend.is_available())
+
+
+@router.get("/{source}/search")
+async def search_source(
+    source: str,
+    _: CurrentUser,
+    registry: SourceRegistryDep,
+    track_repo: TrackRepoDep,
+    q: str = Query(min_length=1),
+    limit: int = Query(20, ge=1, le=50),
+) -> ExternalSearchResponse:
+    backend = registry.searchable(source)  # 404 if unknown, 422 if not searchable
+    if not backend.is_available():
+        raise DependencyUnavailableError(f"Source {source!r} is not available.")
+    results = await backend.search(q, limit=limit)
+    out: list[ExternalSearchResultOut] = []
+    for r in results:
+        existing = await track_repo.get_by_source(r.source, r.source_id)
+        out.append(ExternalSearchResultOut.from_entity(r, existing=existing))
+    return ExternalSearchResponse(results=out, searched_sources=[source])

app/api/v1/storage.py

@@ -1,27 +1,150 @@
 """Storage analysis and cleanup endpoints."""
 
-from typing import Any
+from fastapi import APIRouter, Query
 
-from fastapi import APIRouter
+from app.api.deps import (
+    AlbumRepoDep,
+    ArtistRepoDep,
+    CurrentUser,
+    FileStorageDep,
+    SuperUser,
+    TrackRepoDep,
+)
+from app.api.schemas.pagination import PagedResponse
+from app.api.schemas.storage import (
+    CleanupEnqueuedOut,
+    DiskUsageOut,
+    DuplicateGroupOut,
+    FormatBreakdownOut,
+    GenreCountOut,
+    StorageStatsOut,
+)
+from app.api.schemas.track import TrackOut
+from app.api.v1.tracks import _build_track_out
+from app.domain.entities.track import Track
+from app.workers.queue import enqueue
 
 router = APIRouter(prefix="/storage", tags=["storage"])
 
+# How many of the most common genres the dashboard surfaces.
+_TOP_GENRES = 8
+
+
+async def _tracks_to_out(
+    tracks: list[Track], artist_repo: ArtistRepoDep, album_repo: AlbumRepoDep
+) -> list[TrackOut]:
+    """Hydrate a batch of tracks into ``TrackOut`` (artist/album names + cover
+    flag), resolving each referenced artist/album in a single query."""
+    artist_ids = list({t.artist_id for t in tracks})
+    album_ids = list({t.album_id for t in tracks if t.album_id is not None})
+    artists = {a.id: a for a in await artist_repo.get_many(artist_ids)}
+    albums = {a.id: a for a in await album_repo.get_many(album_ids)}
+    return await _build_track_out(tracks, artists, albums)
+
 
 @router.get("")
-async def get_storage_stats() -> Any: ...
+async def get_storage_stats(
+    track_repo: TrackRepoDep,
+    artist_repo: ArtistRepoDep,
+    album_repo: AlbumRepoDep,
+    storage: FileStorageDep,
+    _: CurrentUser,
+) -> StorageStatsOut:
+    """Library + disk statistics for the Storage dashboard (§A6).
+
+    Aggregates come from the catalogue (cheap GROUP BYs); ``disk`` reflects the
+    real backing volume and is ``None`` for backends without a fixed-capacity
+    disk (e.g. object stores)."""
+    stats = await track_repo.library_stats()
+    total_artists = await artist_repo.count(q=None)
+    total_albums = await album_repo.count(artist_id=None, q=None)
+    genres = await track_repo.genres()
+    disk = await storage.disk_usage()
+
+    return StorageStatsOut(
+        total_tracks=stats.total_tracks,
+        total_artists=total_artists,
+        total_albums=total_albums,
+        total_size=stats.total_size,
+        total_duration_seconds=stats.total_duration_seconds,
+        largest_track_size=stats.largest_track_size,
+        earliest_added=stats.earliest_added,
+        latest_added=stats.latest_added,
+        by_format=[
+            FormatBreakdownOut(
+                file_format=f.file_format,
+                track_count=f.track_count,
+                total_size=f.total_size,
+            )
+            for f in stats.by_format
+        ],
+        by_metadata_status=stats.by_metadata_status,
+        by_source=stats.by_source,
+        top_genres=[
+            GenreCountOut(genre=genre, track_count=count) for genre, count in genres[:_TOP_GENRES]
+        ],
+        disk=DiskUsageOut(total=disk.total, used=disk.used, free=disk.free) if disk else None,
+    )
 
 
 @router.get("/duplicates")
-async def get_duplicates() -> Any: ...
+async def get_duplicates(
+    track_repo: TrackRepoDep,
+    artist_repo: ArtistRepoDep,
+    album_repo: AlbumRepoDep,
+    _: CurrentUser,
+) -> list[DuplicateGroupOut]:
+    """Tracks sharing an acoustic fingerprint, grouped — the library's real
+    duplicates (``(source, source_id)`` is already unique). Cheap DB GROUP BY."""
+    groups = await track_repo.find_duplicate_groups()
+    all_tracks = [track for _, tracks in groups for track in tracks]
+    out = await _tracks_to_out(all_tracks, artist_repo, album_repo)
+    by_id = {item.id: item for item in out}
+    return [
+        DuplicateGroupOut(fingerprint=fingerprint, tracks=[by_id[t.id] for t in tracks])
+        for fingerprint, tracks in groups
+    ]
 
 
 @router.get("/broken")
-async def get_broken_files() -> Any: ...
+async def get_broken_files(
+    track_repo: TrackRepoDep,
+    artist_repo: ArtistRepoDep,
+    album_repo: AlbumRepoDep,
+    _: CurrentUser,
+    limit: int = Query(50, ge=1, le=200),
+    offset: int = Query(0, ge=0),
+) -> PagedResponse[TrackOut]:
+    """Tracks whose last enrichment run failed (``metadata_status=failed``) —
+    each carries its ``metadata_error``. A file gone missing on disk is instead
+    reconciled by ``POST /storage/cleanup`` (that needs a filesystem scan)."""
+    tracks = await track_repo.list_by_metadata_status("failed", limit=limit, offset=offset)
+    total = await track_repo.count_by_metadata_status("failed")
+    items = await _tracks_to_out(tracks, artist_repo, album_repo)
+    return PagedResponse(items=items, total=total, limit=limit, offset=offset)
 
 
 @router.get("/missing-metadata")
-async def get_missing_metadata() -> Any: ...
+async def get_missing_metadata(
+    track_repo: TrackRepoDep,
+    artist_repo: ArtistRepoDep,
+    album_repo: AlbumRepoDep,
+    _: CurrentUser,
+    limit: int = Query(50, ge=1, le=200),
+    offset: int = Query(0, ge=0),
+) -> PagedResponse[TrackOut]:
+    """Tracks still awaiting enrichment (``metadata_status=pending``) — imported
+    but never identified."""
+    tracks = await track_repo.list_by_metadata_status("pending", limit=limit, offset=offset)
+    total = await track_repo.count_by_metadata_status("pending")
+    items = await _tracks_to_out(tracks, artist_repo, album_repo)
+    return PagedResponse(items=items, total=total, limit=limit, offset=offset)
 
 
-@router.post("/cleanup")
+@router.post("/cleanup", status_code=202)
-async def run_cleanup() -> Any: ...
+async def run_cleanup(_: SuperUser) -> CleanupEnqueuedOut:
+    """Admin: enqueue the storage reconciliation job. It scans the catalogue and
+    removes rows whose backing file has vanished (dangling references). Runs in
+    the worker — the filesystem scan must not block the request cycle."""
+    job_id = await enqueue("cleanup_storage")
+    return CleanupEnqueuedOut(status="enqueued", job_id=job_id)

app/api/v1/streaming.py

@@ -1,30 +1,54 @@
-"""Audio streaming endpoint — direct stream with Range support."""
+"""Audio streaming — direct byte-range stream, transcoded quality, and HLS.
 
+``GET /stream/{id}`` streams the master with Range support, or a cached Opus
+rendition when ``?quality=`` is set (a cache miss falls back to the master and
+warms the cache in the background — playback never waits on ffmpeg). ``/hls/*``
+serves the cached HLS rendition (generated by the ``transcode_track`` worker).
+"""
+
+import re
 import uuid
 from typing import Annotated
 
-from fastapi import APIRouter, Header
+import anyio
-from fastapi.responses import StreamingResponse
+from fastapi import APIRouter, Header, Query, Response
+from fastapi.responses import FileResponse, StreamingResponse
 
-from app.api.deps import StreamingServiceDep, StreamUser
+from app.api.deps import StreamingServiceDep, StreamUser, TranscodeServiceDep
+from app.domain.errors import NotFoundError
+from app.workers.queue import enqueue_transcode_quiet
 
 router = APIRouter(prefix="/stream", tags=["streaming"])
 
+_HLS_PLAYLIST_TYPE = "application/vnd.apple.mpegurl"
+_HLS_SEGMENT_TYPE = "video/mp2t"
+_OPUS_TYPE = "audio/ogg"
+_SEGMENT_LINE_RE = re.compile(r"^(seg_\d+\.ts)$", re.MULTILINE)
+
 
 @router.get("/{track_id}")
 async def stream_track(
     track_id: uuid.UUID,
     service: StreamingServiceDep,
+    transcode: TranscodeServiceDep,
     _user: StreamUser,
     range_header: Annotated[str | None, Header(alias="Range")] = None,
-) -> StreamingResponse:
+    quality: Annotated[str | None, Query()] = None,
+) -> Response:
+    # A quality rendition, if one is cached; otherwise fall back to the master
+    # and enqueue generation so the next play gets it (graceful degradation).
+    if quality and quality != "original":
+        cached = await transcode.resolve_quality_file(track_id, quality)
+        if cached is not None:
+            return FileResponse(cached, media_type=_OPUS_TYPE)
+        await enqueue_transcode_quiet(track_id, quality=quality, hls=False)
+
     result = await service.open_stream(track_id, range_header)
 
     headers = {
         "Accept-Ranges": "bytes",
         "Content-Length": str(result.content_length),
     }
-
     if result.is_partial:
         headers["Content-Range"] = f"bytes {result.start}-{result.end}/{result.total_size}"
         status_code = 206
@@ -37,3 +61,38 @@ async def stream_track(
         headers=headers,
         media_type=result.content_type,
     )
+
+
+@router.get("/{track_id}/hls/playlist.m3u8")
+async def stream_hls_playlist(
+    track_id: uuid.UUID,
+    transcode: TranscodeServiceDep,
+    _user: StreamUser,
+    token: Annotated[str | None, Query()] = None,
+) -> Response:
+    """Serve the cached HLS playlist. On a miss, kick off generation and 404 so
+    the client retries. Segment URLs are relative; when the request carried a
+    ``?token=`` (players can't set an Authorization header), it's appended to
+    each segment line so the segment requests authenticate the same way."""
+    path = await transcode.hls_playlist(track_id)
+    if path is None:
+        await enqueue_transcode_quiet(track_id, hls=True)
+        raise NotFoundError("HLS rendition is being prepared; retry shortly.")
+
+    body = await anyio.to_thread.run_sync(path.read_text)
+    if token:
+        body = _SEGMENT_LINE_RE.sub(rf"\1?token={token}", body)
+    return Response(body, media_type=_HLS_PLAYLIST_TYPE)
+
+
+@router.get("/{track_id}/hls/{segment}")
+async def stream_hls_segment(
+    track_id: uuid.UUID,
+    segment: str,
+    transcode: TranscodeServiceDep,
+    _user: StreamUser,
+) -> FileResponse:
+    path = transcode.hls_segment(track_id, segment)
+    if path is None:
+        raise NotFoundError("Segment not found.")
+    return FileResponse(path, media_type=_HLS_SEGMENT_TYPE)

app/api/v1/sync.py

@@ -1,15 +1,103 @@
-"""Client sync endpoints (offline-first event log)."""
+"""Client sync endpoints (offline-first event log).
 
-from typing import Any
+``GET /sync/changes`` pulls everything the caller changed since their cursor;
+``POST /sync/push`` uploads the like/play events a client accumulated offline
+(idempotent — replays are no-ops). See :mod:`app.application.sync_service`.
+"""
+
+import datetime as dt
 
 from fastapi import APIRouter
 
+from app.api.deps import AlbumRepoDep, ArtistRepoDep, CurrentUser, SyncServiceDep
+from app.api.schemas.sync import (
+    LikeEventOut,
+    PlayEventOut,
+    PlaylistSyncOut,
+    SyncChangesOut,
+    SyncPushIn,
+    SyncPushOut,
+)
+from app.api.v1.tracks import _build_track_out
+from app.application.sync_service import LikeEvent, PlayEvent
+
 router = APIRouter(prefix="/sync", tags=["sync"])
 
 
 @router.get("/changes")
-async def get_changes() -> Any: ...
+async def get_changes(
+    service: SyncServiceDep,
+    artist_repo: ArtistRepoDep,
+    album_repo: AlbumRepoDep,
+    user: CurrentUser,
+    since: dt.datetime | None = None,
+) -> SyncChangesOut:
+    """Delta since ``since`` (omit for a full snapshot). Persist ``cursor`` from
+    the response and pass it back as ``?since=`` next time."""
+    changes = await service.get_changes(user.id, since=since)
+
+    artist_ids = list({t.artist_id for t in changes.tracks})
+    album_ids = list({t.album_id for t in changes.tracks if t.album_id is not None})
+    artists = {a.id: a for a in await artist_repo.get_many(artist_ids)}
+    albums = {a.id: a for a in await album_repo.get_many(album_ids)}
+    tracks_out = await _build_track_out(changes.tracks, artists, albums)
+
+    return SyncChangesOut(
+        cursor=changes.cursor,
+        likes=[
+            LikeEventOut(
+                id=lk.id, track_id=lk.track_id, value=lk.value, created_at=lk.created_at
+            )
+            for lk in changes.likes
+        ],
+        plays=[
+            PlayEventOut(
+                id=p.id,
+                track_id=p.track_id,
+                played_at=p.played_at,
+                play_duration_seconds=p.play_duration_seconds,
+                completed=p.completed,
+            )
+            for p in changes.plays
+        ],
+        playlists=[
+            PlaylistSyncOut(
+                id=d.playlist.id,
+                name=d.playlist.name,
+                description=d.playlist.description,
+                version=d.playlist.version,
+                updated_at=d.playlist.updated_at,
+                track_ids=d.track_ids,
+            )
+            for d in changes.playlists
+        ],
+        tracks=tracks_out,
+    )
 
 
 @router.post("/push")
-async def push_changes() -> Any: ...
+async def push_changes(
+    body: SyncPushIn, service: SyncServiceDep, user: CurrentUser
+) -> SyncPushOut:
+    result = await service.push(
+        user.id,
+        likes=[
+            LikeEvent(id=e.id, track_id=e.track_id, value=e.value, created_at=e.created_at)
+            for e in body.likes
+        ],
+        plays=[
+            PlayEvent(
+                id=e.id,
+                track_id=e.track_id,
+                played_at=e.played_at,
+                play_duration_seconds=e.play_duration_seconds,
+                completed=e.completed,
+            )
+            for e in body.plays
+        ],
+    )
+    return SyncPushOut(
+        cursor=result.cursor,
+        accepted_likes=result.accepted_likes,
+        accepted_plays=result.accepted_plays,
+    )

app/api/v1/tracks.py

@@ -1,16 +1,45 @@
 """Track endpoints."""
 
 import uuid
-from typing import Any
+from typing import Annotated
 
+import anyio
 from fastapi import APIRouter, Query, Response
+from fastapi.responses import StreamingResponse
 
-from app.api.deps import AlbumRepoDep, ArtistRepoDep, CurrentUser, FileStorageDep, TrackRepoDep
+from app.api.covers import resolve_album_for_track, stream_cover
+from app.api.deps import (
+    AlbumRepoDep,
+    ArtistRepoDep,
+    CurrentUser,
+    FileStorageDep,
+    LyricsServiceDep,
+    MetadataServiceDep,
+    RecommendationServiceDep,
+    RemoteLibraryServiceDep,
+    StreamUser,
+    TrackRepoDep,
+)
+from app.api.schemas.download import DownloadJobOut
+from app.api.schemas.lyrics import LyricsOut
 from app.api.schemas.pagination import PagedResponse
-from app.api.schemas.track import TrackOut, TrackUpdate
+from app.api.schemas.radio import SimilarTracksOut
+from app.api.schemas.track import (
+    MaterializeResponse,
+    MetadataApply,
+    MetadataMatch,
+    MetadataMatchesOut,
+    RemoteTrackSave,
+    TrackOut,
+    TrackUpdate,
+)
+from app.api.schemas.transcode import OptimizeEnqueuedOut
+from app.application.transcode_service import bitrate_for_quality, remove_track_cache
+from app.core.config import get_settings
 from app.domain.entities.album import Album
 from app.domain.entities.track import Artist, Track
-from app.domain.errors import NotFoundError
+from app.domain.errors import NotFoundError, ValidationError
+from app.workers.queue import enqueue, enqueue_transcode
 
 router = APIRouter(prefix="/tracks", tags=["tracks"])
 
@@ -31,8 +60,15 @@ async def _build_track_out(
             duration_seconds=t.duration_seconds,
             file_format=t.file_format,
             file_size=t.file_size,
+            genre=t.genre,
+            year=t.year,
+            track_number=t.track_number,
             metadata_status=t.metadata_status,
+            metadata_error=t.metadata_error,
+            enriched_at=t.enriched_at,
+            availability=t.availability,
             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,
         )
         for t in tracks
@@ -48,6 +84,7 @@ async def list_tracks(
     artist_id: uuid.UUID | None = None,
     album_id: uuid.UUID | None = None,
     q: str | None = None,
+    source: str | None = Query(None, max_length=32),
     sort_by: str = Query("created_at", pattern="^(title|created_at|artist)$"),
     order: str = Query("desc", pattern="^(asc|desc)$"),
     limit: int = Query(50, ge=1, le=200),
@@ -57,12 +94,13 @@ async def list_tracks(
         artist_id=artist_id,
         album_id=album_id,
         q=q,
+        source=source,
         sort_by=sort_by,
         order=order,
         limit=limit,
         offset=offset,
     )
-    total = await track_repo.count(artist_id=artist_id, album_id=album_id, q=q)
+    total = await track_repo.count(artist_id=artist_id, album_id=album_id, q=q, source=source)
 
     artist_ids = list({t.artist_id for t in tracks})
     album_ids = list({t.album_id for t in tracks if t.album_id is not None})
@@ -73,6 +111,57 @@ async def list_tracks(
     return PagedResponse(items=items, total=total, limit=limit, offset=offset)
 
 
+@router.post("/remote", status_code=201)
+async def save_remote_track(
+    body: RemoteTrackSave,
+    service: RemoteLibraryServiceDep,
+    artist_repo: ArtistRepoDep,
+    album_repo: AlbumRepoDep,
+    user: CurrentUser,
+) -> TrackOut:
+    """Save a remote browse hit (§A4 discover) as a library placeholder —
+    no audio is fetched yet (plan: Model C). Idempotent on ``(source,
+    source_id)``: saving an already-saved hit returns the existing track."""
+    track = await service.save_remote(
+        source=body.source,
+        source_id=body.source_id,
+        title=body.title,
+        artist=body.artist,
+        added_by=user.id,
+    )
+
+    artists = {a.id: a for a in await artist_repo.get_many([track.artist_id])}
+    album_ids = [track.album_id] if track.album_id else []
+    albums = {a.id: a for a in await album_repo.get_many(album_ids)}
+    items = await _build_track_out([track], artists, albums)
+    return items[0]
+
+
+@router.post("/{track_id}/materialize")
+async def materialize_track(
+    track_id: uuid.UUID,
+    service: RemoteLibraryServiceDep,
+    artist_repo: ArtistRepoDep,
+    album_repo: AlbumRepoDep,
+    user: CurrentUser,
+) -> MaterializeResponse:
+    """Fetch a placeholder track's audio on demand (plan: Model C lazy
+    materialization). Already-local tracks return ``job=None`` — nothing to
+    wait for. Otherwise poll ``GET /downloads/{job.id}`` until ``done``, then
+    stream as usual."""
+    outcome = await service.request_materialize(track_id, requested_by=user.id)
+
+    artists = {a.id: a for a in await artist_repo.get_many([outcome.track.artist_id])}
+    album_ids = [outcome.track.album_id] if outcome.track.album_id else []
+    albums = {a.id: a for a in await album_repo.get_many(album_ids)}
+    track_out = (await _build_track_out([outcome.track], artists, albums))[0]
+
+    return MaterializeResponse(
+        track=track_out,
+        job=DownloadJobOut.from_entity(outcome.job) if outcome.job is not None else None,
+    )
+
+
 @router.get("/{track_id}")
 async def get_track(
     track_id: uuid.UUID,
@@ -130,29 +219,182 @@ async def delete_track(
     if track is None:
         raise NotFoundError(f"Track {track_id} not found.")
     await track_repo.delete(track_id)
-    await storage.delete(track.storage_uri)
+    if track.storage_uri is not None:
+        await storage.delete(track.storage_uri)
+    # Drop any cached transcode renditions (Opus + HLS) so they don't dangle.
+    await anyio.to_thread.run_sync(
+        remove_track_cache, get_settings().transcode_cache_path, track_id
+    )
     return Response(status_code=204)
 
 
 @router.get("/{track_id}/similar")
-async def get_similar_tracks(track_id: uuid.UUID, _: CurrentUser) -> Any: ...
+async def get_similar_tracks(
+    track_id: uuid.UUID,
+    service: RecommendationServiceDep,
+    artist_repo: ArtistRepoDep,
+    album_repo: AlbumRepoDep,
+    _: CurrentUser,
+    limit: Annotated[int, Query(ge=1, le=100)] = 20,
+) -> SimilarTracksOut:
+    """Tracks similar to this one (§6.5). Uses ML when configured, else a
+    genre/artist metadata heuristic."""
+    source, tracks = await service.similar_tracks(track_id, limit=limit)
+    artist_ids = list({t.artist_id for t in tracks})
+    album_ids = list({t.album_id for t in tracks if t.album_id is not None})
+    artists = {a.id: a for a in await artist_repo.get_many(artist_ids)}
+    albums = {a.id: a for a in await album_repo.get_many(album_ids)}
+    outs = await _build_track_out(tracks, artists, albums)
+    return SimilarTracksOut(source=source, tracks=outs)
 
 
-@router.post("/{track_id}/optimize")
+@router.post("/{track_id}/optimize", status_code=202)
-async def optimize_track(track_id: uuid.UUID, _: CurrentUser) -> Any: ...
+async def optimize_track(
+    track_id: uuid.UUID,
+    track_repo: TrackRepoDep,
+    _: CurrentUser,
+    quality: Annotated[str, Query()] = "high",
+) -> OptimizeEnqueuedOut:
+    """Enqueue transcoding of a track into a cached Opus rendition + HLS (§6.6).
+    Heavy ffmpeg work runs in the worker; this only queues it."""
+    if bitrate_for_quality(quality) is None:
+        raise ValidationError(
+            f"Unknown quality '{quality}'; expected one of high, medium, low."
+        )
+    track = await track_repo.get_by_id(track_id)
+    if track is None:
+        raise NotFoundError(f"Track {track_id} not found.")
+    job_id = await enqueue_transcode(track_id, quality=quality, hls=True)
+    return OptimizeEnqueuedOut(status="enqueued", job_id=job_id, quality=quality)
+
+
+@router.get("/{track_id}/lyrics")
+async def get_track_lyrics(
+    track_id: uuid.UUID, lyrics: LyricsServiceDep, _: CurrentUser
+) -> LyricsOut:
+    """Cached lyrics for the Now Playing panel (§6.7). A miss is a normal 200
+    with ``status="not_found"`` — the provider (LRCLIB) is queried at most once,
+    then the outcome is cached."""
+    return LyricsOut.from_entity(await lyrics.get_lyrics(track_id))
+
+
+@router.post("/{track_id}/lyrics/refetch")
+async def refetch_track_lyrics(
+    track_id: uuid.UUID, lyrics: LyricsServiceDep, _: CurrentUser
+) -> LyricsOut:
+    """Force a fresh provider lookup, bypassing the cache (user-triggered)."""
+    return LyricsOut.from_entity(await lyrics.get_lyrics(track_id, force=True))
 
 
 @router.get("/{track_id}/cover")
-async def get_track_cover(track_id: uuid.UUID, _: CurrentUser) -> Any: ...
+async def get_track_cover(
+    track_id: uuid.UUID,
+    track_repo: TrackRepoDep,
+    album_repo: AlbumRepoDep,
+    storage: FileStorageDep,
+    _: StreamUser,
+) -> StreamingResponse:
+    # A track's cover is its album's cover. ``<img>`` can't send a bearer
+    # header → StreamUser accepts ``?token=``.
+    album = await resolve_album_for_track(track_repo, album_repo, track_id)
+    if album is None or not album.cover_path:
+        raise NotFoundError("Cover not found.")
+    return await stream_cover(storage, album.cover_path)
 
 
 @router.post("/{track_id}/metadata/enrich")
-async def enrich_metadata(track_id: uuid.UUID, _: CurrentUser) -> Any: ...
+async def enrich_metadata(
+    track_id: uuid.UUID,
+    track_repo: TrackRepoDep,
+    _: CurrentUser,
+) -> dict[str, str]:
+    """Re-run metadata enrichment for a track (admin/user-triggered). The work
+    happens in a worker; this only enqueues it. 503 if the queue is down."""
+    track = await track_repo.get_by_id(track_id)
+    if track is None:
+        raise NotFoundError(f"Track {track_id} not found.")
+    job_id = await enqueue("enrich_track", track_id=str(track_id))
+    return {"track_id": str(track_id), "job_id": job_id}
 
 
 @router.get("/{track_id}/metadata/matches")
-async def get_metadata_matches(track_id: uuid.UUID, _: CurrentUser) -> Any: ...
+async def get_metadata_matches(
+    track_id: uuid.UUID,
+    track_repo: TrackRepoDep,
+    metadata_service: MetadataServiceDep,
+    _: CurrentUser,
+) -> MetadataMatchesOut:
+    """AcoustID candidates for the metadata editor's match picker (§A7).
+
+    Runs the fingerprint lookup inline (single track, user-triggered) and
+    never mutates the track. Degrades to an empty list if fpcalc/AcoustID are
+    unavailable or no match is found.
+    """
+    track = await track_repo.get_by_id(track_id)
+    if track is None:
+        raise NotFoundError(f"Track {track_id} not found.")
+    matches = await metadata_service.find_matches(track_id)
+    return MetadataMatchesOut(
+        items=[
+            MetadataMatch(
+                acoustid=m.acoustid,
+                score=m.score,
+                recording_mbid=m.recording_mbid,
+                release_group_mbid=m.release_group_mbid,
+                title=m.title,
+                artist=m.artist,
+                album=m.album,
+                year=m.year,
+            )
+            for m in matches
+        ]
+    )
 
 
 @router.put("/{track_id}/metadata")
-async def set_metadata(track_id: uuid.UUID, _: CurrentUser) -> Any: ...
+async def set_metadata(
+    track_id: uuid.UUID,
+    body: MetadataApply,
+    track_repo: TrackRepoDep,
+    artist_repo: ArtistRepoDep,
+    album_repo: AlbumRepoDep,
+    _: CurrentUser,
+) -> TrackOut:
+    """Apply manual edits or an accepted AcoustID match (§A7). Sets
+    ``metadata_status = manual`` — never overwritten by auto-enrichment."""
+    track = await track_repo.get_by_id(track_id)
+    if track is None:
+        raise NotFoundError(f"Track {track_id} not found.")
+
+    artist_id: uuid.UUID | None = None
+    if body.artist_name:
+        artist = await artist_repo.get_or_create(body.artist_name)
+        artist_id = artist.id
+
+    album_id: uuid.UUID | None = None
+    if body.album_title:
+        album = await album_repo.get_or_create(
+            title=body.album_title,
+            artist_id=artist_id or track.artist_id,
+            year=body.year,
+            musicbrainz_id=None,
+        )
+        album_id = album.id
+
+    track = await track_repo.update(
+        track_id,
+        title=body.title,
+        genre=body.genre,
+        year=body.year,
+        artist_id=artist_id,
+        album_id=album_id,
+        track_number=body.track_number,
+    )
+
+    artist_ids = [track.artist_id]
+    album_ids = [track.album_id] if track.album_id else []
+    artists = {a.id: a for a in await artist_repo.get_many(artist_ids)}
+    albums = {a.id: a for a in await album_repo.get_many(album_ids)}
+
+    items = await _build_track_out([track], artists, albums)
+    return items[0]

app/api/v1/user_settings.py

@@ -1,23 +1,67 @@
-"""User settings endpoints, including scrobbling configuration."""
+"""User settings endpoints, including scrobbling configuration.
 
-from typing import Any
+Settings are per-caller and created lazily, so a first read returns defaults.
+The scrobbler session key is write-only — accepted on ``PUT`` but never returned.
+"""
 
 from fastapi import APIRouter
 
+from app.api.deps import CurrentUser, UserSettingsServiceDep
+from app.api.schemas.settings import (
+    ScrobblingOut,
+    ScrobblingUpdate,
+    SettingsOut,
+    SettingsUpdate,
+)
+from app.domain.entities.settings import UserSettings
+
 router = APIRouter(prefix="/settings", tags=["settings"])
 
 
+def _to_settings_out(settings: UserSettings) -> SettingsOut:
+    return SettingsOut(theme=settings.theme, stream_quality=settings.stream_quality)
+
+
+def _to_scrobbling_out(settings: UserSettings) -> ScrobblingOut:
+    return ScrobblingOut(
+        enabled=settings.scrobble_enabled,
+        provider=settings.scrobble_provider,
+        username=settings.scrobble_username,
+        configured=settings.scrobble_session_key_enc is not None,
+    )
+
+
 @router.get("")
-async def get_settings() -> Any: ...
+async def get_settings(user: CurrentUser, service: UserSettingsServiceDep) -> SettingsOut:
+    return _to_settings_out(await service.get(user.id))
 
 
 @router.patch("")
-async def update_settings() -> Any: ...
+async def update_settings(
+    body: SettingsUpdate, user: CurrentUser, service: UserSettingsServiceDep
+) -> SettingsOut:
+    settings = await service.update_general(
+        user.id, theme=body.theme, stream_quality=body.stream_quality
+    )
+    return _to_settings_out(settings)
 
 
 @router.get("/scrobbling")
-async def get_scrobbling_settings() -> Any: ...
+async def get_scrobbling_settings(
+    user: CurrentUser, service: UserSettingsServiceDep
+) -> ScrobblingOut:
+    return _to_scrobbling_out(await service.get(user.id))
 
 
 @router.put("/scrobbling")
-async def set_scrobbling_settings() -> Any: ...
+async def set_scrobbling_settings(
+    body: ScrobblingUpdate, user: CurrentUser, service: UserSettingsServiceDep
+) -> ScrobblingOut:
+    settings = await service.set_scrobbling(
+        user.id,
+        enabled=body.enabled,
+        provider=body.provider,
+        username=body.username,
+        session_key=body.session_key,
+    )
+    return _to_scrobbling_out(settings)

app/application/download_service.py

@@ -0,0 +1,183 @@
+"""DownloadService — request external downloads and import their results.
+
+Two roles (plan §6.1):
+
+* **Request side** (HTTP): validate + dedup a download request, create a
+  ``queued`` job, and enqueue the worker. Dedup is on ``(source, source_id)``
+  against both the library (already imported) and in-flight jobs (a double-click
+  must not queue twice) — idempotency per CLAUDE.md.
+* **Worker side**: ``store_result`` turns a backend's :class:`DownloadResult`
+  into a managed file + minimal ``pending`` track (sibling of
+  :class:`~app.application.import_service.LibraryImportService`); enrichment
+  (§6.2) fills the rest.
+
+The fingerprint-level dedup (a different id that turns out to be the same audio)
+happens later in enrichment, where the fingerprint is computed.
+"""
+
+import contextlib
+import uuid
+from collections.abc import Awaitable, Callable
+from dataclasses import dataclass
+
+import anyio
+
+from app.core.logging import get_logger
+from app.domain.entities.download import DownloadJob
+from app.domain.errors import NotFoundError, ValidationError
+from app.domain.ports import (
+    ArtistRepository,
+    DownloadJobRepository,
+    FileStorage,
+    TrackRepository,
+)
+from app.domain.sources import DownloadResult
+
+log = get_logger(__name__)
+
+_UNKNOWN_ARTIST = "Unknown Artist"
+
+# (job_id) -> None — enqueue the download worker, deferred so the job row is
+# committed before the worker reads it (same pattern as enrich).
+DownloadEnqueuer = Callable[[uuid.UUID], Awaitable[None]]
+EnrichEnqueuer = Callable[[uuid.UUID], Awaitable[None]]
+
+
+@dataclass(frozen=True)
+class DownloadRequest:
+    """Outcome of asking for a download.
+
+    Exactly one of the three states holds: the item is already in the library
+    (``track_id`` set, ``already_in_library``), a job already covers it / was
+    just created (``job`` set), so the UI can route to the download manager.
+    """
+
+    job: DownloadJob | None
+    track_id: uuid.UUID | None
+    already_in_library: bool
+
+
+class DownloadService:
+    def __init__(
+        self,
+        *,
+        jobs: DownloadJobRepository,
+        tracks: TrackRepository,
+        artists: ArtistRepository,
+        storage: FileStorage,
+        enqueue_download: DownloadEnqueuer | None = None,
+        enqueue_enrich: EnrichEnqueuer | None = None,
+    ) -> None:
+        self._jobs = jobs
+        self._tracks = tracks
+        self._artists = artists
+        self._storage = storage
+        self._enqueue_download = enqueue_download
+        self._enqueue_enrich = enqueue_enrich
+
+    # -- request side ---------------------------------------------------------
+    async def request(
+        self,
+        *,
+        source: str,
+        source_id: str,
+        query: str | None,
+        requested_by: uuid.UUID | None,
+    ) -> DownloadRequest:
+        source_id = source_id.strip()
+        if not source_id:
+            raise ValidationError("A source_id is required to download.")
+
+        existing = await self._tracks.get_by_source(source, source_id)
+        if existing is not None:
+            return DownloadRequest(job=None, track_id=existing.id, already_in_library=True)
+
+        active = await self._jobs.get_active_for_source(source, source_id)
+        if active is not None:
+            return DownloadRequest(job=active, track_id=None, already_in_library=False)
+
+        job = await self._jobs.add(
+            source=source,
+            source_id=source_id,
+            query=query,
+            requested_by=requested_by,
+        )
+        if self._enqueue_download is not None:
+            await self._enqueue_download(job.id)
+        return DownloadRequest(job=job, track_id=None, already_in_library=False)
+
+    async def list(
+        self,
+        *,
+        requested_by: uuid.UUID | None,
+        status: str | None,
+        limit: int,
+        offset: int,
+    ) -> tuple[list[DownloadJob], int]:
+        jobs = await self._jobs.list(
+            requested_by=requested_by, status=status, limit=limit, offset=offset
+        )
+        total = await self._jobs.count(requested_by=requested_by, status=status)
+        return jobs, total
+
+    async def get(self, job_id: uuid.UUID) -> DownloadJob:
+        job = await self._jobs.get_by_id(job_id)
+        if job is None:
+            raise NotFoundError(f"Download job {job_id} not found.")
+        return job
+
+    async def cancel(self, job_id: uuid.UUID) -> None:
+        """Remove the job record. True mid-flight cancellation of an in-progress
+        yt-dlp download is out of scope (MVP); the worker tolerates a vanished
+        job row (its status writes become no-ops)."""
+        job = await self._jobs.get_by_id(job_id)
+        if job is None:
+            raise NotFoundError(f"Download job {job_id} not found.")
+        await self._jobs.delete(job_id)
+
+    async def retry(self, job_id: uuid.UUID) -> DownloadJob:
+        job = await self.get(job_id)
+        await self._jobs.set_status(job_id, status="queued", error_message=None)
+        if self._enqueue_download is not None:
+            await self._enqueue_download(job_id)
+        refreshed = await self._jobs.get_by_id(job_id)
+        return refreshed if refreshed is not None else job
+
+    # -- worker side ----------------------------------------------------------
+    async def store_result(
+        self,
+        *,
+        source: str,
+        result: DownloadResult,
+        requested_by: uuid.UUID | None,
+    ) -> uuid.UUID:
+        """Store a freshly downloaded file and create a minimal ``pending`` track.
+
+        Returns the new track id (the caller enqueues enrichment after commit).
+        The temp file produced by the backend is always removed."""
+        track_id = uuid.uuid4()
+        key = f"tracks/{str(track_id)[:2]}/{track_id}.{result.file_format}"
+        try:
+            await self._storage.save_file(key, result.path)
+            try:
+                artist = await self._artists.get_or_create(_UNKNOWN_ARTIST)
+                await self._tracks.add(
+                    id=track_id,
+                    title=result.suggested_title,
+                    artist_id=artist.id,
+                    storage_uri=key,
+                    file_format=result.file_format,
+                    file_size=result.file_size,
+                    source=source,
+                    source_id=result.source_id,
+                    metadata_status="pending",
+                    added_by=requested_by,
+                )
+            except Exception:
+                with contextlib.suppress(Exception):
+                    await self._storage.delete(key)
+                raise
+        finally:
+            with contextlib.suppress(Exception):
+                await anyio.Path(result.path).unlink(missing_ok=True)
+        return track_id

app/application/import_service.py

@@ -0,0 +1,106 @@
+"""LibraryImportService — imports files discovered by an indexable source.
+
+Batch sibling of :class:`UploadService`: for each discovered file it dedups on
+``(source, source_id)``, copies the file into managed storage, creates a minimal
+track (artist ``Unknown Artist``, ``metadata_status=pending``), and leaves the
+rest to enrichment (plan §6.2). Per-file failures are isolated — one bad file
+must not abort the whole scan (graceful degradation).
+"""
+
+import contextlib
+import uuid
+from dataclasses import dataclass, field
+
+from app.core.logging import get_logger
+from app.domain.ports import ArtistRepository, FileStorage, IndexableSource, TrackRepository
+from app.domain.sources import SourceFile
+
+log = get_logger(__name__)
+
+_UNKNOWN_ARTIST = "Unknown Artist"
+
+
+@dataclass(frozen=True)
+class ImportSummary:
+    source: str
+    seen: int
+    imported: int
+    skipped: int
+    failed: int
+    # IDs of freshly imported tracks, for the caller to enqueue enrichment
+    # *after* its transaction commits (enqueuing mid-scan would race the worker).
+    imported_ids: list[uuid.UUID] = field(default_factory=list)
+
+
+class LibraryImportService:
+    def __init__(
+        self,
+        *,
+        tracks: TrackRepository,
+        artists: ArtistRepository,
+        storage: FileStorage,
+    ) -> None:
+        self._tracks = tracks
+        self._artists = artists
+        self._storage = storage
+
+    async def scan_and_import(
+        self, source: IndexableSource, *, added_by: uuid.UUID | None
+    ) -> ImportSummary:
+        seen = skipped = failed = 0
+        imported_ids: list[uuid.UUID] = []
+        for file in source.scan():
+            seen += 1
+            try:
+                existing = await self._tracks.get_by_source(source.name, file.source_id)
+                if existing is not None:
+                    skipped += 1
+                    continue
+                track_id = await self._import_one(source.name, file, added_by)
+                imported_ids.append(track_id)
+            except Exception:
+                failed += 1
+                log.warning("import_file_failed", source=source.name, source_id=file.source_id)
+        summary = ImportSummary(
+            source=source.name,
+            seen=seen,
+            imported=len(imported_ids),
+            skipped=skipped,
+            failed=failed,
+            imported_ids=imported_ids,
+        )
+        log.info(
+            "import_complete",
+            source=summary.source,
+            seen=summary.seen,
+            imported=summary.imported,
+            skipped=summary.skipped,
+            failed=summary.failed,
+        )
+        return summary
+
+    async def _import_one(
+        self, source_name: str, file: SourceFile, added_by: uuid.UUID | None
+    ) -> uuid.UUID:
+        track_id = uuid.uuid4()
+        key = f"tracks/{str(track_id)[:2]}/{track_id}.{file.file_format}"
+        await self._storage.save_file(key, file.path)
+        try:
+            artist = await self._artists.get_or_create(_UNKNOWN_ARTIST)
+            await self._tracks.add(
+                id=track_id,
+                title=file.suggested_title,
+                artist_id=artist.id,
+                storage_uri=key,
+                file_format=file.file_format,
+                file_size=file.file_size,
+                source=source_name,
+                source_id=file.source_id,
+                metadata_status="pending",
+                added_by=added_by,
+            )
+        except Exception:
+            with contextlib.suppress(Exception):
+                await self._storage.delete(key)
+            raise
+        return track_id

app/application/lyrics_service.py

@@ -0,0 +1,97 @@
+"""Lyrics service (plan §6.7).
+
+Get-or-fetch with caching: a track's lyrics are served from the DB when present;
+on a miss (or an expired ``not_found``) we ask the provider (LRCLIB) once, then
+cache the outcome. ``not_found`` is cached with a TTL so tracks that genuinely
+have no lyrics aren't looked up on every play, but can eventually be retried.
+
+Degrades gracefully: if the provider is unreachable the lookup just yields a
+``not_found`` — the endpoint still returns 200 with empty lyrics, never an error.
+"""
+
+import datetime as dt
+import uuid
+
+from app.domain.entities.lyrics import Lyrics
+from app.domain.errors import NotFoundError
+from app.domain.ports import (
+    AlbumRepository,
+    ArtistRepository,
+    LyricsProvider,
+    LyricsRepository,
+    TrackRepository,
+)
+
+# Re-lookup a cached "not_found" only after this long — long enough not to spam
+# the provider, short enough that lyrics added upstream eventually surface.
+_NOT_FOUND_TTL = dt.timedelta(days=7)
+
+_STATUS_FOUND = "found"
+_STATUS_NOT_FOUND = "not_found"
+
+
+class LyricsService:
+    def __init__(
+        self,
+        *,
+        lyrics: LyricsRepository,
+        tracks: TrackRepository,
+        artists: ArtistRepository,
+        albums: AlbumRepository,
+        provider: LyricsProvider,
+    ) -> None:
+        self._lyrics = lyrics
+        self._tracks = tracks
+        self._artists = artists
+        self._albums = albums
+        self._provider = provider
+
+    async def get_lyrics(self, track_id: uuid.UUID, *, force: bool = False) -> Lyrics:
+        """Return cached lyrics, fetching from the provider on a miss/expiry.
+        ``force`` (the refetch endpoint) bypasses the cache entirely."""
+        cached = await self._lyrics.get(track_id)
+        if not force and cached is not None and self._is_fresh(cached):
+            return cached
+        return await self._fetch_and_cache(track_id)
+
+    def _is_fresh(self, cached: Lyrics) -> bool:
+        if cached.status == _STATUS_FOUND:
+            return True
+        if cached.status == _STATUS_NOT_FOUND:
+            return dt.datetime.now(dt.UTC) - cached.fetched_at < _NOT_FOUND_TTL
+        # "pending" (never fetched) → not fresh, go fetch.
+        return False
+
+    async def _fetch_and_cache(self, track_id: uuid.UUID) -> Lyrics:
+        track = await self._tracks.get_by_id(track_id)
+        if track is None:
+            raise NotFoundError(f"Track {track_id} not found.")
+
+        artist = await self._artists.get_by_id(track.artist_id)
+        album = (
+            await self._albums.get_by_id(track.album_id)
+            if track.album_id is not None
+            else None
+        )
+        result = await self._provider.fetch(
+            artist=artist.name if artist else "",
+            title=track.title,
+            album=album.title if album else None,
+            duration_seconds=track.duration_seconds,
+        )
+
+        if result is None:
+            return await self._lyrics.upsert(
+                track_id=track_id,
+                synced=None,
+                plain=None,
+                source=None,
+                status=_STATUS_NOT_FOUND,
+            )
+        return await self._lyrics.upsert(
+            track_id=track_id,
+            synced=result.synced,
+            plain=result.plain,
+            source=result.source,
+            status=_STATUS_FOUND,
+        )

app/application/metadata_service.py

@@ -0,0 +1,306 @@
+"""MetadataEnrichmentService — the §6.2 pipeline orchestrator.
+
+Order (tag-first): embedded tags → Chromaprint fingerprint → AcoustID lookup.
+Tags fix the common well-tagged case offline; AcoustID identifies the rest and
+supplies a MusicBrainz id. The result updates the track and sets
+``metadata_status`` to ``enriched`` (identity found) or ``failed`` (nothing).
+
+Invariants (plan §6.2, CLAUDE.md):
+- **Never touch ``manual``** — a user-edited track is returned untouched.
+- **Graceful degradation** — every external step is wrapped; one failure (no
+  fpcalc, no API key, service down) degrades the result, never crashes.
+- **Idempotent** — re-running only fills gaps; ``apply_enrichment`` never erases.
+"""
+
+import tempfile
+import uuid
+from dataclasses import dataclass
+from pathlib import Path
+
+from app.core.logging import get_logger
+from app.domain.entities.album import Album
+from app.domain.entities.cover import CoverArt
+from app.domain.entities.metadata import AudioTags, RecordingMatch
+from app.domain.ports import (
+    AcoustIdClient,
+    AlbumRepository,
+    ArtistRepository,
+    AudioFingerprinter,
+    AudioTagReader,
+    CoverArtExtractor,
+    CoverArtProvider,
+    FileStorage,
+    TrackRepository,
+)
+
+log = get_logger(__name__)
+
+_UNKNOWN_ARTIST = "Unknown Artist"
+
+
+@dataclass(frozen=True)
+class EnrichmentResult:
+    track_id: uuid.UUID
+    status: str  # "enriched" | "failed" | "skipped"
+    matched_mbid: str | None = None
+
+
+class MetadataEnrichmentService:
+    def __init__(
+        self,
+        *,
+        tracks: TrackRepository,
+        artists: ArtistRepository,
+        albums: AlbumRepository,
+        storage: FileStorage,
+        tag_reader: AudioTagReader,
+        fingerprinter: AudioFingerprinter,
+        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
+        self._albums = albums
+        self._storage = storage
+        self._tag_reader = tag_reader
+        self._fingerprinter = fingerprinter
+        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)
+        if track is None:
+            log.info("enrich_track_missing", track_id=str(track_id))
+            return EnrichmentResult(track_id=track_id, status="skipped")
+        if track.metadata_status == "manual":
+            log.info("enrich_skip_manual", track_id=str(track_id))
+            return EnrichmentResult(track_id=track_id, status="skipped")
+        storage_uri = track.storage_uri
+        if storage_uri is None:
+            log.info("enrich_skip_remote", track_id=str(track_id))
+            return EnrichmentResult(track_id=track_id, status="skipped")
+
+        tags = await self._read_local(storage_uri)
+        match = await self._identify(storage_uri)
+
+        # 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
+        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(
+            tags.duration_seconds if tags else None,
+            track.duration_seconds,
+        )
+        bitrate = tags.bitrate if tags else None
+        mbid = match.recording_mbid if match else None
+        acoustid_id = match.acoustid if match else None
+
+        artist_id = await self._resolve_artist(artist_name, fallback=track.artist_id)
+        album = await self._resolve_album(album_title, artist_id=artist_id, year=year, mbid=mbid)
+        album_id = album.id if album is not None else None
+
+        if album is not None:
+            await self._resolve_cover(
+                album,
+                storage_uri=storage_uri,
+                release_group_mbid=match.release_group_mbid if match else None,
+            )
+
+        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,
+            title=title,
+            artist_id=artist_id,
+            album_id=album_id,
+            genre=genre,
+            year=year,
+            track_number=track_number,
+            duration_seconds=duration,
+            bitrate=bitrate,
+            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 find_matches(self, track_id: uuid.UUID) -> list[RecordingMatch]:
+        """AcoustID candidates for the metadata editor's match picker (§A7).
+
+        Read-only — unlike :meth:`enrich`, never touches the track. Runs
+        inline (single track, user-triggered) rather than via the worker.
+        Degrades to ``[]`` whenever fingerprinting/AcoustID is unavailable or
+        the file can't be read, same as the enrichment pipeline.
+        """
+        track = await self._tracks.get_by_id(track_id)
+        if track is None:
+            return []
+        if not self._acoustid.is_available() or not self._fingerprinter.is_available():
+            return []
+        if track.storage_uri is None:
+            return []
+        try:
+            async with self._storage.as_local_path(track.storage_uri) as path:
+                fingerprint = await self._fingerprinter.calculate(path)
+            if fingerprint is None:
+                return []
+            return await self._acoustid.lookup_all(fingerprint)
+        except Exception:
+            log.warning("find_matches_failed", track_id=str(track_id))
+            return []
+
+    async def _read_local(self, storage_uri: str) -> AudioTags | None:
+        try:
+            async with self._storage.as_local_path(storage_uri) as path:
+                return await self._tag_reader.read(path)
+        except Exception:
+            log.warning("enrich_tag_step_failed", storage_uri=storage_uri)
+            return None
+
+    async def _identify(self, storage_uri: str) -> RecordingMatch | None:
+        if not self._acoustid.is_available() or not self._fingerprinter.is_available():
+            return None
+        try:
+            async with self._storage.as_local_path(storage_uri) as path:
+                fingerprint = await self._fingerprinter.calculate(path)
+            if fingerprint is None:
+                return None
+            return await self._acoustid.lookup(fingerprint)
+        except Exception:
+            log.warning("enrich_identify_step_failed", storage_uri=storage_uri)
+            return None
+
+    async def _resolve_artist(self, name: str | None, *, fallback: uuid.UUID) -> uuid.UUID:
+        if not name or name == _UNKNOWN_ARTIST:
+            return fallback
+        artist = await self._artists.get_or_create(name)
+        return artist.id
+
+    async def _resolve_album(
+        self,
+        title: str | None,
+        *,
+        artist_id: uuid.UUID,
+        year: int | None,
+        mbid: str | None,
+    ) -> Album | None:
+        if not title:
+            return None
+        return await self._albums.get_or_create(
+            title=title,
+            artist_id=artist_id,
+            year=year,
+            musicbrainz_id=mbid,
+        )
+
+    async def _resolve_cover(
+        self,
+        album: Album,
+        *,
+        storage_uri: str,
+        release_group_mbid: str | None,
+    ) -> None:
+        """Fill in an album cover when it has none. Source order mirrors the
+        tag-first pipeline: embedded artwork (offline) → Cover Art Archive
+        (network, by release-group). Best-effort — any failure is swallowed so a
+        missing cover never affects enrichment status."""
+        if album.cover_path:
+            return  # already has one — never overwrite (idempotent)
+
+        cover = await self._extract_cover(storage_uri)
+        if cover is None:
+            cover = await self._fetch_cover(release_group_mbid)
+        if cover is None:
+            return
+
+        try:
+            key = await self._save_cover(album.id, cover)
+            await self._albums.set_cover_path(album.id, key)
+            log.info("cover_resolved", album_id=str(album.id), content_type=cover.content_type)
+        except Exception:
+            log.warning("cover_save_failed", album_id=str(album.id))
+
+    async def _extract_cover(self, storage_uri: str) -> CoverArt | None:
+        if self._cover_extractor is None:
+            return None
+        try:
+            async with self._storage.as_local_path(storage_uri) as path:
+                return await self._cover_extractor.extract(path)
+        except Exception:
+            log.warning("cover_extract_step_failed", storage_uri=storage_uri)
+            return None
+
+    async def _fetch_cover(self, release_group_mbid: str | None) -> CoverArt | None:
+        if self._cover_provider is None or not release_group_mbid:
+            return None
+        if not self._cover_provider.is_available():
+            return None
+        try:
+            return await self._cover_provider.fetch_release_group(release_group_mbid)
+        except Exception:
+            log.warning("cover_fetch_step_failed", release_group=release_group_mbid)
+            return None
+
+    async def _save_cover(self, album_id: uuid.UUID, cover: CoverArt) -> str:
+        key = f"covers/{album_id}.{cover.extension}"
+        with tempfile.NamedTemporaryFile(suffix=f".{cover.extension}") as tmp:
+            tmp.write(cover.data)
+            tmp.flush()
+            await self._storage.save_file(key, Path(tmp.name))
+        return key
+
+
+def _opt_str(*values: str | None) -> str | None:
+    for value in values:
+        if value:
+            return value
+    return None
+
+
+def _first_int(*values: int | None) -> int | None:
+    for value in values:
+        if value is not None:
+            return value
+    return None

app/application/recommendation_service.py

@@ -0,0 +1,189 @@
+"""Recommendation / radio service (plan §6.5).
+
+Tries the external ML recommender first; when it's unavailable or declines
+(returns ``None``), falls back to metadata heuristics over the catalogue — so
+similar/radio always work, worse, without ML (graceful-degradation invariant).
+
+``reason`` values are short codes (``ml`` / ``similar`` / ``from_likes`` /
+``discover``) the client localizes for the "why is this playing?" affordance.
+"""
+
+import random
+import uuid
+from dataclasses import dataclass
+
+from app.domain.entities.track import Artist, Track
+from app.domain.errors import NotFoundError
+from app.domain.ports import (
+    ArtistRepository,
+    LikeRepository,
+    Recommender,
+    TrackRepository,
+)
+
+REASON_ML = "ml"
+REASON_SIMILAR = "similar"
+REASON_FROM_LIKES = "from_likes"
+REASON_DISCOVER = "discover"
+
+_LIKED_SEED_POOL = 50
+
+
+@dataclass(frozen=True, slots=True)
+class RadioPick:
+    track: Track
+    reason: str
+
+
+class RecommendationService:
+    def __init__(
+        self,
+        *,
+        recommender: Recommender,
+        tracks: TrackRepository,
+        artists: ArtistRepository,
+        likes: LikeRepository,
+    ) -> None:
+        self._recommender = recommender
+        self._tracks = tracks
+        self._artists = artists
+        self._likes = likes
+
+    # -- similar ---------------------------------------------------------------
+
+    async def similar_tracks(
+        self, track_id: uuid.UUID, *, limit: int
+    ) -> tuple[str, list[Track]]:
+        seed = await self._tracks.get_by_id(track_id)
+        if seed is None:
+            raise NotFoundError(f"Track {track_id} not found.")
+
+        if self._recommender.is_available():
+            ids = await self._recommender.similar_track_ids(
+                track_id, limit=limit, exclude_ids=[track_id]
+            )
+            if ids is not None:
+                return REASON_ML, await self._hydrate_tracks(ids)
+
+        found = await self._tracks.list_similar(
+            genre=seed.genre,
+            artist_id=seed.artist_id,
+            exclude_ids=[track_id],
+            limit=limit,
+        )
+        return REASON_SIMILAR, found
+
+    async def similar_artists(
+        self, artist_id: uuid.UUID, *, limit: int
+    ) -> tuple[str, list[Artist]]:
+        if await self._artists.get_by_id(artist_id) is None:
+            raise NotFoundError(f"Artist {artist_id} not found.")
+
+        if self._recommender.is_available():
+            ids = await self._recommender.similar_artist_ids(artist_id, limit=limit)
+            if ids is not None:
+                by_id = {a.id: a for a in await self._artists.get_many(ids)}
+                return REASON_ML, [by_id[i] for i in ids if i in by_id]
+
+        found = await self._artists.list_similar(artist_id=artist_id, limit=limit)
+        return REASON_SIMILAR, found
+
+    # -- radio -----------------------------------------------------------------
+
+    async def radio(
+        self,
+        *,
+        user_id: uuid.UUID,
+        seed_track_id: uuid.UUID | None,
+        from_likes: bool,
+        exploration: float,
+        limit: int,
+        exclude_ids: list[uuid.UUID],
+    ) -> tuple[str, list[RadioPick]]:
+        exploration = min(1.0, max(0.0, exploration))
+
+        if self._recommender.is_available():
+            ids = await self._recommender.radio_track_ids(
+                seed_track_id=seed_track_id,
+                exploration=exploration,
+                limit=limit,
+                exclude_ids=exclude_ids,
+            )
+            if ids is not None:
+                picks = [
+                    RadioPick(track=t, reason=REASON_ML)
+                    for t in await self._hydrate_tracks(ids)
+                ]
+                return REASON_ML, picks
+
+        return "metadata", await self._radio_fallback(
+            user_id=user_id,
+            seed_track_id=seed_track_id,
+            from_likes=from_likes,
+            exploration=exploration,
+            limit=limit,
+            exclude_ids=exclude_ids,
+        )
+
+    async def _radio_fallback(
+        self,
+        *,
+        user_id: uuid.UUID,
+        seed_track_id: uuid.UUID | None,
+        from_likes: bool,
+        exploration: float,
+        limit: int,
+        exclude_ids: list[uuid.UUID],
+    ) -> list[RadioPick]:
+        exclude = list(dict.fromkeys(exclude_ids))  # de-dupe, keep order
+        explore_n = round(limit * exploration)
+        similar_n = limit - explore_n
+        picks: list[RadioPick] = []
+
+        seed, seed_reason = await self._resolve_seed(
+            user_id, seed_track_id, from_likes
+        )
+        if seed is not None and similar_n > 0:
+            for track in await self._tracks.list_similar(
+                genre=seed.genre,
+                artist_id=seed.artist_id,
+                exclude_ids=exclude,
+                limit=similar_n,
+            ):
+                picks.append(RadioPick(track=track, reason=seed_reason))
+                exclude.append(track.id)
+
+        # Fill the remainder (exploration + any similarity shortfall) with random
+        # playable tracks — this is also the total fallback when there's no seed.
+        remaining = limit - len(picks)
+        if remaining > 0:
+            for track in await self._tracks.sample_playable(
+                exclude_ids=exclude, limit=remaining
+            ):
+                picks.append(RadioPick(track=track, reason=REASON_DISCOVER))
+                exclude.append(track.id)
+
+        random.shuffle(picks)
+        return picks
+
+    async def _resolve_seed(
+        self,
+        user_id: uuid.UUID,
+        seed_track_id: uuid.UUID | None,
+        from_likes: bool,
+    ) -> tuple[Track | None, str]:
+        if seed_track_id is not None:
+            return await self._tracks.get_by_id(seed_track_id), REASON_SIMILAR
+        if from_likes:
+            liked = await self._likes.list_liked_tracks(
+                user_id=user_id, limit=_LIKED_SEED_POOL, offset=0
+            )
+            if liked:
+                return random.choice(liked), REASON_FROM_LIKES
+        return None, REASON_DISCOVER
+
+    async def _hydrate_tracks(self, ids: list[uuid.UUID]) -> list[Track]:
+        """Resolve ids → tracks preserving order, skipping any that vanished.
+        One batched query rather than N per-id round-trips."""
+        by_id = {t.id: t for t in await self._tracks.get_many(ids)}
+        return [by_id[i] for i in ids if i in by_id]

app/application/remote_library_service.py

@@ -0,0 +1,122 @@
+"""RemoteLibraryService — save-to-library + materialize for remote browse hits
+(plan: Model C, on-demand YTM library).
+
+Two operations:
+
+* ``save_remote`` persists a placeholder ``Track`` (``availability="remote"``,
+  ``storage_uri=None``) for a remote browse hit. Idempotent on
+  ``(source, source_id)`` — CLAUDE.md dedup.
+* ``request_materialize`` lazily fills a placeholder's audio in place: it
+  creates (or reuses) a ``DownloadJob`` pointing at the existing track and
+  enqueues the materialize worker, which calls ``TrackRepository.materialize``
+  on completion. ``track.id`` never changes (CLAUDE.md), so likes/playlists/
+  queue entries referencing the placeholder keep working once it's filled in.
+"""
+
+import uuid
+from collections.abc import Awaitable, Callable
+from dataclasses import dataclass
+
+from app.domain.entities.download import DownloadJob
+from app.domain.entities.track import Track
+from app.domain.errors import NotFoundError, ValidationError
+from app.domain.ports import ArtistRepository, DownloadJobRepository, TrackRepository
+
+_UNKNOWN_ARTIST = "Unknown Artist"
+
+# (job_id) -> None — enqueue the materialize worker, same deferred pattern as
+# download/enrich enqueuers.
+MaterializeEnqueuer = Callable[[uuid.UUID], Awaitable[None]]
+
+
+@dataclass(frozen=True)
+class MaterializeOutcome:
+    """Result of requesting materialization.
+
+    ``job`` is ``None`` when the track is already ``local`` — nothing to do,
+    the caller can stream immediately. Otherwise it's the (new or already
+    in-flight) job filling the placeholder."""
+
+    track: Track
+    job: DownloadJob | None
+
+
+class RemoteLibraryService:
+    def __init__(
+        self,
+        *,
+        tracks: TrackRepository,
+        artists: ArtistRepository,
+        jobs: DownloadJobRepository,
+        enqueue_materialize: MaterializeEnqueuer | None = None,
+    ) -> None:
+        self._tracks = tracks
+        self._artists = artists
+        self._jobs = jobs
+        self._enqueue_materialize = enqueue_materialize
+
+    async def save_remote(
+        self,
+        *,
+        source: str,
+        source_id: str,
+        title: str,
+        artist: str | None,
+        added_by: uuid.UUID | None,
+    ) -> Track:
+        """Persist a placeholder for a remote browse hit. Idempotent: a hit
+        already saved (by ``(source, source_id)``) is returned as-is."""
+        source_id = source_id.strip()
+        if not source_id:
+            raise ValidationError("A source_id is required to save.")
+
+        existing = await self._tracks.get_by_source(source, source_id)
+        if existing is not None:
+            return existing
+
+        artist_entity = await self._artists.get_or_create(artist or _UNKNOWN_ARTIST)
+        return await self._tracks.add(
+            id=uuid.uuid4(),
+            title=title,
+            artist_id=artist_entity.id,
+            storage_uri=None,
+            file_format=None,
+            file_size=None,
+            source=source,
+            source_id=source_id,
+            metadata_status="pending",
+            added_by=added_by,
+            availability="remote",
+        )
+
+    async def request_materialize(
+        self, track_id: uuid.UUID, *, requested_by: uuid.UUID | None
+    ) -> MaterializeOutcome:
+        """Kick off (or report on) materializing a placeholder track.
+
+        Already-local tracks are a no-op (``job=None``). A track with no
+        remote ``source_id`` (e.g. a deleted upload row reused for something
+        else) can't be materialized."""
+        track = await self._tracks.get_by_id(track_id)
+        if track is None:
+            raise NotFoundError(f"Track {track_id} not found.")
+        if track.availability == "local":
+            return MaterializeOutcome(track=track, job=None)
+        if track.source_id is None:
+            raise ValidationError("Track has no remote source to materialize from.")
+
+        active = await self._jobs.get_active_for_source(track.source, track.source_id)
+        if active is not None:
+            return MaterializeOutcome(track=track, job=active)
+
+        job = await self._jobs.add(
+            source=track.source,
+            source_id=track.source_id,
+            query=None,
+            requested_by=requested_by,
+        )
+        await self._jobs.set_status(job.id, status="queued", track_id=track.id)
+        if self._enqueue_materialize is not None:
+            await self._enqueue_materialize(job.id)
+        refreshed = await self._jobs.get_by_id(job.id)
+        return MaterializeOutcome(track=track, job=refreshed if refreshed is not None else job)

app/application/streaming_service.py

@@ -72,16 +72,19 @@ class StreamingService:
         track = await self._tracks.get_by_id(track_id)
         if track is None:
             raise NotFoundError("Track not found.")
+        storage_uri = track.storage_uri
+        if storage_uri is None:
+            raise NotFoundError("Track is not yet downloaded.")
 
-        stat = await self._storage.stat(track.storage_uri)
+        stat = await self._storage.stat(storage_uri)
         total_size = stat.size
         content_type = stat.content_type or _FORMAT_CONTENT_TYPE.get(
-            track.file_format.lower(), "application/octet-stream"
+            (track.file_format or "").lower(), "application/octet-stream"
         )
 
         start, end, is_partial = _parse_range(range_header, total_size)
 
-        stream, _ = await self._storage.open_range(track.storage_uri, start, end)
+        stream, _ = await self._storage.open_range(storage_uri, start, end)
 
         actual_end = end if end is not None else total_size - 1
         content_length = actual_end - start + 1

app/application/sync_service.py

@@ -0,0 +1,140 @@
+"""Offline-first sync use cases: delta pull + idempotent push.
+
+The cursor is a server-clock timestamp obtained from the DB (``now``), so it is
+immune to app/DB clock skew. A pull returns everything a user changed in the
+half-open window ``(since, cursor]``; a push appends the event-log entries a
+client accumulated offline. Events carry a client-generated id, so a replay is a
+no-op (append is ``ON CONFLICT DO NOTHING``). Events for tracks this server does
+not have are skipped rather than rejected (graceful degradation).
+
+Known limitation (v1): the delta is unpaginated and, being wall-clock based, a
+write committing right on the cursor boundary under concurrency can slip a cycle
+— a periodic full resync (``since=None``) heals it.
+"""
+
+import datetime as dt
+import uuid
+from collections.abc import Awaitable, Callable
+from dataclasses import dataclass
+
+from app.domain.entities.history import PlayHistoryEntry
+from app.domain.entities.like import Like
+from app.domain.entities.playlist import Playlist
+from app.domain.entities.track import Track
+from app.domain.ports import (
+    HistoryRepository,
+    LikeRepository,
+    PlaylistRepository,
+    TrackRepository,
+)
+
+
+@dataclass(frozen=True, slots=True)
+class LikeEvent:
+    id: uuid.UUID
+    track_id: uuid.UUID
+    value: str
+    created_at: dt.datetime
+
+
+@dataclass(frozen=True, slots=True)
+class PlayEvent:
+    id: uuid.UUID
+    track_id: uuid.UUID
+    played_at: dt.datetime
+    play_duration_seconds: int | None
+    completed: bool
+
+
+@dataclass(frozen=True, slots=True)
+class PlaylistDelta:
+    playlist: Playlist
+    track_ids: list[uuid.UUID]
+
+
+@dataclass(frozen=True, slots=True)
+class SyncChanges:
+    cursor: dt.datetime
+    likes: list[Like]
+    plays: list[PlayHistoryEntry]
+    playlists: list[PlaylistDelta]
+    tracks: list[Track]
+
+
+@dataclass(frozen=True, slots=True)
+class SyncPushResult:
+    cursor: dt.datetime
+    accepted_likes: int
+    accepted_plays: int
+
+
+class SyncService:
+    def __init__(
+        self,
+        *,
+        likes: LikeRepository,
+        history: HistoryRepository,
+        playlists: PlaylistRepository,
+        tracks: TrackRepository,
+        now: Callable[[], Awaitable[dt.datetime]],
+    ) -> None:
+        self._likes = likes
+        self._history = history
+        self._playlists = playlists
+        self._tracks = tracks
+        self._now = now
+
+    async def get_changes(self, user_id: uuid.UUID, *, since: dt.datetime | None) -> SyncChanges:
+        until = await self._now()
+        likes = await self._likes.list_since(user_id, since=since, until=until)
+        plays = await self._history.list_since(user_id, since=since, until=until)
+        changed = await self._playlists.list_changed_since(
+            owner_id=user_id, since=since, until=until
+        )
+        playlists = [
+            PlaylistDelta(playlist=p, track_ids=await self._playlists.track_ids(p.id))
+            for p in changed
+        ]
+        tracks = await self._tracks.list_changed_since(since=since, until=until)
+        return SyncChanges(
+            cursor=until, likes=likes, plays=plays, playlists=playlists, tracks=tracks
+        )
+
+    async def push(
+        self,
+        user_id: uuid.UUID,
+        *,
+        likes: list[LikeEvent],
+        plays: list[PlayEvent],
+    ) -> SyncPushResult:
+        accepted_likes = 0
+        for event in likes:
+            if await self._tracks.get_by_id(event.track_id) is None:
+                continue  # skip events for tracks this server doesn't have
+            if await self._likes.add_event(
+                id=event.id,
+                user_id=user_id,
+                track_id=event.track_id,
+                value=event.value,
+                created_at=event.created_at,
+            ):
+                accepted_likes += 1
+
+        accepted_plays = 0
+        for play in plays:
+            if await self._tracks.get_by_id(play.track_id) is None:
+                continue
+            if await self._history.add_event(
+                id=play.id,
+                user_id=user_id,
+                track_id=play.track_id,
+                played_at=play.played_at,
+                play_duration_seconds=play.play_duration_seconds,
+                completed=play.completed,
+            ):
+                accepted_plays += 1
+
+        cursor = await self._now()
+        return SyncPushResult(
+            cursor=cursor, accepted_likes=accepted_likes, accepted_plays=accepted_plays
+        )

app/application/transcode_service.py

@@ -0,0 +1,110 @@
+"""Transcode service + cache-path helpers (Group B / plan §6.6).
+
+Cache layout under ``transcode_cache_path``::
+
+    {track_id}/opus_{kbps}.opus     # direct quality renditions
+    {track_id}/hls/playlist.m3u8    # HLS rendition (AAC-in-TS)
+    {track_id}/hls/seg_000.ts …
+
+The request side (streaming router) only *reads* the cache — misses fall back to
+the original file and enqueue generation. The worker (``transcode_task``) writes
+it. Path helpers are module-level so both sides agree on locations without one
+importing the other.
+"""
+
+import re
+import shutil
+import uuid
+from pathlib import Path
+
+import anyio
+
+from app.domain.errors import NotFoundError
+from app.domain.ports import TrackRepository
+
+# Stream-quality name (matches the user-settings ``StreamQuality``) → Opus
+# bitrate. ``original`` is absent: it means "serve the master, no transcode".
+QUALITY_BITRATE: dict[str, int] = {"high": 128, "medium": 96, "low": 64}
+
+# Single HLS rendition bitrate (AAC). One rendition keeps the MVP simple; a
+# multi-bitrate ladder can come later.
+HLS_BITRATE = 128
+
+# Only these segment names may be served, guarding the segment route against
+# path traversal.
+_SEGMENT_RE = re.compile(r"^seg_\d{3,}\.ts$")
+
+
+def bitrate_for_quality(quality: str) -> int | None:
+    """Opus bitrate for a quality name, or ``None`` for ``original``/unknown."""
+    return QUALITY_BITRATE.get(quality)
+
+
+def track_cache_dir(root: Path, track_id: uuid.UUID) -> Path:
+    return root / str(track_id)
+
+
+def opus_path(root: Path, track_id: uuid.UUID, bitrate_kbps: int) -> Path:
+    return track_cache_dir(root, track_id) / f"opus_{bitrate_kbps}.opus"
+
+
+def hls_dir(root: Path, track_id: uuid.UUID) -> Path:
+    return track_cache_dir(root, track_id) / "hls"
+
+
+def hls_playlist_path(root: Path, track_id: uuid.UUID) -> Path:
+    return hls_dir(root, track_id) / "playlist.m3u8"
+
+
+def hls_segment_path(root: Path, track_id: uuid.UUID, name: str) -> Path | None:
+    """Resolve a segment file, or ``None`` if the name is not a valid segment."""
+    if not _SEGMENT_RE.fullmatch(name):
+        return None
+    return hls_dir(root, track_id) / name
+
+
+def remove_track_cache(root: Path, track_id: uuid.UUID) -> None:
+    """Delete every cached rendition for a track (Opus + HLS). Best-effort — used
+    when a track is deleted so its transcode cache doesn't dangle forever."""
+    shutil.rmtree(track_cache_dir(root, track_id), ignore_errors=True)
+
+
+class TranscodeService:
+    """Request-side cache lookups for transcoded renditions."""
+
+    def __init__(self, *, tracks: TrackRepository, cache_root: Path) -> None:
+        self._tracks = tracks
+        self._root = cache_root
+
+    async def _require_streamable(self, track_id: uuid.UUID) -> None:
+        track = await self._tracks.get_by_id(track_id)
+        if track is None:
+            raise NotFoundError("Track not found.")
+        if track.storage_uri is None:
+            raise NotFoundError("Track is not yet downloaded.")
+
+    async def resolve_quality_file(
+        self, track_id: uuid.UUID, quality: str
+    ) -> Path | None:
+        """Cached Opus file for ``quality`` if present, else ``None`` (caller
+        falls back to the master and enqueues generation). ``original`` → None."""
+        bitrate = bitrate_for_quality(quality)
+        if bitrate is None:
+            return None
+        path = opus_path(self._root, track_id, bitrate)
+        exists = await anyio.to_thread.run_sync(path.exists)
+        return path if exists else None
+
+    async def hls_playlist(self, track_id: uuid.UUID) -> Path | None:
+        """Cached HLS playlist if generated, else ``None`` (validates the track
+        exists so an unknown id 404s rather than silently missing)."""
+        await self._require_streamable(track_id)
+        path = hls_playlist_path(self._root, track_id)
+        exists = await anyio.to_thread.run_sync(path.exists)
+        return path if exists else None
+
+    def hls_segment(self, track_id: uuid.UUID, name: str) -> Path | None:
+        path = hls_segment_path(self._root, track_id, name)
+        if path is None or not path.exists():
+            return None
+        return path

app/application/upload_service.py

@@ -5,6 +5,7 @@ import hashlib
 import os
 import tempfile
 import uuid
+from collections.abc import Awaitable, Callable
 from dataclasses import dataclass
 from pathlib import Path
 from typing import Protocol
@@ -14,6 +15,8 @@ import anyio
 from app.domain.entities.user import User
 from app.domain.ports import ArtistRepository, FileStorage, TrackRepository
 
+EnrichEnqueuer = Callable[[uuid.UUID], Awaitable[None]]
+
 
 class UploadFileProtocol(Protocol):
     filename: str | None
@@ -49,11 +52,13 @@ class UploadService:
         artists: ArtistRepository,
         storage: FileStorage,
         tmp_dir: Path | None = None,
+        enqueue_enrich: EnrichEnqueuer | None = None,
     ) -> None:
         self._tracks = tracks
         self._artists = artists
         self._storage = storage
         self._tmp_dir = tmp_dir
+        self._enqueue_enrich = enqueue_enrich
 
     async def handle_upload(
         self,
@@ -105,7 +110,8 @@ class UploadService:
                     await self._storage.delete(key)
                 raise
 
-            # TODO(1D): enqueue metadata enrichment task
+            if self._enqueue_enrich is not None:
+                await self._enqueue_enrich(track.id)
 
             return UploadResult(
                 track_id=track.id,

app/application/user_settings_service.py

@@ -0,0 +1,62 @@
+"""User-settings use cases: general preferences + scrobbling configuration.
+
+Settings rows are created lazily — a user who never saved anything reads clean
+defaults. Partial updates are merged against current values here, so the
+repository always persists the complete desired state. The scrobbler session
+key is encrypted before it touches the DB (never stored or returned in plain).
+"""
+
+import uuid
+from dataclasses import replace
+
+from app.domain.entities.settings import UserSettings
+from app.domain.ports import SubsonicCipher, UserSettingsRepository
+
+
+class UserSettingsService:
+    def __init__(self, *, settings: UserSettingsRepository, cipher: SubsonicCipher) -> None:
+        self._settings = settings
+        # Same Fernet cipher used for the Subsonic app-password — reused here to
+        # encrypt the scrobbler session key at rest (symmetric, recoverable).
+        self._cipher = cipher
+
+    async def get(self, user_id: uuid.UUID) -> UserSettings:
+        return await self._settings.get(user_id) or UserSettings.defaults(user_id)
+
+    async def update_general(
+        self, user_id: uuid.UUID, *, theme: str | None, stream_quality: str | None
+    ) -> UserSettings:
+        current = await self.get(user_id)
+        merged = replace(
+            current,
+            theme=theme if theme is not None else current.theme,
+            stream_quality=stream_quality if stream_quality is not None else current.stream_quality,
+        )
+        return await self._settings.upsert(merged)
+
+    async def set_scrobbling(
+        self,
+        user_id: uuid.UUID,
+        *,
+        enabled: bool,
+        provider: str | None,
+        username: str | None,
+        session_key: str | None,
+    ) -> UserSettings:
+        """Replace the scrobbling config. ``session_key`` is write-only: a new
+        value is encrypted and stored; omitting it keeps the existing key (the
+        client can't read it back to re-send it)."""
+        current = await self.get(user_id)
+        session_key_enc: str | None
+        if session_key is not None:
+            session_key_enc = self._cipher.encrypt(session_key)
+        else:
+            session_key_enc = current.scrobble_session_key_enc
+        merged = replace(
+            current,
+            scrobble_enabled=enabled,
+            scrobble_provider=provider,
+            scrobble_username=username,
+            scrobble_session_key_enc=session_key_enc,
+        )
+        return await self._settings.upsert(merged)

app/core/config.py

@@ -5,12 +5,25 @@ development). Access the cached singleton via :func:`get_settings`.
 """
 
 from functools import lru_cache
+from importlib.metadata import PackageNotFoundError, version
 from pathlib import Path
 from typing import Literal
 
 from pydantic import Field, SecretStr, field_validator
 from pydantic_settings import BaseSettings, SettingsConfigDict
 
+# App identity for outbound API calls (e.g. the MusicBrainz/AcoustID
+# User-Agent). Name is fixed; version comes from the installed package.
+APP_NAME = "MCMA"
+_PROJECT_URL = "https://git.ollyhearn.ru/olly/mcma-backend"
+
+
+def app_version() -> str:
+    try:
+        return version("mcma-backend")
+    except PackageNotFoundError:
+        return "0.0.0"
+
 
 class Settings(BaseSettings):
     model_config = SettingsConfigDict(
@@ -44,6 +57,20 @@ class Settings(BaseSettings):
     jwt_algorithm: str = "HS256"
     access_token_ttl_seconds: int = 60 * 15  # 15 min
     refresh_token_ttl_seconds: int = 60 * 60 * 24 * 30  # 30 days (offline-first)
+    # Public self-service sign-up. When disabled, accounts are created
+    # admin-only (POST /admin/users). Registered users are never superusers.
+    allow_registration: bool = True
+
+    # -- CORS -------------------------------------------------------------
+    # Origins allowed to call the API from a browser. The web UI is multi-
+    # instance — it connects to whatever origin the operator types on the
+    # connect screen — so a page served from origin A may call this backend at
+    # origin B (e.g. the direct :8000 port, a LAN IP, or 127.0.0.1 vs localhost).
+    # Auth rides in the ``Authorization`` bearer header (not cookies), so the
+    # wildcard default is safe here — it is paired with ``allow_credentials=False``.
+    # Set explicit origins in hardened deployments. Accepts a comma-separated
+    # string in ``.env`` (``CORS_ALLOW_ORIGINS=https://a,https://b``) or ``*``.
+    cors_allow_origins: list[str] = Field(default_factory=lambda: ["*"])
 
     # -- subsonic ---------------------------------------------------------
     # Symmetric key (any string) used to encrypt each user's recoverable
@@ -54,10 +81,21 @@ class Settings(BaseSettings):
     # -- media / storage --------------------------------------------------
     media_path: Path = Path("/data/media")
     transcode_cache_path: Path = Path("/data/transcode-cache")
+    # ffmpeg binary for transcoding/HLS (on PATH in the image); override for a
+    # non-standard location.
+    ffmpeg_path: str = "ffmpeg"
     max_parallel_downloads: int = 2
+    # How many times the download worker retries a failed fetch (yt-dlp fails
+    # often) before marking the job ``failed`` — exponential backoff between tries.
+    download_max_retries: int = 3
     storage_backend: Literal["local", "s3"] = "local"
     upload_tmp_dir: Path | None = None
 
+    # -- sources ----------------------------------------------------------
+    # Mounted folder the ``local`` source indexes (copies into managed storage).
+    # Unset → the local source is simply not registered.
+    local_media_import_path: Path | None = None
+
     # -- S3 storage (deferred; set storage_backend="s3" to use) ----------
     s3_endpoint_url: str | None = None
     s3_bucket: str | None = None
@@ -68,9 +106,34 @@ class Settings(BaseSettings):
     # -- external services (all optional; graceful degradation) ----------
     ml_service_url: str | None = None
     acoustid_api_key: SecretStr | None = None
-    musicbrainz_user_agent: str = "mcma-backend/0.1.0 ( https://github.com/your/repo )"
+    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
+    # deployments should set their own contact email; see
+    # ``musicbrainz_user_agent`` below for how it's used.
+    musicbrainz_owner_email: str | None = None
+    # ``youtube`` fetch source (search + download via ytmusicapi/yt-dlp). Enabled
+    # by default; the source still reports unavailable if the libs aren't present.
+    youtube_enabled: bool = True
+    # Optional cookies file (Netscape format) for yt-dlp — lets it fetch
+    # age-restricted / region-locked items via an authenticated session.
     youtube_cookies_path: Path | None = None
 
+    # -- enrichment -------------------------------------------------------
+    # ``fpcalc`` (Chromaprint) binary; resolved on PATH by default. The Docker
+    # image installs it via libchromaprint-tools.
+    fpcalc_path: str = "fpcalc"
+
+    # Cover Art Archive — network fallback for album covers (after embedded art).
+    # Disable to keep enrichment fully offline; embedded artwork still works.
+    coverart_enabled: bool = True
+    coverart_base_url: str = "https://coverartarchive.org"
+
     @field_validator("database_url")
     @classmethod
     def _require_async_driver(cls, v: str) -> str:
@@ -78,10 +141,29 @@ class Settings(BaseSettings):
             raise ValueError("database_url must use the asyncpg driver: postgresql+asyncpg://")
         return v
 
+    @field_validator("cors_allow_origins", mode="before")
+    @classmethod
+    def _split_cors_origins(cls, v: object) -> object:
+        # Allow a plain comma-separated string in .env (pydantic would otherwise
+        # try to JSON-decode a list field): "a, b" -> ["a", "b"]; "*" -> ["*"].
+        if isinstance(v, str):
+            return [origin.strip() for origin in v.split(",") if origin.strip()]
+        return v
+
     @property
     def is_prod(self) -> bool:
         return self.environment == "prod"
 
+    @property
+    def musicbrainz_user_agent(self) -> str:
+        """User-Agent sent to MusicBrainz/AcoustID: ``MCMA/<version> ( <contact> )``.
+
+        Falls back to the project URL if the deployment hasn't set
+        ``musicbrainz_owner_email``.
+        """
+        contact = self.musicbrainz_owner_email or _PROJECT_URL
+        return f"{APP_NAME}/{app_version()} ( {contact} )"
+
 
 @lru_cache
 def get_settings() -> Settings:

app/domain/entities/__init__.py

@@ -1,21 +1,37 @@
 """Domain entities and value objects — pure, framework-free."""
 
 from app.domain.entities.album import Album
+from app.domain.entities.cover import CoverArt
+from app.domain.entities.download import DownloadJob
 from app.domain.entities.history import PlayHistoryEntry
 from app.domain.entities.like import Like
+from app.domain.entities.metadata import AudioTags, Fingerprint, RecordingMatch
 from app.domain.entities.playlist import Playlist
-from app.domain.entities.storage import ObjectStat
+from app.domain.entities.storage import (
+    DiskUsage,
+    FormatBreakdown,
+    LibraryStats,
+    ObjectStat,
+)
 from app.domain.entities.track import Artist, Track
 from app.domain.entities.user import Credentials, SubsonicCredentials, User
 
 __all__ = [
     "Album",
     "Artist",
+    "AudioTags",
+    "CoverArt",
     "Credentials",
+    "DiskUsage",
+    "DownloadJob",
+    "Fingerprint",
+    "FormatBreakdown",
+    "LibraryStats",
     "Like",
     "ObjectStat",
     "PlayHistoryEntry",
     "Playlist",
+    "RecordingMatch",
     "SubsonicCredentials",
     "Track",
     "User",

app/domain/entities/album.py

@@ -13,5 +13,7 @@ class Album:
     year: int | None
     cover_path: str | None
     musicbrainz_id: str | None
+    source: str | None
+    source_id: str | None
     created_at: dt.datetime
     updated_at: dt.datetime

app/domain/entities/cover.py

@@ -0,0 +1,28 @@
+"""Cover-art value object — raw image bytes plus their MIME type.
+
+Crosses the domain boundary between the cover sources (embedded extractor,
+Cover Art Archive) and the storage/serving layers. The bytes are the encoded
+image as-is; we never decode/resize in Phase 1.
+"""
+
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True, slots=True)
+class CoverArt:
+    data: bytes
+    content_type: str  # "image/jpeg" | "image/png" | …
+
+    @property
+    def extension(self) -> str:
+        """File extension for the content type (no leading dot)."""
+        return _EXT_BY_TYPE.get(self.content_type.lower(), "jpg")
+
+
+_EXT_BY_TYPE: dict[str, str] = {
+    "image/jpeg": "jpg",
+    "image/jpg": "jpg",
+    "image/png": "png",
+    "image/webp": "webp",
+    "image/gif": "gif",
+}

app/domain/entities/download.py

@@ -0,0 +1,26 @@
+"""Download job domain entity (plan §6.1).
+
+A queued fetch from an external source, tracked through its lifecycle so the UI
+download manager (screen §A5) can show progress, errors, and retries. The
+``status`` strings mirror :class:`~app.infrastructure.db.models.enums.DownloadStatus`.
+"""
+
+import datetime as dt
+import uuid
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True, slots=True)
+class DownloadJob:
+    id: uuid.UUID
+    source: str
+    source_id: str | None
+    query: str | None
+    requested_by: uuid.UUID | None
+    status: str
+    progress: float
+    error_message: str | None
+    retry_count: int
+    track_id: uuid.UUID | None
+    created_at: dt.datetime
+    updated_at: dt.datetime

app/domain/entities/lyrics.py

@@ -0,0 +1,41 @@
+"""Lyrics value objects (plan §6.7).
+
+``Lyrics`` is the cached row for a track; ``LyricsResult`` is what a provider
+(LRCLIB) returns for a lookup. Both cross the domain boundary — no framework
+imports. Status values mirror ``LyricsStatus`` in the ORM enum ("found" /
+"not_found" / "pending") but are kept as plain strings here so the domain stays
+independent of the persistence layer.
+"""
+
+import datetime as dt
+import uuid
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True, slots=True)
+class LyricsResult:
+    """A provider hit: synced (timestamped LRC) and/or plain text."""
+
+    synced: str | None
+    plain: str | None
+    source: str
+
+
+@dataclass(frozen=True, slots=True)
+class Lyrics:
+    """Cached lyrics for one track. ``status`` is ``found`` / ``not_found`` /
+    ``pending``; ``not_found`` is cached too (with a TTL in the service) so a
+    track with no lyrics doesn't hammer the provider on every play."""
+
+    track_id: uuid.UUID
+    synced: str | None
+    plain: str | None
+    source: str | None
+    status: str
+    fetched_at: dt.datetime
+
+    @property
+    def has_lyrics(self) -> bool:
+        return self.status == "found" and (
+            self.synced is not None or self.plain is not None
+        )

app/domain/entities/metadata.py

@@ -0,0 +1,54 @@
+"""Value objects for the metadata-enrichment pipeline (plan §6.2).
+
+Pure data carriers between the enrichment service and its adapters (tag reader,
+fingerprinter, AcoustID). No framework imports — these cross the domain boundary.
+"""
+
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True, slots=True)
+class AudioTags:
+    """Embedded tags read from the file itself (ID3 / Vorbis / MP4 …).
+
+    Every field is optional — files are tagged inconsistently. The reader fills
+    what it can and leaves the rest ``None`` for downstream identification.
+    """
+
+    title: str | None = None
+    artist: str | None = None
+    album: str | None = None
+    album_artist: str | None = None
+    genre: str | None = None
+    year: int | None = None
+    track_number: int | None = None
+    duration_seconds: int | None = None
+    bitrate: int | None = None
+
+
+@dataclass(frozen=True, slots=True)
+class Fingerprint:
+    """Chromaprint fingerprint plus the decoded duration (both needed by AcoustID)."""
+
+    fingerprint: str
+    duration_seconds: int
+
+
+@dataclass(frozen=True, slots=True)
+class RecordingMatch:
+    """A single AcoustID result, flattened to the fields enrichment cares about.
+
+    ``acoustid`` is the stable AcoustID identifier (a UUID) — used as the
+    dedup key persisted on ``track.acoustid_fingerprint`` (fits the 64-char
+    column; the raw fingerprint does not). ``recording_mbid`` is the MusicBrainz
+    recording id when present.
+    """
+
+    acoustid: str
+    score: float
+    recording_mbid: str | None = None
+    release_group_mbid: str | None = None
+    title: str | None = None
+    artist: str | None = None
+    album: str | None = None
+    year: int | None = None

app/domain/entities/settings.py

@@ -0,0 +1,34 @@
+"""User settings domain entity (general preferences + scrobbling config)."""
+
+import uuid
+from dataclasses import dataclass
+
+# Defaults for a user who has never saved settings — the row is created lazily,
+# so reads return these before the first write.
+DEFAULT_THEME = "system"
+DEFAULT_STREAM_QUALITY = "original"
+
+
+@dataclass(frozen=True, slots=True)
+class UserSettings:
+    user_id: uuid.UUID
+    theme: str
+    stream_quality: str
+    scrobble_enabled: bool
+    scrobble_provider: str | None
+    scrobble_username: str | None
+    # Scrobbler session key / user token, encrypted at rest (never leaves the
+    # server in plaintext). ``None`` until the user configures scrobbling.
+    scrobble_session_key_enc: str | None
+
+    @classmethod
+    def defaults(cls, user_id: uuid.UUID) -> UserSettings:
+        return cls(
+            user_id=user_id,
+            theme=DEFAULT_THEME,
+            stream_quality=DEFAULT_STREAM_QUALITY,
+            scrobble_enabled=False,
+            scrobble_provider=None,
+            scrobble_username=None,
+            scrobble_session_key_enc=None,
+        )

app/domain/entities/storage.py

@@ -1,5 +1,6 @@
 """Value objects for file storage."""
 
+import datetime as dt
 from dataclasses import dataclass
 
 
@@ -7,3 +8,39 @@ from dataclasses import dataclass
 class ObjectStat:
     size: int
     content_type: str | None
+
+
+@dataclass(frozen=True, slots=True)
+class DiskUsage:
+    """Capacity of the volume backing the media store. ``None`` for backends
+    (e.g. object stores) that expose no notion of total disk capacity."""
+
+    total: int
+    used: int
+    free: int
+
+
+@dataclass(frozen=True, slots=True)
+class FormatBreakdown:
+    """Per-container-format slice of the library (e.g. ``flac`` → 312 tracks)."""
+
+    file_format: str
+    track_count: int
+    total_size: int
+
+
+@dataclass(frozen=True, slots=True)
+class LibraryStats:
+    """Aggregate facts about everything the instance has stored. Computed from
+    the catalogue (DB), not the filesystem — ``total_size`` is the sum of the
+    recorded ``file_size`` of every track."""
+
+    total_tracks: int
+    total_size: int
+    total_duration_seconds: int
+    by_format: list[FormatBreakdown]
+    by_metadata_status: dict[str, int]
+    by_source: dict[str, int]
+    largest_track_size: int
+    earliest_added: dt.datetime | None
+    latest_added: dt.datetime | None

app/domain/entities/track.py

@@ -9,6 +9,8 @@ from dataclasses import dataclass
 class Artist:
     id: uuid.UUID
     name: str
+    source: str | None
+    source_id: str | None
     created_at: dt.datetime
     updated_at: dt.datetime
 
@@ -19,14 +21,18 @@ class Track:
     title: str
     artist_id: uuid.UUID
     album_id: uuid.UUID | None
-    storage_uri: str
+    storage_uri: str | None
-    file_format: str
+    file_format: str | None
-    file_size: int
+    file_size: int | None
     source: str
     source_id: str
     duration_seconds: int | None
     genre: str | None
     year: int | None
+    track_number: int | None
     metadata_status: str
+    metadata_error: str | None
+    enriched_at: dt.datetime | None
+    availability: str
     created_at: dt.datetime
     updated_at: dt.datetime

app/domain/errors.py

@@ -54,6 +54,13 @@ class PermissionDeniedError(DomainError):
     code = "permission_denied"
 
 
+class NotSupportedError(DomainError):
+    """Operation is intentionally unsupported (e.g. a config knob that is managed
+    via environment, not mutable at runtime)."""
+
+    code = "not_supported"
+
+
 class DependencyUnavailableError(DomainError):
     """An external dependency (source, ML, MusicBrainz) is unavailable.
 
@@ -69,6 +76,13 @@ class StorageError(DomainError):
     code = "storage_error"
 
 
+class TranscodeError(DomainError):
+    """Transcoding (ffmpeg) failed. Raised in the worker; a play falls back to
+    the original file rather than surfacing this."""
+
+    code = "transcode_error"
+
+
 class RangeNotSatisfiableError(DomainError):
     """Requested byte range cannot be satisfied."""
 

app/domain/ports.py

@@ -7,24 +7,39 @@ are bound to these ports at the composition root (``app.api.deps``).
 
 import datetime as dt
 import uuid
-from collections.abc import AsyncIterator
+from collections.abc import AsyncIterator, Awaitable, Callable, Iterator
 from contextlib import AbstractAsyncContextManager
 from pathlib import Path
 from typing import Protocol
 
 from app.domain.entities import (
     Album,
+    AudioTags,
+    CoverArt,
     Credentials,
+    DiskUsage,
+    DownloadJob,
+    Fingerprint,
+    LibraryStats,
     Like,
     ObjectStat,
     PlayHistoryEntry,
     Playlist,
+    RecordingMatch,
     SubsonicCredentials,
     User,
 )
+from app.domain.entities.lyrics import Lyrics, LyricsResult
+from app.domain.entities.settings import UserSettings
 from app.domain.entities.track import Artist, Track
+from app.domain.sources import DownloadResult, RawMetadata, SearchResult, SourceFile, SourceInfo
 from app.domain.tokens import IssuedToken, TokenClaims, TokenType
 
+# A fetch source reports download progress as a fraction in [0.0, 1.0]. It's a
+# plain callback (not a port) because it's an inversion of control supplied per
+# call by the worker, which persists it to the download job.
+ProgressCallback = Callable[[float], Awaitable[None]]
+
 
 class UserRepository(Protocol):
     async def get_by_id(self, user_id: uuid.UUID) -> User | None: ...
@@ -43,6 +58,11 @@ class UserRepository(Protocol):
     async def set_subsonic_password_enc(self, user_id: uuid.UUID, password_enc: str) -> None: ...
 
 
+class UserSettingsRepository(Protocol):
+    async def get(self, user_id: uuid.UUID) -> UserSettings | None: ...
+    async def upsert(self, settings: UserSettings) -> UserSettings: ...
+
+
 class SubsonicCipher(Protocol):
     """Symmetric encrypt/decrypt for the recoverable Subsonic app-password."""
 
@@ -93,12 +113,27 @@ class FileStorage(Protocol):
     async def exists(self, key: str) -> bool: ...
     async def delete(self, key: str) -> None: ...
     def as_local_path(self, key: str) -> AbstractAsyncContextManager[Path]: ...
+    async def disk_usage(self) -> DiskUsage | None:
+        """Capacity of the volume backing the store, or ``None`` when the
+        backend has no addressable disk (e.g. an object store)."""
+        ...
 
 
 class ArtistRepository(Protocol):
     async def get_or_create(self, name: str) -> Artist: ...
+    async def get_or_create_remote(self, *, name: str, source: str, source_id: str) -> Artist:
+        """Resolve/create an artist bound to a remote ``(source, source_id)``
+        (lazy materialization save-to-library)."""
+        ...
+
     async def get_by_id(self, artist_id: uuid.UUID) -> Artist | None: ...
     async def get_many(self, ids: list[uuid.UUID]) -> list[Artist]: ...
+    async def list_similar(self, *, artist_id: uuid.UUID, limit: int) -> list[Artist]:
+        """Artists sharing the seed artist's genres, ranked by overlap. Metadata
+        fallback for ``GET /artists/{id}/similar``. Defined before ``list`` so the
+        ``list[Artist]`` annotation isn't shadowed by the method named ``list``."""
+        ...
+
     async def list(self, *, q: str | None, limit: int, offset: int) -> list[Artist]: ...
     async def count(self, *, q: str | None) -> int: ...
     async def album_count(self, artist_id: uuid.UUID) -> int: ...
@@ -107,6 +142,11 @@ class ArtistRepository(Protocol):
 
 class TrackRepository(Protocol):
     async def get_by_id(self, track_id: uuid.UUID) -> Track | None: ...
+    async def get_many(self, ids: list[uuid.UUID]) -> list[Track]:
+        """Resolve multiple ids in one query (unordered) — batches the per-id
+        lookups radio/similar would otherwise fan out into N round-trips."""
+        ...
+
     async def get_by_source(self, source: str, source_id: str) -> Track | None: ...
     async def add(
         self,
@@ -114,24 +154,68 @@ class TrackRepository(Protocol):
         id: uuid.UUID,
         title: str,
         artist_id: uuid.UUID,
-        storage_uri: str,
+        storage_uri: str | None,
-        file_format: str,
+        file_format: str | None,
-        file_size: int,
+        file_size: int | None,
         source: str,
         source_id: str,
         metadata_status: str,
         added_by: uuid.UUID | None,
+        availability: str = ...,
     ) -> Track: ...
+    async def materialize(
+        self,
+        track_id: uuid.UUID,
+        *,
+        storage_uri: str,
+        file_format: str,
+        file_size: int,
+        bitrate: int | None,
+    ) -> Track:
+        """Fill in a remote placeholder's audio fields after a download
+        (lazy materialization), flipping ``availability`` to ``local``."""
+        ...
+
     async def delete(self, track_id: uuid.UUID) -> None: ...
-    # genres must come before ``list`` — the method named ``list`` shadows the
+    # genres / library_stats must come before ``list`` — the method named
-    # builtin in later annotations (same pattern as AlbumRepository below).
+    # ``list`` shadows the builtin in later annotations (same pattern as
+    # AlbumRepository below).
     async def genres(self) -> list[tuple[str, int]]: ...
+    async def library_stats(self) -> LibraryStats: ...
+    async def list_similar(
+        self,
+        *,
+        genre: str | None,
+        artist_id: uuid.UUID,
+        exclude_ids: list[uuid.UUID],
+        limit: int,
+    ) -> list[Track]:
+        """Playable tracks resembling a seed (same genre and/or artist), ranked
+        by match strength then shuffled. The metadata fallback for §6.5 radio /
+        similar when no ML service is configured."""
+        ...
+
+    async def sample_playable(
+        self, *, exclude_ids: list[uuid.UUID], limit: int
+    ) -> list[Track]:
+        """Random playable tracks — the exploration filler for radio."""
+        ...
+    async def find_duplicate_groups(self) -> list[tuple[str, list[Track]]]: ...
+    async def list_by_metadata_status(
+        self, status: str, *, limit: int, offset: int
+    ) -> list[Track]: ...
+    async def all_storage_refs(self) -> list[tuple[uuid.UUID, str]]: ...
+    async def count_by_metadata_status(self, status: str) -> int: ...
+    async def list_changed_since(
+        self, *, since: dt.datetime | None, until: dt.datetime
+    ) -> list[Track]: ...
     async def list(
         self,
         *,
         artist_id: uuid.UUID | None,
         album_id: uuid.UUID | None,
         q: str | None,
+        source: str | None = None,
         sort_by: str,
         order: str,
         limit: int,
@@ -143,6 +227,7 @@ class TrackRepository(Protocol):
         artist_id: uuid.UUID | None,
         album_id: uuid.UUID | None,
         q: str | None,
+        source: str | None = None,
     ) -> int: ...
     async def update(
         self,
@@ -152,9 +237,60 @@ class TrackRepository(Protocol):
         genre: str | None,
         year: int | None,
     ) -> Track: ...
+    async def apply_enrichment(
+        self,
+        track_id: uuid.UUID,
+        *,
+        title: str,
+        artist_id: uuid.UUID,
+        album_id: uuid.UUID | None,
+        genre: str | None,
+        year: int | None,
+        track_number: int | None,
+        duration_seconds: int | None,
+        bitrate: int | None,
+        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, 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."""
+        ...
 
 
 class AlbumRepository(Protocol):
+    async def get_or_create(
+        self,
+        *,
+        title: str,
+        artist_id: uuid.UUID,
+        year: int | None,
+        musicbrainz_id: str | None,
+    ) -> Album: ...
+    async def get_or_create_remote(
+        self,
+        *,
+        title: str,
+        artist_id: uuid.UUID,
+        year: int | None,
+        musicbrainz_id: str | None,
+        source: str,
+        source_id: str,
+    ) -> Album:
+        """Resolve/create an album bound to a remote ``(source, source_id)``
+        (lazy materialization save-to-library)."""
+        ...
+
+    async def set_cover_path(self, album_id: uuid.UUID, cover_path: str) -> None: ...
     async def get_by_id(self, album_id: uuid.UUID) -> Album | None: ...
     async def get_many(self, ids: list[uuid.UUID]) -> list[Album]: ...
     async def count(self, *, artist_id: uuid.UUID | None, q: str | None) -> int: ...
@@ -187,17 +323,38 @@ class PlaylistRepository(Protocol):
         self, playlist_id: uuid.UUID, *, limit: int, offset: int
     ) -> list[Track]: ...
     async def get_track_total(self, playlist_id: uuid.UUID) -> int: ...
+    async def has_track(self, playlist_id: uuid.UUID, track_id: uuid.UUID) -> bool: ...
     async def add_track(
         self, playlist_id: uuid.UUID, track_id: uuid.UUID, *, position: float
     ) -> None: ...
     async def remove_track(self, playlist_id: uuid.UUID, track_id: uuid.UUID) -> None: ...
     async def max_position(self, playlist_id: uuid.UUID) -> float: ...
+    async def reorder_tracks(
+        self, playlist_id: uuid.UUID, ordered_track_ids: list[uuid.UUID]
+    ) -> None: ...
+    async def get_cover_path(self, playlist_id: uuid.UUID) -> str | None: ...
+    async def list_changed_since(
+        self, *, owner_id: uuid.UUID, since: dt.datetime | None, until: dt.datetime
+    ) -> list[Playlist]: ...
+    async def track_ids(self, playlist_id: uuid.UUID) -> list[uuid.UUID]: ...
     # list must come after any method using list[...] in its signature (name shadowing)
     async def list(self, *, owner_id: uuid.UUID, limit: int, offset: int) -> list[Playlist]: ...
 
 
 class LikeRepository(Protocol):
     async def add(self, *, user_id: uuid.UUID, track_id: uuid.UUID, value: str) -> Like: ...
+    async def add_event(
+        self,
+        *,
+        id: uuid.UUID,
+        user_id: uuid.UUID,
+        track_id: uuid.UUID,
+        value: str,
+        created_at: dt.datetime,
+    ) -> bool: ...
+    async def list_since(
+        self, user_id: uuid.UUID, *, since: dt.datetime | None, until: dt.datetime
+    ) -> list[Like]: ...
     async def get_latest_state(
         self, *, user_id: uuid.UUID, track_ids: list[uuid.UUID]
     ) -> list[Like]: ...
@@ -217,7 +374,218 @@ class HistoryRepository(Protocol):
         play_duration_seconds: int | None,
         completed: bool,
     ) -> PlayHistoryEntry: ...
+    async def add_event(
+        self,
+        *,
+        id: uuid.UUID,
+        user_id: uuid.UUID,
+        track_id: uuid.UUID,
+        played_at: dt.datetime,
+        play_duration_seconds: int | None,
+        completed: bool,
+    ) -> bool: ...
+    async def list_since(
+        self, user_id: uuid.UUID, *, since: dt.datetime | None, until: dt.datetime
+    ) -> list[PlayHistoryEntry]: ...
     async def list(
         self, *, user_id: uuid.UUID, limit: int, offset: int
     ) -> list[PlayHistoryEntry]: ...
     async def count(self, *, user_id: uuid.UUID) -> int: ...
+
+
+class DownloadJobRepository(Protocol):
+    """Persistence for download jobs (plan §6.1). Drives the §A5 download manager
+    and the worker's retry/backoff loop."""
+
+    async def add(
+        self,
+        *,
+        source: str,
+        source_id: str | None,
+        query: str | None,
+        requested_by: uuid.UUID | None,
+    ) -> DownloadJob: ...
+    async def get_by_id(self, job_id: uuid.UUID) -> DownloadJob | None: ...
+    async def get_active_for_source(self, source: str, source_id: str) -> DownloadJob | None:
+        """An unfinished (queued/downloading/enriching) job for the same item, if
+        any — used to dedup before enqueuing so a double-click can't queue twice."""
+        ...
+
+    async def list(
+        self,
+        *,
+        requested_by: uuid.UUID | None,
+        status: str | None,
+        limit: int,
+        offset: int,
+    ) -> list[DownloadJob]: ...
+    async def count(self, *, requested_by: uuid.UUID | None, status: str | None) -> int: ...
+    async def set_status(
+        self,
+        job_id: uuid.UUID,
+        *,
+        status: str,
+        error_message: str | None = None,
+        track_id: uuid.UUID | None = None,
+    ) -> None: ...
+    async def set_progress(self, job_id: uuid.UUID, progress: float) -> None: ...
+    async def increment_retry(self, job_id: uuid.UUID) -> int:
+        """Bump ``retry_count`` and return the new value."""
+        ...
+
+    async def delete(self, job_id: uuid.UUID) -> None: ...
+    async def failure_rate(self, source: str, *, since: dt.datetime) -> float:
+        """Fraction of jobs for ``source`` created since ``since`` that ended
+        ``failed`` (0.0 when there are none) — drives the §A5 "source unhealthy"
+        banner."""
+        ...
+
+
+class SourceBackend(Protocol):
+    """A registered source of tracks (mounted folder, YouTube, …).
+
+    ``name`` is the stable identifier used in URLs and stored on ``track.source``.
+    """
+
+    name: str
+
+    def info(self) -> SourceInfo: ...
+    def is_available(self) -> bool: ...
+
+
+class IndexableSource(SourceBackend, Protocol):
+    """A source that enumerates files already on disk (e.g. the local folder)."""
+
+    def scan(self) -> Iterator[SourceFile]: ...
+
+
+class SearchableSource(SourceBackend, Protocol):
+    """A source that can be searched by free text (e.g. YouTube Music).
+
+    Returns ``[]`` (never raises) on no results / the service being down — the
+    discover screen degrades to "nothing found" rather than erroring."""
+
+    async def search(self, query: str, *, limit: int) -> list[SearchResult]: ...
+
+
+class FetchableSource(SourceBackend, Protocol):
+    """A source that can download a previously-discovered item to local disk.
+
+    ``fetch`` resolves a ``source_id`` (from a :class:`SearchResult`) into a file
+    and reports progress through ``on_progress``. It runs only in a worker (heavy
+    I/O) and raises on failure so the download task can retry with backoff."""
+
+    async def fetch(
+        self, source_id: str, *, on_progress: ProgressCallback | None = None
+    ) -> DownloadResult: ...
+
+    async def get_metadata(self, source_id: str) -> RawMetadata | None: ...
+
+
+# -- metadata enrichment (plan §6.2) -----------------------------------------
+class AudioTagReader(Protocol):
+    """Reads embedded tags from a local audio file. Returns ``None`` only when
+    the file can't be parsed at all — never raises (graceful degradation)."""
+
+    async def read(self, path: Path) -> AudioTags | None: ...
+
+
+class AudioFingerprinter(Protocol):
+    """Chromaprint (fpcalc) wrapper. ``is_available`` reflects whether the
+    binary is present; ``calculate`` returns ``None`` on any failure."""
+
+    def is_available(self) -> bool: ...
+    async def calculate(self, path: Path) -> Fingerprint | None: ...
+
+
+class AcoustIdClient(Protocol):
+    """AcoustID lookup. ``is_available`` is False without an API key (the whole
+    fingerprint path is then skipped). ``lookup`` returns the best match or
+    ``None`` (no result / service down), never raising. ``lookup_all`` returns
+    the same candidates ranked by confidence (``[]`` on no result / unavailable
+    / error), for the metadata editor's match picker."""
+
+    def is_available(self) -> bool: ...
+    async def lookup(self, fingerprint: Fingerprint) -> RecordingMatch | None: ...
+    async def lookup_all(self, fingerprint: Fingerprint) -> list[RecordingMatch]: ...
+
+
+class CoverArtExtractor(Protocol):
+    """Pulls embedded cover art out of a local audio file (offline, no network).
+    Returns ``None`` when the file has no picture or can't be parsed — never raises."""
+
+    async def extract(self, path: Path) -> CoverArt | None: ...
+
+
+class CoverArtProvider(Protocol):
+    """Fetches cover art from an external service (Cover Art Archive) by
+    MusicBrainz release-group id. ``is_available`` may gate it off; ``fetch``
+    returns ``None`` (not found / service down), never raising."""
+
+    def is_available(self) -> bool: ...
+    async def fetch_release_group(self, release_group_mbid: str) -> CoverArt | None: ...
+
+
+class Recommender(Protocol):
+    """External ML recommender (plan §6.5, ``ML_SERVICE_URL``). Returns ordered
+    track/artist ids, or ``None`` when unavailable/erroring so the service falls
+    back to metadata heuristics — ML is never a hard dependency (invariant)."""
+
+    def is_available(self) -> bool: ...
+
+    async def similar_track_ids(
+        self, track_id: uuid.UUID, *, limit: int, exclude_ids: list[uuid.UUID]
+    ) -> list[uuid.UUID] | None: ...
+
+    async def similar_artist_ids(
+        self, artist_id: uuid.UUID, *, limit: int
+    ) -> list[uuid.UUID] | None: ...
+
+    async def radio_track_ids(
+        self,
+        *,
+        seed_track_id: uuid.UUID | None,
+        exploration: float,
+        limit: int,
+        exclude_ids: list[uuid.UUID],
+    ) -> list[uuid.UUID] | None: ...
+
+
+class Transcoder(Protocol):
+    """Transcodes an audio file with ffmpeg (plan §6.6 / Group B). ``to_opus``
+    writes a single Opus rendition; ``to_hls`` writes an HLS playlist + segments
+    (AAC-in-TS) into ``out_dir``. Both raise ``TranscodeError`` on failure; heavy
+    work always runs in a worker, never the request cycle."""
+
+    async def to_opus(self, src: Path, dest: Path, *, bitrate_kbps: int) -> None: ...
+    async def to_hls(self, src: Path, out_dir: Path, *, bitrate_kbps: int) -> None: ...
+
+
+class LyricsProvider(Protocol):
+    """Fetches lyrics from an external database (LRCLIB) by artist/title/album/
+    duration. Returns a hit or ``None`` (no match / service down), never raising."""
+
+    async def fetch(
+        self,
+        *,
+        artist: str,
+        title: str,
+        album: str | None,
+        duration_seconds: int | None,
+    ) -> LyricsResult | None: ...
+
+
+class LyricsRepository(Protocol):
+    """Cached lyrics, one row per track. ``upsert`` also caches a ``not_found``
+    (empty text) so misses aren't re-fetched until the service's TTL lapses."""
+
+    async def get(self, track_id: uuid.UUID) -> Lyrics | None: ...
+    async def upsert(
+        self,
+        *,
+        track_id: uuid.UUID,
+        synced: str | None,
+        plain: str | None,
+        source: str | None,
+        status: str,
+    ) -> Lyrics: ...

app/domain/sources.py

@@ -0,0 +1,95 @@
+"""Source-backend value objects — framework-free.
+
+A *source* is a place tracks come from (a mounted folder, YouTube, an upload).
+Backends are driven adapters (``app.infrastructure.sources``); these are the
+shapes they speak in, and the ports they satisfy live in ``app.domain.ports``.
+
+The first backend, ``local``, is *indexable*: it enumerates files already on
+disk. Concrete metadata (artist/album/tags) is intentionally **not** resolved
+here — a source yields a file plus a minimal title; enrichment (plan §6.2) fills
+the rest later, so this stays a thin discovery layer (CLAUDE.md: no duplicated
+business logic)."""
+
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any
+
+# A source's ``kind`` describes which ports it satisfies, so the UI/admin can
+# tell an indexed folder from a searchable fetch-source. A backend may be both.
+KIND_INDEXABLE = "indexable"  # enumerates files already on disk (local folder)
+KIND_FETCH = "fetch"  # searches + downloads from an external service (YTM, …)
+
+
+@dataclass(frozen=True, slots=True)
+class SourceInfo:
+    """Describes a registered source for enumeration / health (UI, admin)."""
+
+    name: str
+    label: str
+    kind: str  # KIND_INDEXABLE | KIND_FETCH
+    available: bool
+
+
+@dataclass(frozen=True, slots=True)
+class SourceFile:
+    """A single importable file discovered by an indexable source.
+
+    ``source_id`` is stable per source (the local backend uses the path relative
+    to its root) so re-scans are idempotent — already-imported files are skipped.
+    """
+
+    source_id: str
+    path: Path
+    suggested_title: str
+    file_format: str
+    file_size: int
+
+
+@dataclass(frozen=True, slots=True)
+class SearchResult:
+    """One hit from a searchable source (plan §5), shown on the discover screen.
+
+    ``source_id`` is the stable handle the same backend later resolves in
+    ``fetch`` — it must round-trip a download request without re-searching.
+    ``raw`` carries the backend's untouched payload for debugging / future use.
+    """
+
+    source: str
+    source_id: str
+    title: str
+    artist: str | None
+    album: str | None
+    duration_seconds: int | None
+    thumbnail_url: str | None
+    raw: dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass(frozen=True, slots=True)
+class RawMetadata:
+    """Metadata a fetch-source can offer about an item *before* enrichment.
+
+    Best-effort and source-shaped — the canonical metadata still comes from the
+    enrichment pipeline (plan §6.2). Used to seed a more useful provisional
+    title than a bare id while a download is queued."""
+
+    title: str | None
+    artist: str | None
+    album: str | None
+    year: int | None
+    extra: dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass(frozen=True, slots=True)
+class DownloadResult:
+    """A file a fetch-source produced on local disk (plan §5).
+
+    ``path`` is a temp file the caller owns: it is stored into managed storage
+    and then removed (same lifecycle as an upload). ``source_id`` is echoed back
+    because some backends only learn the canonical id during the download."""
+
+    source_id: str
+    path: Path
+    file_format: str
+    file_size: int
+    bitrate: int | None
+    suggested_title: str

app/infrastructure/db/models/__init__.py

@@ -14,6 +14,7 @@ from app.infrastructure.db.models.play_history import PlayHistoryModel
 from app.infrastructure.db.models.playlist import PlaylistModel, PlaylistTrackModel
 from app.infrastructure.db.models.track import TrackModel
 from app.infrastructure.db.models.user import RefreshTokenModel, UserModel
+from app.infrastructure.db.models.user_settings import UserSettingsModel
 
 __all__ = [
     "AlbumModel",
@@ -27,4 +28,5 @@ __all__ = [
     "RefreshTokenModel",
     "TrackModel",
     "UserModel",
+    "UserSettingsModel",
 ]

app/infrastructure/db/models/album.py

@@ -2,7 +2,7 @@
 
 import uuid
 
-from sqlalchemy import ForeignKey, Integer, String
+from sqlalchemy import ForeignKey, Integer, String, UniqueConstraint
 from sqlalchemy.orm import Mapped, mapped_column
 
 from app.infrastructure.db.base import Base
@@ -11,6 +11,12 @@ from app.infrastructure.db.models.mixins import TimestampMixin, UUIDPrimaryKeyMi
 
 class AlbumModel(UUIDPrimaryKeyMixin, TimestampMixin, Base):
     __tablename__ = "albums"
+    __table_args__ = (
+        # Binds a remote (browsable) album to its local row for re-browse/save
+        # dedup. Multiple NULLs are allowed by Postgres, so locally-created
+        # albums (source/source_id both NULL) never collide on this.
+        UniqueConstraint("source", "source_id", name="uq_albums_source_source_id"),
+    )
 
     title: Mapped[str] = mapped_column(String(1024), index=True, nullable=False)
     artist_id: Mapped[uuid.UUID] = mapped_column(
@@ -21,3 +27,7 @@ class AlbumModel(UUIDPrimaryKeyMixin, TimestampMixin, Base):
     year: Mapped[int | None] = mapped_column(Integer, nullable=True)
     cover_path: Mapped[str | None] = mapped_column(String(1024), nullable=True)
     musicbrainz_id: Mapped[str | None] = mapped_column(String(36), index=True, nullable=True)
+
+    # -- remote identity (lazy materialization) --------------------------
+    source: Mapped[str | None] = mapped_column(String(32), nullable=True)
+    source_id: Mapped[str | None] = mapped_column(String(512), nullable=True)

app/infrastructure/db/models/artist.py

@@ -1,6 +1,6 @@
 """ORM model for artists."""
 
-from sqlalchemy import String
+from sqlalchemy import String, UniqueConstraint
 from sqlalchemy.orm import Mapped, mapped_column
 
 from app.infrastructure.db.base import Base
@@ -9,6 +9,16 @@ from app.infrastructure.db.models.mixins import TimestampMixin, UUIDPrimaryKeyMi
 
 class ArtistModel(UUIDPrimaryKeyMixin, TimestampMixin, Base):
     __tablename__ = "artists"
+    __table_args__ = (
+        # Binds a remote (browsable) artist to its local row for re-browse/save
+        # dedup. Multiple NULLs are allowed by Postgres, so locally-created
+        # artists (source/source_id both NULL) never collide on this.
+        UniqueConstraint("source", "source_id", name="uq_artists_source_source_id"),
+    )
 
     name: Mapped[str] = mapped_column(String(512), index=True, nullable=False)
     musicbrainz_id: Mapped[str | None] = mapped_column(String(36), index=True, nullable=True)
+
+    # -- remote identity (lazy materialization) --------------------------
+    source: Mapped[str | None] = mapped_column(String(32), nullable=True)
+    source_id: Mapped[str | None] = mapped_column(String(512), nullable=True)

app/infrastructure/db/models/download_job.py

@@ -35,3 +35,9 @@ class DownloadJobModel(UUIDPrimaryKeyMixin, TimestampMixin, Base):
     progress: Mapped[float] = mapped_column(Float, nullable=False, default=0.0)
     error_message: Mapped[str | None] = mapped_column(Text, nullable=True)
     retry_count: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
+    # Set once the download finishes and the track is imported — lets the UI
+    # link a completed job to its library track.
+    track_id: Mapped[uuid.UUID | None] = mapped_column(
+        ForeignKey("tracks.id", ondelete="SET NULL"),
+        nullable=True,
+    )

app/infrastructure/db/models/enums.py

@@ -64,3 +64,12 @@ class LyricsStatus(enum.StrEnum):
     FOUND = "found"
     NOT_FOUND = "not_found"
     PENDING = "pending"
+
+
+class TrackAvailability(enum.StrEnum):
+    """Whether a track's audio is on local storage or still a remote placeholder
+    (plan: lazy materialization). ``remote`` tracks have ``storage_uri = NULL``
+    until ``TrackRepository.materialize`` fills it in."""
+
+    LOCAL = "local"
+    REMOTE = "remote"

app/infrastructure/db/models/like.py

@@ -37,3 +37,12 @@ class LikeModel(UUIDPrimaryKeyMixin, Base):
         server_default=func.now(),
         nullable=False,
     )
+    # Server ingestion time — the delta-sync ordering key. Set once at insert and
+    # never changed; distinct from ``created_at`` (the real event time, which a
+    # sync push preserves from the client even when it happened offline earlier).
+    synced_at: Mapped[dt.datetime] = mapped_column(
+        DateTime(timezone=True),
+        server_default=func.now(),
+        nullable=False,
+        index=True,
+    )

app/infrastructure/db/models/play_history.py

@@ -34,3 +34,11 @@ class PlayHistoryModel(UUIDPrimaryKeyMixin, Base):
     )
     play_duration_seconds: Mapped[int | None] = mapped_column(Integer, nullable=True)
     completed: Mapped[bool] = mapped_column(Boolean, nullable=False, default=False)
+    # Server ingestion time — the delta-sync ordering key (see LikeModel). Distinct
+    # from ``played_at`` (the real play time), which a sync push preserves.
+    synced_at: Mapped[dt.datetime] = mapped_column(
+        DateTime(timezone=True),
+        server_default=func.now(),
+        nullable=False,
+        index=True,
+    )

app/infrastructure/db/models/track.py

@@ -6,13 +6,14 @@
 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
-from app.infrastructure.db.models.enums import MetadataStatus, StoragePolicy
+from app.infrastructure.db.models.enums import MetadataStatus, StoragePolicy, TrackAvailability
 from app.infrastructure.db.models.mixins import TimestampMixin, UUIDPrimaryKeyMixin
 
 
@@ -40,11 +41,20 @@ class TrackModel(UUIDPrimaryKeyMixin, TimestampMixin, Base):
     year: Mapped[int | None] = mapped_column(Integer, nullable=True)
 
     # -- file (original, stored as-is) -----------------------------------
-    storage_uri: Mapped[str] = mapped_column(String(2048), nullable=False)
+    # NULL on a remote placeholder (not yet materialized) — see ``availability``.
-    file_format: Mapped[str] = mapped_column(String(32), nullable=False)
+    storage_uri: Mapped[str | None] = mapped_column(String(2048), nullable=True)
-    file_size: Mapped[int] = mapped_column(Integer, nullable=False)
+    file_format: Mapped[str | None] = mapped_column(String(32), nullable=True)
+    file_size: Mapped[int | None] = mapped_column(Integer, nullable=True)
     bitrate: Mapped[int | None] = mapped_column(Integer, nullable=True)
 
+    # ``remote`` = placeholder with no local audio yet; materialize() flips this
+    # to ``local`` once the file is downloaded and ``storage_uri`` is filled in.
+    availability: Mapped[str] = mapped_column(
+        String(16),
+        nullable=False,
+        default=TrackAvailability.LOCAL.value,
+    )
+
     # -- dedup / external ids --------------------------------------------
     acoustid_fingerprint: Mapped[str | None] = mapped_column(String(64), index=True, nullable=True)
     musicbrainz_id: Mapped[str | None] = mapped_column(String(36), index=True, nullable=True)
@@ -63,6 +73,16 @@ 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/models/user_settings.py

@@ -0,0 +1,29 @@
+"""ORM model for per-user settings (general preferences + scrobbling)."""
+
+import uuid
+
+from sqlalchemy import Boolean, ForeignKey, String
+from sqlalchemy.orm import Mapped, mapped_column
+
+from app.infrastructure.db.base import Base
+from app.infrastructure.db.models.mixins import TimestampMixin
+
+
+class UserSettingsModel(TimestampMixin, Base):
+    """One row per user, created lazily on first save. The primary key *is* the
+    user id (a 1:1 extension of ``users``), so there's no separate surrogate id."""
+
+    __tablename__ = "user_settings"
+
+    user_id: Mapped[uuid.UUID] = mapped_column(
+        ForeignKey("users.id", ondelete="CASCADE"),
+        primary_key=True,
+    )
+    theme: Mapped[str] = mapped_column(String(16), default="system", nullable=False)
+    stream_quality: Mapped[str] = mapped_column(String(16), default="original", nullable=False)
+    scrobble_enabled: Mapped[bool] = mapped_column(Boolean, default=False, nullable=False)
+    scrobble_provider: Mapped[str | None] = mapped_column(String(16), nullable=True)
+    scrobble_username: Mapped[str | None] = mapped_column(String(255), nullable=True)
+    # Fernet-encrypted scrobbler session key / token (see core.security). Never
+    # the plaintext — mirrors how the Subsonic app-password is stored.
+    scrobble_session_key_enc: Mapped[str | None] = mapped_column(String(512), nullable=True)

app/infrastructure/db/repositories/__init__.py

@@ -2,22 +2,32 @@
 
 from app.infrastructure.db.repositories.album_repository import SqlAlchemyAlbumRepository
 from app.infrastructure.db.repositories.artist_repository import SqlAlchemyArtistRepository
+from app.infrastructure.db.repositories.download_job_repository import (
+    SqlAlchemyDownloadJobRepository,
+)
 from app.infrastructure.db.repositories.history_repository import SqlAlchemyHistoryRepository
 from app.infrastructure.db.repositories.like_repository import SqlAlchemyLikeRepository
+from app.infrastructure.db.repositories.lyrics_repository import SqlAlchemyLyricsRepository
 from app.infrastructure.db.repositories.playlist_repository import SqlAlchemyPlaylistRepository
 from app.infrastructure.db.repositories.refresh_token_repository import (
     SqlAlchemyRefreshTokenRepository,
 )
 from app.infrastructure.db.repositories.track_repository import SqlAlchemyTrackRepository
 from app.infrastructure.db.repositories.user_repository import SqlAlchemyUserRepository
+from app.infrastructure.db.repositories.user_settings_repository import (
+    SqlAlchemyUserSettingsRepository,
+)
 
 __all__ = [
     "SqlAlchemyAlbumRepository",
     "SqlAlchemyArtistRepository",
+    "SqlAlchemyDownloadJobRepository",
     "SqlAlchemyHistoryRepository",
     "SqlAlchemyLikeRepository",
+    "SqlAlchemyLyricsRepository",
     "SqlAlchemyPlaylistRepository",
     "SqlAlchemyRefreshTokenRepository",
     "SqlAlchemyTrackRepository",
     "SqlAlchemyUserRepository",
+    "SqlAlchemyUserSettingsRepository",
 ]

app/infrastructure/db/repositories/album_repository.py

@@ -18,6 +18,8 @@ def _to_entity(row: AlbumModel) -> Album:
         year=row.year,
         cover_path=row.cover_path,
         musicbrainz_id=row.musicbrainz_id,
+        source=row.source,
+        source_id=row.source_id,
         created_at=row.created_at,
         updated_at=row.updated_at,
     )
@@ -27,6 +29,100 @@ class SqlAlchemyAlbumRepository:
     def __init__(self, session: AsyncSession) -> None:
         self._session = session
 
+    async def get_or_create(
+        self,
+        *,
+        title: str,
+        artist_id: uuid.UUID,
+        year: int | None,
+        musicbrainz_id: str | None,
+    ) -> Album:
+        """Resolve an album by ``(title, artist_id)``, creating it if absent.
+        Backfills ``year``/``musicbrainz_id`` onto an existing row when it lacks
+        them and enrichment now has values (gap-fill, never overwrite)."""
+        row = (
+            await self._session.execute(
+                select(AlbumModel).where(
+                    AlbumModel.title == title,
+                    AlbumModel.artist_id == artist_id,
+                )
+            )
+        ).scalar_one_or_none()
+        if row is None:
+            row = AlbumModel(
+                title=title,
+                artist_id=artist_id,
+                year=year,
+                musicbrainz_id=musicbrainz_id,
+            )
+            self._session.add(row)
+        else:
+            if row.year is None and year is not None:
+                row.year = year
+            if row.musicbrainz_id is None and musicbrainz_id is not None:
+                row.musicbrainz_id = musicbrainz_id
+        await self._session.flush()
+        await self._session.refresh(row)
+        return _to_entity(row)
+
+    async def get_or_create_remote(
+        self,
+        *,
+        title: str,
+        artist_id: uuid.UUID,
+        year: int | None,
+        musicbrainz_id: str | None,
+        source: str,
+        source_id: str,
+    ) -> Album:
+        """Resolve an album by ``(source, source_id)`` first (re-browse/save
+        dedup), falling back to ``(title, artist_id)`` and gap-filling the
+        remote ids onto an existing row, else creating a new remote-bound row."""
+        row = (
+            await self._session.execute(
+                select(AlbumModel).where(
+                    AlbumModel.source == source,
+                    AlbumModel.source_id == source_id,
+                )
+            )
+        ).scalar_one_or_none()
+        if row is None:
+            row = (
+                await self._session.execute(
+                    select(AlbumModel).where(
+                        AlbumModel.title == title,
+                        AlbumModel.artist_id == artist_id,
+                    )
+                )
+            ).scalar_one_or_none()
+        if row is None:
+            row = AlbumModel(
+                title=title,
+                artist_id=artist_id,
+                year=year,
+                musicbrainz_id=musicbrainz_id,
+                source=source,
+                source_id=source_id,
+            )
+            self._session.add(row)
+        else:
+            if row.year is None and year is not None:
+                row.year = year
+            if row.musicbrainz_id is None and musicbrainz_id is not None:
+                row.musicbrainz_id = musicbrainz_id
+            if row.source is None and row.source_id is None:
+                row.source = source
+                row.source_id = source_id
+        await self._session.flush()
+        await self._session.refresh(row)
+        return _to_entity(row)
+
+    async def set_cover_path(self, album_id: uuid.UUID, cover_path: str) -> None:
+        row = await self._session.get(AlbumModel, album_id)
+        if row is not None:
+            row.cover_path = cover_path
+            await self._session.flush()
+
     async def get_by_id(self, album_id: uuid.UUID) -> Album | None:
         row = await self._session.get(AlbumModel, album_id)
         return _to_entity(row) if row is not None else None

app/infrastructure/db/repositories/artist_repository.py

@@ -15,6 +15,8 @@ def _to_entity(row: ArtistModel) -> Artist:
     return Artist(
         id=row.id,
         name=row.name,
+        source=row.source,
+        source_id=row.source_id,
         created_at=row.created_at,
         updated_at=row.updated_at,
     )
@@ -35,6 +37,32 @@ class SqlAlchemyArtistRepository:
             await self._session.refresh(row)
         return _to_entity(row)
 
+    async def get_or_create_remote(self, *, name: str, source: str, source_id: str) -> Artist:
+        """Resolve an artist by ``(source, source_id)`` first (re-browse/save
+        dedup), falling back to ``name`` and gap-filling the remote ids onto an
+        existing row, else creating a new remote-bound row."""
+        row = (
+            await self._session.execute(
+                select(ArtistModel).where(
+                    ArtistModel.source == source,
+                    ArtistModel.source_id == source_id,
+                )
+            )
+        ).scalar_one_or_none()
+        if row is None:
+            row = (
+                await self._session.execute(select(ArtistModel).where(ArtistModel.name == name))
+            ).scalar_one_or_none()
+        if row is None:
+            row = ArtistModel(name=name, source=source, source_id=source_id)
+            self._session.add(row)
+        elif row.source is None and row.source_id is None:
+            row.source = source
+            row.source_id = source_id
+        await self._session.flush()
+        await self._session.refresh(row)
+        return _to_entity(row)
+
     async def get_by_id(self, artist_id: uuid.UUID) -> Artist | None:
         row = await self._session.get(ArtistModel, artist_id)
         return _to_entity(row) if row is not None else None
@@ -49,6 +77,26 @@ class SqlAlchemyArtistRepository:
         )
         return [_to_entity(r) for r in rows]
 
+    async def list_similar(self, *, artist_id: uuid.UUID, limit: int) -> list[Artist]:
+        # Artists whose tracks fall in the seed artist's genres, ranked by how
+        # many such tracks they have. Defined before ``list`` so the ``list[Artist]``
+        # return annotation isn't shadowed by the method named ``list``.
+        seed_genres = (
+            select(TrackModel.genre)
+            .where(TrackModel.artist_id == artist_id, TrackModel.genre.is_not(None))
+            .distinct()
+        )
+        stmt = (
+            select(ArtistModel)
+            .join(TrackModel, TrackModel.artist_id == ArtistModel.id)
+            .where(TrackModel.genre.in_(seed_genres), ArtistModel.id != artist_id)
+            .group_by(ArtistModel.id)
+            .order_by(func.count(TrackModel.id).desc())
+            .limit(limit)
+        )
+        rows = (await self._session.execute(stmt)).scalars().all()
+        return [_to_entity(r) for r in rows]
+
     async def list(self, *, q: str | None, limit: int, offset: int) -> list[Artist]:
         stmt = select(ArtistModel)
         if q:
@@ -80,3 +128,4 @@ class SqlAlchemyArtistRepository:
                 .where(TrackModel.artist_id == artist_id)
             )
         ).scalar_one()
+

app/infrastructure/db/repositories/download_job_repository.py

@@ -0,0 +1,164 @@
+"""Download job repository — adapter over ``AsyncSession`` (plan §6.1)."""
+
+import datetime as dt
+import uuid
+
+from sqlalchemy import func, select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.domain.entities.download import DownloadJob
+from app.infrastructure.db.models.download_job import DownloadJobModel
+from app.infrastructure.db.models.enums import DownloadStatus
+
+# Jobs that are not yet finished — used to dedup an in-flight download.
+_ACTIVE_STATUSES = (
+    DownloadStatus.QUEUED.value,
+    DownloadStatus.DOWNLOADING.value,
+    DownloadStatus.ENRICHING.value,
+)
+
+
+def _to_entity(row: DownloadJobModel) -> DownloadJob:
+    return DownloadJob(
+        id=row.id,
+        source=row.source,
+        source_id=row.source_id,
+        query=row.query,
+        requested_by=row.requested_by,
+        status=row.status,
+        progress=row.progress,
+        error_message=row.error_message,
+        retry_count=row.retry_count,
+        track_id=row.track_id,
+        created_at=row.created_at,
+        updated_at=row.updated_at,
+    )
+
+
+class SqlAlchemyDownloadJobRepository:
+    def __init__(self, session: AsyncSession) -> None:
+        self._session = session
+
+    async def add(
+        self,
+        *,
+        source: str,
+        source_id: str | None,
+        query: str | None,
+        requested_by: uuid.UUID | None,
+    ) -> DownloadJob:
+        row = DownloadJobModel(
+            source=source,
+            source_id=source_id,
+            query=query,
+            requested_by=requested_by,
+            status=DownloadStatus.QUEUED.value,
+            progress=0.0,
+            retry_count=0,
+        )
+        self._session.add(row)
+        await self._session.flush()
+        await self._session.refresh(row)
+        return _to_entity(row)
+
+    async def get_by_id(self, job_id: uuid.UUID) -> DownloadJob | None:
+        row = await self._session.get(DownloadJobModel, job_id)
+        return _to_entity(row) if row is not None else None
+
+    async def get_active_for_source(self, source: str, source_id: str) -> DownloadJob | None:
+        row = (
+            await self._session.execute(
+                select(DownloadJobModel)
+                .where(
+                    DownloadJobModel.source == source,
+                    DownloadJobModel.source_id == source_id,
+                    DownloadJobModel.status.in_(_ACTIVE_STATUSES),
+                )
+                .order_by(DownloadJobModel.created_at.desc())
+                .limit(1)
+            )
+        ).scalar_one_or_none()
+        return _to_entity(row) if row is not None else None
+
+    async def list(
+        self,
+        *,
+        requested_by: uuid.UUID | None,
+        status: str | None,
+        limit: int,
+        offset: int,
+    ) -> list[DownloadJob]:
+        stmt = select(DownloadJobModel)
+        if requested_by is not None:
+            stmt = stmt.where(DownloadJobModel.requested_by == requested_by)
+        if status is not None:
+            stmt = stmt.where(DownloadJobModel.status == status)
+        stmt = stmt.order_by(DownloadJobModel.created_at.desc()).limit(limit).offset(offset)
+        rows = (await self._session.execute(stmt)).scalars().all()
+        return [_to_entity(r) for r in rows]
+
+    async def count(self, *, requested_by: uuid.UUID | None, status: str | None) -> int:
+        stmt = select(func.count()).select_from(DownloadJobModel)
+        if requested_by is not None:
+            stmt = stmt.where(DownloadJobModel.requested_by == requested_by)
+        if status is not None:
+            stmt = stmt.where(DownloadJobModel.status == status)
+        return (await self._session.execute(stmt)).scalar_one()
+
+    async def set_status(
+        self,
+        job_id: uuid.UUID,
+        *,
+        status: str,
+        error_message: str | None = None,
+        track_id: uuid.UUID | None = None,
+    ) -> None:
+        row = await self._session.get(DownloadJobModel, job_id)
+        if row is None:
+            return
+        row.status = status
+        # ``error_message`` is always written: a successful transition clears a
+        # stale reason from an earlier failed attempt.
+        row.error_message = error_message
+        if track_id is not None:
+            row.track_id = track_id
+        if status == DownloadStatus.DONE.value:
+            row.progress = 1.0
+        await self._session.flush()
+
+    async def set_progress(self, job_id: uuid.UUID, progress: float) -> None:
+        row = await self._session.get(DownloadJobModel, job_id)
+        if row is None:
+            return
+        row.progress = max(0.0, min(1.0, progress))
+        await self._session.flush()
+
+    async def increment_retry(self, job_id: uuid.UUID) -> int:
+        row = await self._session.get(DownloadJobModel, job_id)
+        if row is None:
+            return 0
+        row.retry_count += 1
+        await self._session.flush()
+        return row.retry_count
+
+    async def delete(self, job_id: uuid.UUID) -> None:
+        row = await self._session.get(DownloadJobModel, job_id)
+        if row is not None:
+            await self._session.delete(row)
+            await self._session.flush()
+
+    async def failure_rate(self, source: str, *, since: dt.datetime) -> float:
+        total, failed = (
+            await self._session.execute(
+                select(
+                    func.count(),
+                    func.count().filter(DownloadJobModel.status == DownloadStatus.FAILED.value),
+                )
+                .select_from(DownloadJobModel)
+                .where(
+                    DownloadJobModel.source == source,
+                    DownloadJobModel.created_at >= since,
+                )
+            )
+        ).one()
+        return (failed / total) if total else 0.0

app/infrastructure/db/repositories/history_repository.py

@@ -4,6 +4,7 @@ import datetime as dt
 import uuid
 
 from sqlalchemy import func, select
+from sqlalchemy.dialects.postgresql import insert as pg_insert
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.domain.entities.history import PlayHistoryEntry
@@ -46,6 +47,51 @@ class SqlAlchemyHistoryRepository:
         await self._session.refresh(row)
         return _to_entity(row)
 
+    async def add_event(
+        self,
+        *,
+        id: uuid.UUID,
+        user_id: uuid.UUID,
+        track_id: uuid.UUID,
+        played_at: dt.datetime,
+        play_duration_seconds: int | None,
+        completed: bool,
+    ) -> bool:
+        """Idempotent append for sync push: insert a client-generated play event,
+        skipping it if the ``id`` already exists (a replay). Returns whether a new
+        row was stored. Defined before ``list`` (name-shadowing)."""
+        stmt = (
+            pg_insert(PlayHistoryModel)
+            .values(
+                id=id,
+                user_id=user_id,
+                track_id=track_id,
+                played_at=played_at,
+                play_duration_seconds=play_duration_seconds,
+                completed=completed,
+            )
+            .on_conflict_do_nothing(index_elements=["id"])
+            .returning(PlayHistoryModel.id)
+        )
+        inserted = (await self._session.execute(stmt)).scalar_one_or_none()
+        return inserted is not None
+
+    async def list_since(
+        self, user_id: uuid.UUID, *, since: dt.datetime | None, until: dt.datetime
+    ) -> list[PlayHistoryEntry]:
+        """Play events for a user ingested in the half-open window ``(since,
+        until]`` (by ``synced_at``, the server-side sync key — so events pushed
+        with an older ``played_at`` still surface), oldest first. ``since=None``
+        returns everything up to ``until``."""
+        stmt = select(PlayHistoryModel).where(
+            PlayHistoryModel.user_id == user_id, PlayHistoryModel.synced_at <= until
+        )
+        if since is not None:
+            stmt = stmt.where(PlayHistoryModel.synced_at > since)
+        stmt = stmt.order_by(PlayHistoryModel.synced_at)
+        rows = (await self._session.execute(stmt)).scalars().all()
+        return [_to_entity(r) for r in rows]
+
     async def list(self, *, user_id: uuid.UUID, limit: int, offset: int) -> list[PlayHistoryEntry]:
         rows = (
             (

app/infrastructure/db/repositories/like_repository.py

@@ -3,9 +3,11 @@
 Likes are an append-only event log. Current state = latest event per (user, track).
 """
 
+import datetime as dt
 import uuid
 
-from sqlalchemy import func, select
+from sqlalchemy import Subquery, func, select
+from sqlalchemy.dialects.postgresql import insert as pg_insert
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.domain.entities.like import Like
@@ -38,7 +40,11 @@ def _track_to_entity(row: TrackModel) -> Track:
         duration_seconds=row.duration_seconds,
         genre=row.genre,
         year=row.year,
+        track_number=row.track_number,
         metadata_status=row.metadata_status,
+        metadata_error=row.metadata_error,
+        enriched_at=row.enriched_at,
+        availability=row.availability,
         created_at=row.created_at,
         updated_at=row.updated_at,
     )
@@ -55,31 +61,96 @@ class SqlAlchemyLikeRepository:
         await self._session.refresh(row)
         return _to_entity(row)
 
+    async def add_event(
+        self,
+        *,
+        id: uuid.UUID,
+        user_id: uuid.UUID,
+        track_id: uuid.UUID,
+        value: str,
+        created_at: dt.datetime,
+    ) -> bool:
+        """Idempotent append for sync push: insert a client-generated like event,
+        skipping it if the ``id`` already exists (a replay). Preserves the
+        client's ``created_at`` (the event happened offline earlier). Returns
+        whether a new row was stored."""
+        stmt = (
+            pg_insert(LikeModel)
+            .values(
+                id=id,
+                user_id=user_id,
+                track_id=track_id,
+                value=value,
+                created_at=created_at,
+            )
+            .on_conflict_do_nothing(index_elements=["id"])
+            .returning(LikeModel.id)
+        )
+        inserted = (await self._session.execute(stmt)).scalar_one_or_none()
+        return inserted is not None
+
+    async def list_since(
+        self, user_id: uuid.UUID, *, since: dt.datetime | None, until: dt.datetime
+    ) -> list[Like]:
+        """Like events for a user ingested in the half-open window ``(since,
+        until]`` (by ``synced_at``, the server-side sync key — so events pushed
+        with an older ``created_at`` still surface), oldest first. ``since=None``
+        returns everything up to ``until``."""
+        stmt = select(LikeModel).where(
+            LikeModel.user_id == user_id, LikeModel.synced_at <= until
+        )
+        if since is not None:
+            stmt = stmt.where(LikeModel.synced_at > since)
+        stmt = stmt.order_by(LikeModel.synced_at)
+        rows = (await self._session.execute(stmt)).scalars().all()
+        return [_to_entity(r) for r in rows]
+
+    def _latest_events_sq(
+        self, user_id: uuid.UUID, track_ids: list[uuid.UUID] | None
+    ) -> Subquery:
+        """The latest like event per ``track_id`` for a user, as a subquery.
+
+        ``DISTINCT ON (track_id)`` with a deterministic tiebreaker (``created_at``
+        then ``id``) picks exactly one row per track even when two events share an
+        identical ``created_at`` — likes carry a client-supplied timestamp from
+        offline sync, so ties are realistic and a plain ``max()``+equality-join
+        would return both rows (double-counting the track)."""
+        stmt = select(
+            LikeModel.track_id,
+            LikeModel.value.label("value"),
+            LikeModel.created_at.label("created_at"),
+        ).where(LikeModel.user_id == user_id)
+        if track_ids is not None:
+            stmt = stmt.where(LikeModel.track_id.in_(track_ids))
+        return (
+            stmt.distinct(LikeModel.track_id)
+            .order_by(
+                LikeModel.track_id,
+                LikeModel.created_at.desc(),
+                LikeModel.id.desc(),
+            )
+            .subquery()
+        )
+
     async def get_latest_state(
         self, *, user_id: uuid.UUID, track_ids: list[uuid.UUID]
     ) -> list[Like]:
         if not track_ids:
             return []
-        # Subquery: max(created_at) per track for this user
-        max_sq = (
-            select(
-                LikeModel.track_id,
-                func.max(LikeModel.created_at).label("latest"),
-            )
-            .where(LikeModel.user_id == user_id, LikeModel.track_id.in_(track_ids))
-            .group_by(LikeModel.track_id)
-            .subquery()
-        )
         rows = (
             (
                 await self._session.execute(
                     select(LikeModel)
-                    .join(
+                    .where(
-                        max_sq,
+                        LikeModel.user_id == user_id,
-                        (LikeModel.track_id == max_sq.c.track_id)
+                        LikeModel.track_id.in_(track_ids),
-                        & (LikeModel.created_at == max_sq.c.latest),
+                    )
+                    .distinct(LikeModel.track_id)
+                    .order_by(
+                        LikeModel.track_id,
+                        LikeModel.created_at.desc(),
+                        LikeModel.id.desc(),
                     )
-                    .where(LikeModel.user_id == user_id)
                 )
             )
             .scalars()
@@ -91,31 +162,14 @@ class SqlAlchemyLikeRepository:
         self, *, user_id: uuid.UUID, limit: int, offset: int
     ) -> list[Track]:
         # Tracks where the latest like event has value='like', ordered by like time desc
-        max_sq = (
+        latest_sq = self._latest_events_sq(user_id, None)
-            select(
-                LikeModel.track_id,
-                func.max(LikeModel.created_at).label("latest"),
-            )
-            .where(LikeModel.user_id == user_id)
-            .group_by(LikeModel.track_id)
-            .subquery()
-        )
-        liked_sq = (
-            select(LikeModel.track_id, LikeModel.created_at)
-            .join(
-                max_sq,
-                (LikeModel.track_id == max_sq.c.track_id)
-                & (LikeModel.created_at == max_sq.c.latest),
-            )
-            .where(LikeModel.user_id == user_id, LikeModel.value == "like")
-            .subquery()
-        )
         rows = (
             (
                 await self._session.execute(
                     select(TrackModel)
-                    .join(liked_sq, TrackModel.id == liked_sq.c.track_id)
+                    .join(latest_sq, TrackModel.id == latest_sq.c.track_id)
-                    .order_by(liked_sq.c.created_at.desc())
+                    .where(latest_sq.c.value == "like")
+                    .order_by(latest_sq.c.created_at.desc())
                     .limit(limit)
                     .offset(offset)
                 )
@@ -126,25 +180,11 @@ class SqlAlchemyLikeRepository:
         return [_track_to_entity(r) for r in rows]
 
     async def count_liked_tracks(self, *, user_id: uuid.UUID) -> int:
-        max_sq = (
+        latest_sq = self._latest_events_sq(user_id, None)
-            select(
-                LikeModel.track_id,
-                func.max(LikeModel.created_at).label("latest"),
-            )
-            .where(LikeModel.user_id == user_id)
-            .group_by(LikeModel.track_id)
-            .subquery()
-        )
-        liked_sq = (
-            select(LikeModel.track_id)
-            .join(
-                max_sq,
-                (LikeModel.track_id == max_sq.c.track_id)
-                & (LikeModel.created_at == max_sq.c.latest),
-            )
-            .where(LikeModel.user_id == user_id, LikeModel.value == "like")
-            .subquery()
-        )
         return (
-            await self._session.execute(select(func.count()).select_from(liked_sq))
+            await self._session.execute(
+                select(func.count())
+                .select_from(latest_sq)
+                .where(latest_sq.c.value == "like")
+            )
         ).scalar_one()

app/infrastructure/db/repositories/lyrics_repository.py

@@ -0,0 +1,72 @@
+"""Lyrics repository — adapter over ``AsyncSession``.
+
+One cached row per track (``track_id`` unique). ``upsert`` refreshes the row and
+bumps ``fetched_at`` so the service's TTL is measured from the last fetch.
+"""
+
+import uuid
+
+from sqlalchemy import func, select
+from sqlalchemy.dialects.postgresql import insert as pg_insert
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.domain.entities.lyrics import Lyrics
+from app.infrastructure.db.models.lyrics import LyricsModel
+
+
+def _to_entity(row: LyricsModel) -> Lyrics:
+    return Lyrics(
+        track_id=row.track_id,
+        synced=row.synced,
+        plain=row.plain,
+        source=row.source,
+        status=row.status,
+        fetched_at=row.fetched_at,
+    )
+
+
+class SqlAlchemyLyricsRepository:
+    def __init__(self, session: AsyncSession) -> None:
+        self._session = session
+
+    async def get(self, track_id: uuid.UUID) -> Lyrics | None:
+        row = await self._session.scalar(
+            select(LyricsModel).where(LyricsModel.track_id == track_id)
+        )
+        return _to_entity(row) if row is not None else None
+
+    async def upsert(
+        self,
+        *,
+        track_id: uuid.UUID,
+        synced: str | None,
+        plain: str | None,
+        source: str | None,
+        status: str,
+    ) -> Lyrics:
+        values = {
+            "track_id": track_id,
+            "synced": synced,
+            "plain": plain,
+            "source": source,
+            "status": status,
+            "fetched_at": func.now(),
+        }
+        stmt = (
+            pg_insert(LyricsModel)
+            .values(**values)
+            .on_conflict_do_update(
+                index_elements=[LyricsModel.track_id],
+                set_={
+                    "synced": synced,
+                    "plain": plain,
+                    "source": source,
+                    "status": status,
+                    "fetched_at": func.now(),
+                },
+            )
+            .returning(LyricsModel)
+        )
+        row = (await self._session.scalars(stmt)).one()
+        await self._session.flush()
+        return _to_entity(row)

app/infrastructure/db/repositories/playlist_repository.py

@@ -1,5 +1,6 @@
 """Playlist repository — adapter over ``AsyncSession``."""
 
+import datetime as dt
 import uuid
 
 from sqlalchemy import func, select
@@ -37,7 +38,11 @@ def _track_to_entity(row: TrackModel) -> Track:
         duration_seconds=row.duration_seconds,
         genre=row.genre,
         year=row.year,
+        track_number=row.track_number,
         metadata_status=row.metadata_status,
+        metadata_error=row.metadata_error,
+        enriched_at=row.enriched_at,
+        availability=row.availability,
         created_at=row.created_at,
         updated_at=row.updated_at,
     )
@@ -134,9 +139,22 @@ class SqlAlchemyPlaylistRepository:
     async def get_track_total(self, playlist_id: uuid.UUID) -> int:
         return await self.track_count(playlist_id)
 
+    async def has_track(self, playlist_id: uuid.UUID, track_id: uuid.UUID) -> bool:
+        row = (
+            await self._session.execute(
+                select(PlaylistTrackModel.id).where(
+                    PlaylistTrackModel.playlist_id == playlist_id,
+                    PlaylistTrackModel.track_id == track_id,
+                )
+            )
+        ).scalar_one_or_none()
+        return row is not None
+
     async def add_track(
         self, playlist_id: uuid.UUID, track_id: uuid.UUID, *, position: float
     ) -> None:
+        if await self.has_track(playlist_id, track_id):
+            return
         row = PlaylistTrackModel(playlist_id=playlist_id, track_id=track_id, position=position)
         self._session.add(row)
         playlist = await self._session.get(PlaylistModel, playlist_id)
@@ -144,6 +162,26 @@ class SqlAlchemyPlaylistRepository:
             playlist.version = playlist.version + 1
         await self._session.flush()
 
+    async def reorder_tracks(
+        self, playlist_id: uuid.UUID, ordered_track_ids: list[uuid.UUID]
+    ) -> None:
+        rows = (
+            (
+                await self._session.execute(
+                    select(PlaylistTrackModel).where(PlaylistTrackModel.playlist_id == playlist_id)
+                )
+            )
+            .scalars()
+            .all()
+        )
+        by_track_id = {row.track_id: row for row in rows}
+        for position, track_id in enumerate(ordered_track_ids, start=1):
+            by_track_id[track_id].position = float(position)
+        playlist = await self._session.get(PlaylistModel, playlist_id)
+        if playlist is not None:
+            playlist.version = playlist.version + 1
+        await self._session.flush()
+
     async def remove_track(self, playlist_id: uuid.UUID, track_id: uuid.UUID) -> None:
         row = (
             await self._session.execute(
@@ -170,6 +208,41 @@ class SqlAlchemyPlaylistRepository:
         ).scalar_one_or_none()
         return float(result) if result is not None else 0.0
 
+    async def get_cover_path(self, playlist_id: uuid.UUID) -> str | None:
+        """The playlist's stored cover key, or ``None`` (missing playlist or no
+        cover). Read directly — the entity doesn't carry the storage key."""
+        return (
+            await self._session.execute(
+                select(PlaylistModel.cover_path).where(PlaylistModel.id == playlist_id)
+            )
+        ).scalar_one_or_none()
+
+    async def list_changed_since(
+        self, *, owner_id: uuid.UUID, since: dt.datetime | None, until: dt.datetime
+    ) -> list[Playlist]:
+        """A user's playlists changed in the window ``(since, until]`` (metadata
+        or membership — every mutation bumps ``updated_at``/``version``). Defined
+        before ``list`` (name-shadowing)."""
+        stmt = select(PlaylistModel).where(
+            PlaylistModel.owner_id == owner_id, PlaylistModel.updated_at <= until
+        )
+        if since is not None:
+            stmt = stmt.where(PlaylistModel.updated_at > since)
+        stmt = stmt.order_by(PlaylistModel.updated_at)
+        rows = (await self._session.execute(stmt)).scalars().all()
+        return [_to_entity(r) for r in rows]
+
+    async def track_ids(self, playlist_id: uuid.UUID) -> list[uuid.UUID]:
+        """Ordered track ids of a playlist (by position) — for sync payloads."""
+        rows = (
+            await self._session.execute(
+                select(PlaylistTrackModel.track_id)
+                .where(PlaylistTrackModel.playlist_id == playlist_id)
+                .order_by(PlaylistTrackModel.position)
+            )
+        ).scalars().all()
+        return list(rows)
+
     # list must come after methods using list[...] in signatures (builtin name shadowing)
     async def list(self, *, owner_id: uuid.UUID, limit: int, offset: int) -> list[Playlist]:
         rows = (

app/infrastructure/db/repositories/track_repository.py

@@ -1,13 +1,16 @@
 """Track repository — adapter over ``AsyncSession``."""
 
+import datetime as dt
 import uuid
 
-from sqlalchemy import func, select
+from sqlalchemy import case, func, or_, select
 from sqlalchemy.ext.asyncio import AsyncSession
 
+from app.domain.entities.storage import FormatBreakdown, LibraryStats
 from app.domain.entities.track import Track
 from app.domain.errors import NotFoundError
 from app.infrastructure.db.models.artist import ArtistModel
+from app.infrastructure.db.models.enums import TrackAvailability
 from app.infrastructure.db.models.track import TrackModel
 
 
@@ -25,7 +28,11 @@ def _to_entity(row: TrackModel) -> Track:
         duration_seconds=row.duration_seconds,
         genre=row.genre,
         year=row.year,
+        track_number=row.track_number,
         metadata_status=row.metadata_status,
+        metadata_error=row.metadata_error,
+        enriched_at=row.enriched_at,
+        availability=row.availability,
         created_at=row.created_at,
         updated_at=row.updated_at,
     )
@@ -39,6 +46,16 @@ class SqlAlchemyTrackRepository:
         row = await self._session.get(TrackModel, track_id)
         return _to_entity(row) if row is not None else None
 
+    async def get_many(self, ids: list[uuid.UUID]) -> list[Track]:
+        if not ids:
+            return []
+        rows = (
+            (await self._session.execute(select(TrackModel).where(TrackModel.id.in_(ids))))
+            .scalars()
+            .all()
+        )
+        return [_to_entity(r) for r in rows]
+
     async def get_by_source(self, source: str, source_id: str) -> Track | None:
         row = (
             await self._session.execute(
@@ -56,13 +73,14 @@ class SqlAlchemyTrackRepository:
         id: uuid.UUID,
         title: str,
         artist_id: uuid.UUID,
-        storage_uri: str,
+        storage_uri: str | None,
-        file_format: str,
+        file_format: str | None,
-        file_size: int,
+        file_size: int | None,
         source: str,
         source_id: str,
         metadata_status: str,
         added_by: uuid.UUID | None,
+        availability: str = TrackAvailability.LOCAL.value,
     ) -> Track:
         row = TrackModel(
             id=id,
@@ -75,12 +93,38 @@ class SqlAlchemyTrackRepository:
             source_id=source_id,
             metadata_status=metadata_status,
             added_by=added_by,
+            availability=availability,
         )
         self._session.add(row)
         await self._session.flush()
         await self._session.refresh(row)
         return _to_entity(row)
 
+    async def materialize(
+        self,
+        track_id: uuid.UUID,
+        *,
+        storage_uri: str,
+        file_format: str,
+        file_size: int,
+        bitrate: int | None,
+    ) -> Track:
+        """Fill in a remote placeholder's audio fields after a download (lazy
+        materialization). ``track.id`` is unchanged, so likes/playlists/queue
+        entries that already reference it keep working."""
+        row = await self._session.get(TrackModel, track_id)
+        if row is None:
+            raise NotFoundError(f"Track {track_id} not found.")
+        row.storage_uri = storage_uri
+        row.file_format = file_format
+        row.file_size = file_size
+        if bitrate is not None:
+            row.bitrate = bitrate
+        row.availability = TrackAvailability.LOCAL.value
+        await self._session.flush()
+        await self._session.refresh(row)
+        return _to_entity(row)
+
     async def delete(self, track_id: uuid.UUID) -> None:
         row = await self._session.get(TrackModel, track_id)
         if row is not None:
@@ -102,12 +146,192 @@ class SqlAlchemyTrackRepository:
         ).all()
         return [(row.genre, row.cnt) for row in rows]
 
+    async def list_similar(
+        self,
+        *,
+        genre: str | None,
+        artist_id: uuid.UUID,
+        exclude_ids: list[uuid.UUID],
+        limit: int,
+    ) -> list[Track]:
+        # Rank a same-genre hit above a same-artist hit; shuffle within a tier so
+        # the mix varies. Only playable (locally-stored) tracks are candidates.
+        if genre is not None:
+            match = or_(TrackModel.genre == genre, TrackModel.artist_id == artist_id)
+            score = case((TrackModel.genre == genre, 2), else_=0) + case(
+                (TrackModel.artist_id == artist_id, 1), else_=0
+            )
+        else:
+            match = TrackModel.artist_id == artist_id
+            score = case((TrackModel.artist_id == artist_id, 1), else_=0)
+
+        stmt = select(TrackModel).where(TrackModel.storage_uri.is_not(None), match)
+        if exclude_ids:
+            stmt = stmt.where(TrackModel.id.not_in(exclude_ids))
+        stmt = stmt.order_by(score.desc(), func.random()).limit(limit)
+        rows = (await self._session.execute(stmt)).scalars().all()
+        return [_to_entity(r) for r in rows]
+
+    async def sample_playable(
+        self, *, exclude_ids: list[uuid.UUID], limit: int
+    ) -> list[Track]:
+        stmt = select(TrackModel).where(TrackModel.storage_uri.is_not(None))
+        if exclude_ids:
+            stmt = stmt.where(TrackModel.id.not_in(exclude_ids))
+        stmt = stmt.order_by(func.random()).limit(limit)
+        rows = (await self._session.execute(stmt)).scalars().all()
+        return [_to_entity(r) for r in rows]
+
+    async def library_stats(self) -> LibraryStats:
+        """One-shot aggregate over the whole catalogue (no pagination). Defined
+        before ``list`` for the same shadowing reason as ``genres``."""
+        totals = (
+            await self._session.execute(
+                select(
+                    func.count(TrackModel.id),
+                    func.coalesce(func.sum(TrackModel.file_size), 0),
+                    func.coalesce(func.sum(TrackModel.duration_seconds), 0),
+                    func.coalesce(func.max(TrackModel.file_size), 0),
+                    func.min(TrackModel.created_at),
+                    func.max(TrackModel.created_at),
+                )
+            )
+        ).one()
+
+        fmt_rows = (
+            await self._session.execute(
+                select(
+                    TrackModel.file_format,
+                    func.count(TrackModel.id),
+                    func.coalesce(func.sum(TrackModel.file_size), 0),
+                )
+                .where(TrackModel.file_format.is_not(None))
+                .group_by(TrackModel.file_format)
+                .order_by(func.sum(TrackModel.file_size).desc())
+            )
+        ).all()
+
+        status_rows = (
+            await self._session.execute(
+                select(TrackModel.metadata_status, func.count(TrackModel.id)).group_by(
+                    TrackModel.metadata_status
+                )
+            )
+        ).all()
+
+        source_rows = (
+            await self._session.execute(
+                select(TrackModel.source, func.count(TrackModel.id)).group_by(TrackModel.source)
+            )
+        ).all()
+
+        return LibraryStats(
+            total_tracks=totals[0],
+            total_size=totals[1],
+            total_duration_seconds=totals[2],
+            largest_track_size=totals[3],
+            earliest_added=totals[4],
+            latest_added=totals[5],
+            by_format=[
+                FormatBreakdown(file_format=fmt, track_count=cnt, total_size=size)
+                for fmt, cnt, size in fmt_rows
+            ],
+            by_metadata_status={status: cnt for status, cnt in status_rows},
+            by_source={source: cnt for source, cnt in source_rows},
+        )
+
+    async def find_duplicate_groups(self) -> list[tuple[str, list[Track]]]:
+        """Tracks that share an ``acoustid_fingerprint`` (the dedup key), grouped
+        by it — only fingerprints with more than one track. Empty when clean.
+        Defined before ``list`` for the same name-shadowing reason as ``genres``."""
+        dup_fps = (
+            select(TrackModel.acoustid_fingerprint)
+            .where(TrackModel.acoustid_fingerprint.is_not(None))
+            .group_by(TrackModel.acoustid_fingerprint)
+            .having(func.count(TrackModel.id) > 1)
+            .scalar_subquery()
+        )
+        rows = (
+            (
+                await self._session.execute(
+                    select(TrackModel)
+                    .where(TrackModel.acoustid_fingerprint.in_(dup_fps))
+                    .order_by(TrackModel.acoustid_fingerprint, TrackModel.created_at)
+                )
+            )
+            .scalars()
+            .all()
+        )
+        groups: dict[str, list[Track]] = {}
+        for row in rows:
+            fingerprint = row.acoustid_fingerprint
+            assert fingerprint is not None  # filtered to non-null above
+            groups.setdefault(fingerprint, []).append(_to_entity(row))
+        return list(groups.items())
+
+    async def list_by_metadata_status(
+        self, status: str, *, limit: int, offset: int
+    ) -> list[Track]:
+        """Tracks in a given ``metadata_status`` (e.g. ``pending``/``failed``),
+        newest first. Defined before ``list`` (name-shadowing)."""
+        rows = (
+            (
+                await self._session.execute(
+                    select(TrackModel)
+                    .where(TrackModel.metadata_status == status)
+                    .order_by(TrackModel.created_at.desc())
+                    .limit(limit)
+                    .offset(offset)
+                )
+            )
+            .scalars()
+            .all()
+        )
+        return [_to_entity(r) for r in rows]
+
+    async def all_storage_refs(self) -> list[tuple[uuid.UUID, str]]:
+        """``(id, storage_uri)`` for every *local* track — for the cleanup
+        worker's filesystem reconciliation. Remote placeholders have no local
+        file (``availability != local``) and are skipped. No entity hydration."""
+        rows = (
+            await self._session.execute(
+                select(TrackModel.id, TrackModel.storage_uri).where(
+                    TrackModel.availability == TrackAvailability.LOCAL.value,
+                    TrackModel.storage_uri.is_not(None),
+                )
+            )
+        ).all()
+        return [(row.id, row.storage_uri) for row in rows]
+
+    async def count_by_metadata_status(self, status: str) -> int:
+        return (
+            await self._session.execute(
+                select(func.count())
+                .select_from(TrackModel)
+                .where(TrackModel.metadata_status == status)
+            )
+        ).scalar_one()
+
+    async def list_changed_since(
+        self, *, since: dt.datetime | None, until: dt.datetime
+    ) -> list[Track]:
+        """Catalogue tracks changed in the window ``(since, until]`` (by
+        ``updated_at``), oldest first — the delta a client caches for offline use.
+        Defined before ``list`` (name-shadowing)."""
+        stmt = select(TrackModel).where(TrackModel.updated_at <= until)
+        if since is not None:
+            stmt = stmt.where(TrackModel.updated_at > since)
+        stmt = stmt.order_by(TrackModel.updated_at)
+        rows = (await self._session.execute(stmt)).scalars().all()
+        return [_to_entity(r) for r in rows]
+
     async def list(
         self,
         *,
         artist_id: uuid.UUID | None,
         album_id: uuid.UUID | None,
         q: str | None,
+        source: str | None = None,
         sort_by: str = "created_at",
         order: str = "desc",
         limit: int = 50,
@@ -118,6 +342,8 @@ class SqlAlchemyTrackRepository:
             stmt = stmt.where(TrackModel.artist_id == artist_id)
         if album_id is not None:
             stmt = stmt.where(TrackModel.album_id == album_id)
+        if source is not None:
+            stmt = stmt.where(TrackModel.source == source)
         if q:
             stmt = stmt.where(TrackModel.title.ilike(f"%{q}%"))
 
@@ -142,12 +368,15 @@ class SqlAlchemyTrackRepository:
         artist_id: uuid.UUID | None,
         album_id: uuid.UUID | None,
         q: str | None,
+        source: str | None = None,
     ) -> int:
         stmt = select(func.count()).select_from(TrackModel)
         if artist_id is not None:
             stmt = stmt.where(TrackModel.artist_id == artist_id)
         if album_id is not None:
             stmt = stmt.where(TrackModel.album_id == album_id)
+        if source is not None:
+            stmt = stmt.where(TrackModel.source == source)
         if q:
             stmt = stmt.where(TrackModel.title.ilike(f"%{q}%"))
         return (await self._session.execute(stmt)).scalar_one()
@@ -159,6 +388,9 @@ class SqlAlchemyTrackRepository:
         title: str | None,
         genre: str | None,
         year: int | None,
+        artist_id: uuid.UUID | None = None,
+        album_id: uuid.UUID | None = None,
+        track_number: int | None = None,
     ) -> Track:
         row = await self._session.get(TrackModel, track_id)
         if row is None:
@@ -169,7 +401,75 @@ class SqlAlchemyTrackRepository:
             row.genre = genre
         if year is not None:
             row.year = year
+        if artist_id is not None:
+            row.artist_id = artist_id
+        if album_id is not None:
+            row.album_id = album_id
+        if track_number is not None:
+            row.track_number = track_number
         row.metadata_status = "manual"
         await self._session.flush()
         await self._session.refresh(row)
         return _to_entity(row)
+
+    async def apply_enrichment(
+        self,
+        track_id: uuid.UUID,
+        *,
+        title: str,
+        artist_id: uuid.UUID,
+        album_id: uuid.UUID | None,
+        genre: str | None,
+        year: int | None,
+        track_number: int | None,
+        duration_seconds: int | None,
+        bitrate: int | None,
+        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:
+            raise NotFoundError(f"Track {track_id} not found.")
+        # Identity + status are authoritative for an enrichment run.
+        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
+        if genre is not None:
+            row.genre = genre
+        if year is not None:
+            row.year = year
+        if track_number is not None:
+            row.track_number = track_number
+        if duration_seconds is not None:
+            row.duration_seconds = duration_seconds
+        if bitrate is not None:
+            row.bitrate = bitrate
+        if acoustid_fingerprint is not None:
+            row.acoustid_fingerprint = acoustid_fingerprint
+        if musicbrainz_id is not None:
+            row.musicbrainz_id = musicbrainz_id
+        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/infrastructure/db/repositories/user_settings_repository.py

@@ -0,0 +1,47 @@
+"""User-settings repository — adapter over ``AsyncSession``."""
+
+import uuid
+
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.domain.entities.settings import UserSettings
+from app.infrastructure.db.models.user_settings import UserSettingsModel
+
+
+def _to_entity(row: UserSettingsModel) -> UserSettings:
+    return UserSettings(
+        user_id=row.user_id,
+        theme=row.theme,
+        stream_quality=row.stream_quality,
+        scrobble_enabled=row.scrobble_enabled,
+        scrobble_provider=row.scrobble_provider,
+        scrobble_username=row.scrobble_username,
+        scrobble_session_key_enc=row.scrobble_session_key_enc,
+    )
+
+
+class SqlAlchemyUserSettingsRepository:
+    def __init__(self, session: AsyncSession) -> None:
+        self._session = session
+
+    async def get(self, user_id: uuid.UUID) -> UserSettings | None:
+        row = await self._session.get(UserSettingsModel, user_id)
+        return _to_entity(row) if row is not None else None
+
+    async def upsert(self, settings: UserSettings) -> UserSettings:
+        """Create or replace the caller's settings row with ``settings`` in full.
+        The service merges partial updates against current values before calling
+        this, so the write always carries the complete desired state."""
+        row = await self._session.get(UserSettingsModel, settings.user_id)
+        if row is None:
+            row = UserSettingsModel(user_id=settings.user_id)
+            self._session.add(row)
+        row.theme = settings.theme
+        row.stream_quality = settings.stream_quality
+        row.scrobble_enabled = settings.scrobble_enabled
+        row.scrobble_provider = settings.scrobble_provider
+        row.scrobble_username = settings.scrobble_username
+        row.scrobble_session_key_enc = settings.scrobble_session_key_enc
+        await self._session.flush()
+        await self._session.refresh(row)
+        return _to_entity(row)

app/infrastructure/metadata/__init__.py

@@ -0,0 +1 @@
+"""Metadata-enrichment adapters: tag reader, fingerprinter, AcoustID client."""

app/infrastructure/metadata/acoustid.py

@@ -0,0 +1,161 @@
+"""AcoustIdHttpClient — identifies a recording from its Chromaprint fingerprint.
+
+One ``/v2/lookup`` call with ``meta=recordings+releasegroups`` returns the
+AcoustID id, the MusicBrainz recording id, and canonical title/artist/album —
+metadata that itself originates from MusicBrainz, so a separate MB call is not
+needed for Phase 1 (plan §6.2 steps 2-3 collapsed into one request).
+
+Graceful degradation: no API key → ``is_available()`` is False and the whole
+fingerprint path is skipped; any network/parse error → ``lookup`` returns
+``None``. A small inter-call delay keeps us within AcoustID's rate limit.
+"""
+
+import asyncio
+import time
+
+import httpx
+
+from app.core.logging import get_logger
+from app.domain.entities.metadata import Fingerprint, RecordingMatch
+
+log = get_logger(__name__)
+
+_DEFAULT_URL = "https://api.acoustid.org/v2/lookup"
+_TIMEOUT_SECONDS = 10.0
+_MIN_INTERVAL_SECONDS = 0.34  # AcoustID allows ~3 req/s; stay polite
+
+
+class AcoustIdHttpClient:
+    """Implements :class:`app.domain.ports.AcoustIdClient`."""
+
+    _throttle_lock = asyncio.Lock()
+    _last_call_monotonic = 0.0
+
+    def __init__(
+        self,
+        *,
+        api_key: str | None,
+        user_agent: str,
+        api_url: str = _DEFAULT_URL,
+    ) -> None:
+        self._api_key = api_key
+        self._user_agent = user_agent
+        self._api_url = api_url
+
+    def is_available(self) -> bool:
+        return bool(self._api_key)
+
+    async def lookup(self, fingerprint: Fingerprint) -> RecordingMatch | None:
+        payload = await self._lookup_raw(fingerprint)
+        if payload is None:
+            return None
+        return _parse_best_match(payload)
+
+    async def lookup_all(self, fingerprint: Fingerprint) -> list[RecordingMatch]:
+        payload = await self._lookup_raw(fingerprint)
+        if payload is None:
+            return []
+        return _parse_matches(payload)
+
+    async def _lookup_raw(self, fingerprint: Fingerprint) -> object | None:
+        if not self._api_key:
+            return None
+        try:
+            await self._throttle()
+            async with httpx.AsyncClient(
+                timeout=_TIMEOUT_SECONDS,
+                headers={"User-Agent": self._user_agent},
+            ) as client:
+                resp = await client.get(
+                    self._api_url,
+                    params={
+                        "client": self._api_key,
+                        "duration": str(fingerprint.duration_seconds),
+                        "fingerprint": fingerprint.fingerprint,
+                        "meta": "recordings releasegroups",
+                        "format": "json",
+                    },
+                )
+            resp.raise_for_status()
+            return resp.json()  # type: ignore[no-any-return]
+        except httpx.HTTPError, ValueError:
+            log.warning("acoustid_lookup_failed")
+            return None
+
+    @classmethod
+    async def _throttle(cls) -> None:
+        async with cls._throttle_lock:
+            elapsed = time.monotonic() - cls._last_call_monotonic
+            wait = _MIN_INTERVAL_SECONDS - elapsed
+            if wait > 0:
+                await asyncio.sleep(wait)
+            cls._last_call_monotonic = time.monotonic()
+
+
+_MAX_MATCHES = 5
+
+
+def _parse_best_match(payload: object) -> RecordingMatch | None:
+    matches = _parse_matches(payload)
+    return matches[0] if matches else None
+
+
+def _parse_matches(payload: object) -> list[RecordingMatch]:
+    if not isinstance(payload, dict) or payload.get("status") != "ok":
+        return []
+    results = payload.get("results")
+    if not isinstance(results, list) or not results:
+        return []
+
+    # Results are returned best-score-first, but sort defensively and cap the
+    # number of candidates surfaced to the editor.
+    candidates = [r for r in results if isinstance(r, dict)]
+    candidates.sort(key=lambda r: r.get("score", 0.0), reverse=True)
+
+    matches: list[RecordingMatch] = []
+    for result in candidates[:_MAX_MATCHES]:
+        match = _parse_one(result)
+        if match is not None:
+            matches.append(match)
+    return matches
+
+
+def _parse_one(result: dict[str, object]) -> RecordingMatch | None:
+    acoustid = result.get("id")
+    if not isinstance(acoustid, str):
+        return None
+    score = float(result.get("score", 0.0))  # type: ignore[arg-type]
+
+    recording_mbid: str | None = None
+    release_group_mbid: str | None = None
+    title: str | None = None
+    artist: str | None = None
+    album: str | None = None
+
+    recordings = result.get("recordings")
+    if isinstance(recordings, list) and recordings and isinstance(recordings[0], dict):
+        rec = recordings[0]
+        recording_mbid = rec.get("id") if isinstance(rec.get("id"), str) else None
+        title = rec.get("title") if isinstance(rec.get("title"), str) else None
+        artists = rec.get("artists")
+        if isinstance(artists, list) and artists and isinstance(artists[0], dict):
+            name = artists[0].get("name")
+            artist = name if isinstance(name, str) else None
+        groups = rec.get("releasegroups")
+        if isinstance(groups, list) and groups and isinstance(groups[0], dict):
+            group = groups[0]
+            gtitle = group.get("title")
+            album = gtitle if isinstance(gtitle, str) else None
+            gid = group.get("id")
+            release_group_mbid = gid if isinstance(gid, str) else None
+
+    return RecordingMatch(
+        acoustid=acoustid,
+        score=score,
+        recording_mbid=recording_mbid,
+        release_group_mbid=release_group_mbid,
+        title=title,
+        artist=artist,
+        album=album,
+        year=None,
+    )

app/infrastructure/metadata/cover_extractor.py

@@ -0,0 +1,111 @@
+"""MutagenCoverExtractor — pulls embedded cover art from a local audio file.
+
+The offline-first cover source (mirrors the tag pre-pass): a well-tagged file
+often already carries front-cover artwork (ID3 ``APIC``, FLAC/OGG picture
+blocks, MP4 ``covr``). We read it without any network call. Parsing is blocking,
+so it runs in a worker thread. Any failure degrades to ``None`` — never raises.
+
+mutagen ships no type stubs, so its objects are handled as ``Any`` and accessed
+defensively (``getattr``) — the format zoo doesn't fit one static shape anyway.
+"""
+
+import base64
+from pathlib import Path
+from typing import Any
+
+import anyio
+from mutagen import File as MutagenFile  # type: ignore[attr-defined]
+from mutagen.flac import Picture
+from mutagen.mp4 import MP4Cover
+
+from app.core.logging import get_logger
+from app.domain.entities.cover import CoverArt
+
+log = get_logger(__name__)
+
+# MP4 cover format flag → MIME (mutagen exposes an int, not a content type).
+_MP4_FORMATS: dict[int, str] = {
+    MP4Cover.FORMAT_JPEG: "image/jpeg",
+    MP4Cover.FORMAT_PNG: "image/png",
+}
+_FRONT_COVER = 3  # APIC/Picture "type" value for the front cover
+
+
+class MutagenCoverExtractor:
+    """Implements :class:`app.domain.ports.CoverArtExtractor`."""
+
+    async def extract(self, path: Path) -> CoverArt | None:
+        try:
+            return await anyio.to_thread.run_sync(self._extract_sync, path)
+        except Exception:
+            log.warning("cover_extract_failed", path=str(path))
+            return None
+
+    def _extract_sync(self, path: Path) -> CoverArt | None:
+        audio: Any = MutagenFile(str(path))
+        if audio is None:
+            return None
+
+        # FLAC / OGG-FLAC: typed picture blocks on the file object.
+        pictures = getattr(audio, "pictures", None)
+        if pictures:
+            cover = _from_picture(_front_or_first(pictures))
+            if cover is not None:
+                return cover
+
+        tags = audio.tags
+        if tags is None:
+            return None
+
+        # MP3 / anything with ID3 frames: APIC frames keyed as "APIC:...".
+        apics = [frame for frame in tags.values() if frame.__class__.__name__ == "APIC"]
+        if apics:
+            cover = _from_picture(_front_or_first(apics))
+            if cover is not None:
+                return cover
+
+        get = getattr(tags, "get", None)
+        if get is None:
+            return None
+
+        # MP4 / M4A: "covr" atom holds a list of MP4Cover (a bytes subclass).
+        covr = get("covr")
+        if covr:
+            mp4_cover = covr[0]
+            content_type = _MP4_FORMATS.get(getattr(mp4_cover, "imageformat", -1), "image/jpeg")
+            return CoverArt(data=bytes(mp4_cover), content_type=content_type)
+
+        # OGG Vorbis: base64 picture block in METADATA_BLOCK_PICTURE.
+        block = get("metadata_block_picture")
+        if block:
+            cover = _from_picture(_decode_vorbis_picture(block[0]))
+            if cover is not None:
+                return cover
+
+        return None
+
+
+def _from_picture(picture: Any) -> CoverArt | None:
+    """Build a :class:`CoverArt` from a mutagen picture/APIC frame, or ``None``."""
+    if picture is None:
+        return None
+    data = getattr(picture, "data", None)
+    if not data:
+        return None
+    mime = getattr(picture, "mime", None) or "image/jpeg"
+    return CoverArt(data=bytes(data), content_type=str(mime))
+
+
+def _front_or_first(pictures: list[Any]) -> Any:
+    """Prefer the front-cover picture (type 3), else the first available."""
+    for pic in pictures:
+        if getattr(pic, "type", None) == _FRONT_COVER:
+            return pic
+    return pictures[0] if pictures else None
+
+
+def _decode_vorbis_picture(encoded: str) -> Any:
+    try:
+        return Picture(base64.b64decode(encoded))  # type: ignore[no-untyped-call]
+    except Exception:
+        return None

app/infrastructure/metadata/coverart.py

@@ -0,0 +1,83 @@
+"""CoverArtArchiveClient — fetches front cover art from the Cover Art Archive.
+
+The network fallback when a file carries no embedded artwork: given a
+MusicBrainz **release-group** id (supplied by the AcoustID lookup), request the
+front image from ``coverartarchive.org``. The CAA redirects to the Internet
+Archive, so redirects are followed. ``thumbnail`` 500px keeps payloads small.
+
+Graceful degradation (CLAUDE.md): no release-group id → never called; any
+network/HTTP error (incl. 404 "no cover") → returns ``None``, never raises. A
+small inter-call delay respects the shared MusicBrainz/CAA infrastructure.
+"""
+
+import asyncio
+import time
+
+import httpx
+
+from app.core.logging import get_logger
+from app.domain.entities.cover import CoverArt
+
+log = get_logger(__name__)
+
+_DEFAULT_BASE_URL = "https://coverartarchive.org"
+_TIMEOUT_SECONDS = 15.0
+_MIN_INTERVAL_SECONDS = 1.0  # CAA piggybacks on MusicBrainz infra; stay polite
+_MAX_BYTES = 10 * 1024 * 1024  # ignore absurdly large images
+
+
+class CoverArtArchiveClient:
+    """Implements :class:`app.domain.ports.CoverArtProvider`."""
+
+    _throttle_lock = asyncio.Lock()
+    _last_call_monotonic = 0.0
+
+    def __init__(
+        self,
+        *,
+        user_agent: str,
+        enabled: bool = True,
+        base_url: str = _DEFAULT_BASE_URL,
+    ) -> None:
+        self._user_agent = user_agent
+        self._enabled = enabled
+        self._base_url = base_url.rstrip("/")
+
+    def is_available(self) -> bool:
+        return self._enabled
+
+    async def fetch_release_group(self, release_group_mbid: str) -> CoverArt | None:
+        if not self._enabled or not release_group_mbid:
+            return None
+        url = f"{self._base_url}/release-group/{release_group_mbid}/front-500"
+        try:
+            await self._throttle()
+            async with httpx.AsyncClient(
+                timeout=_TIMEOUT_SECONDS,
+                follow_redirects=True,
+                headers={"User-Agent": self._user_agent},
+            ) as client:
+                resp = await client.get(url)
+            if resp.status_code == 404:
+                return None  # no cover for this release group — normal, not an error
+            resp.raise_for_status()
+        except httpx.HTTPError:
+            log.warning("coverart_fetch_failed", release_group=release_group_mbid)
+            return None
+
+        data = resp.content
+        if not data or len(data) > _MAX_BYTES:
+            return None
+        content_type = resp.headers.get("content-type", "image/jpeg").split(";")[0].strip()
+        if not content_type.startswith("image/"):
+            return None
+        return CoverArt(data=data, content_type=content_type)
+
+    @classmethod
+    async def _throttle(cls) -> None:
+        async with cls._throttle_lock:
+            elapsed = time.monotonic() - cls._last_call_monotonic
+            wait = _MIN_INTERVAL_SECONDS - elapsed
+            if wait > 0:
+                await asyncio.sleep(wait)
+            cls._last_call_monotonic = time.monotonic()

app/infrastructure/metadata/fingerprint.py

@@ -0,0 +1,62 @@
+"""FpcalcFingerprinter — Chromaprint fingerprint via the ``fpcalc`` binary.
+
+``fpcalc -json <file>`` emits ``{"duration": float, "fingerprint": str}``. The
+binary ships in the Docker image (``libchromaprint-tools``). Any failure (binary
+missing, bad file, timeout) degrades to ``None`` — the pipeline then falls back
+to tag-only metadata (plan §6.2: one external dependency must never crash it).
+"""
+
+import asyncio
+import json
+import shutil
+from pathlib import Path
+
+from app.core.logging import get_logger
+from app.domain.entities.metadata import Fingerprint
+
+log = get_logger(__name__)
+
+_TIMEOUT_SECONDS = 30
+
+
+class FpcalcFingerprinter:
+    """Implements :class:`app.domain.ports.AudioFingerprinter`."""
+
+    def __init__(self, binary: str = "fpcalc") -> None:
+        self._binary = binary
+
+    def is_available(self) -> bool:
+        return shutil.which(self._binary) is not None
+
+    async def calculate(self, path: Path) -> Fingerprint | None:
+        if not self.is_available():
+            return None
+        try:
+            proc = await asyncio.create_subprocess_exec(
+                self._binary,
+                "-json",
+                str(path),
+                stdout=asyncio.subprocess.PIPE,
+                stderr=asyncio.subprocess.PIPE,
+            )
+            async with asyncio.timeout(_TIMEOUT_SECONDS):
+                stdout, _stderr = await proc.communicate()
+        except TimeoutError, OSError:
+            log.warning("fpcalc_failed", path=str(path))
+            return None
+
+        if proc.returncode != 0:
+            log.warning("fpcalc_nonzero", path=str(path), returncode=proc.returncode)
+            return None
+
+        try:
+            data = json.loads(stdout)
+            fingerprint = str(data["fingerprint"])
+            duration = round(float(data["duration"]))
+        except json.JSONDecodeError, KeyError, ValueError:
+            log.warning("fpcalc_bad_output", path=str(path))
+            return None
+
+        if not fingerprint or duration <= 0:
+            return None
+        return Fingerprint(fingerprint=fingerprint, duration_seconds=duration)

app/infrastructure/metadata/lrclib.py

@@ -0,0 +1,100 @@
+"""LrclibHttpClient — fetches lyrics from LRCLIB (plan §6.7).
+
+LRCLIB is a free, keyless lyrics database. ``/api/get`` does an exact match on
+artist+track+album+duration; if that misses we fall back to ``/api/search`` and
+take the best-scoring hit. Graceful degradation: any network/parse error →
+``fetch`` returns ``None`` (the service then caches a ``not_found``), never
+raising. No API key is needed, so this provider is always "available".
+"""
+
+import httpx
+
+from app.core.logging import get_logger
+from app.domain.entities.lyrics import LyricsResult
+
+log = get_logger(__name__)
+
+_BASE_URL = "https://lrclib.net"
+_TIMEOUT_SECONDS = 10.0
+_SOURCE = "lrclib"
+
+
+class LrclibHttpClient:
+    """Implements :class:`app.domain.ports.LyricsProvider`."""
+
+    def __init__(self, *, user_agent: str, base_url: str = _BASE_URL) -> None:
+        self._user_agent = user_agent
+        self._base_url = base_url.rstrip("/")
+
+    async def fetch(
+        self,
+        *,
+        artist: str,
+        title: str,
+        album: str | None,
+        duration_seconds: int | None,
+    ) -> LyricsResult | None:
+        try:
+            async with httpx.AsyncClient(
+                timeout=_TIMEOUT_SECONDS,
+                headers={"User-Agent": self._user_agent},
+                base_url=self._base_url,
+            ) as client:
+                hit = await self._get(client, artist, title, album, duration_seconds)
+                if hit is None:
+                    hit = await self._search(client, artist, title)
+        except (httpx.HTTPError, ValueError) as exc:
+            log.warning("lrclib.fetch_failed", error=str(exc))
+            return None
+        return hit
+
+    async def _get(
+        self,
+        client: httpx.AsyncClient,
+        artist: str,
+        title: str,
+        album: str | None,
+        duration_seconds: int | None,
+    ) -> LyricsResult | None:
+        """Exact match via ``/api/get`` (404 when nothing matches exactly)."""
+        params = {"artist_name": artist, "track_name": title}
+        if album:
+            params["album_name"] = album
+        if duration_seconds is not None:
+            params["duration"] = str(duration_seconds)
+        resp = await client.get("/api/get", params=params)
+        if resp.status_code == httpx.codes.NOT_FOUND:
+            return None
+        resp.raise_for_status()
+        return _to_result(resp.json())
+
+    async def _search(
+        self, client: httpx.AsyncClient, artist: str, title: str
+    ) -> LyricsResult | None:
+        """Fuzzy fallback via ``/api/search`` — take the first usable hit."""
+        resp = await client.get(
+            "/api/search", params={"artist_name": artist, "track_name": title}
+        )
+        resp.raise_for_status()
+        results = resp.json()
+        if not isinstance(results, list):
+            return None
+        for item in results:
+            result = _to_result(item)
+            if result is not None:
+                return result
+        return None
+
+
+def _to_result(payload: object) -> LyricsResult | None:
+    """Map an LRCLIB record to a ``LyricsResult``. Instrumental tracks and empty
+    records yield ``None`` (nothing worth caching as "found")."""
+    if not isinstance(payload, dict):
+        return None
+    if payload.get("instrumental"):
+        return None
+    synced = payload.get("syncedLyrics") or None
+    plain = payload.get("plainLyrics") or None
+    if synced is None and plain is None:
+        return None
+    return LyricsResult(synced=synced, plain=plain, source=_SOURCE)

app/infrastructure/metadata/tags.py

@@ -0,0 +1,88 @@
+"""MutagenTagReader — reads embedded tags from a local audio file.
+
+The offline first pass of enrichment (plan §6.2): well-tagged files get correct
+artist/album/title without any network call. mutagen's ``easy=True`` mode
+normalises tag keys across ID3 / Vorbis / MP4, so one code path covers all the
+formats the library accepts. Parsing is blocking, so it runs in a worker thread.
+"""
+
+import re
+from pathlib import Path
+
+import anyio
+from mutagen import File as MutagenFile  # type: ignore[attr-defined]
+
+from app.core.logging import get_logger
+from app.domain.entities.metadata import AudioTags
+
+log = get_logger(__name__)
+
+_YEAR_RE = re.compile(r"(\d{4})")
+
+
+def _first(value: object) -> str | None:
+    """EasyXxx tags expose values as lists; take the first non-empty string."""
+    if isinstance(value, list):
+        value = value[0] if value else None
+    if value is None:
+        return None
+    text = str(value).strip()
+    return text or None
+
+
+def _parse_year(value: object) -> int | None:
+    text = _first(value)
+    if text is None:
+        return None
+    m = _YEAR_RE.search(text)
+    return int(m.group(1)) if m else None
+
+
+def _parse_track_number(value: object) -> int | None:
+    text = _first(value)
+    if text is None:
+        return None
+    # "3" or "3/12" → 3
+    head = text.split("/", 1)[0].strip()
+    return int(head) if head.isdigit() else None
+
+
+class MutagenTagReader:
+    """Implements :class:`app.domain.ports.AudioTagReader`."""
+
+    async def read(self, path: Path) -> AudioTags | None:
+        try:
+            return await anyio.to_thread.run_sync(self._read_sync, path)
+        except Exception:
+            log.warning("tag_read_failed", path=str(path))
+            return None
+
+    def _read_sync(self, path: Path) -> AudioTags | None:
+        audio = MutagenFile(str(path), easy=True)
+        if audio is None:
+            return None  # unrecognised container
+
+        tags = audio.tags or {}
+        info = getattr(audio, "info", None)
+
+        duration = None
+        bitrate = None
+        if info is not None:
+            length = getattr(info, "length", None)
+            if length:
+                duration = round(float(length))
+            raw_bitrate = getattr(info, "bitrate", None)
+            if raw_bitrate:
+                bitrate = int(raw_bitrate) // 1000  # bits/s → kbps for display
+
+        return AudioTags(
+            title=_first(tags.get("title")),
+            artist=_first(tags.get("artist")),
+            album=_first(tags.get("album")),
+            album_artist=_first(tags.get("albumartist")),
+            genre=_first(tags.get("genre")),
+            year=_parse_year(tags.get("date") or tags.get("year")),
+            track_number=_parse_track_number(tags.get("tracknumber")),
+            duration_seconds=duration,
+            bitrate=bitrate,
+        )

app/infrastructure/ml/__init__.py

@@ -0,0 +1 @@
+"""ML/recommender adapters (plan §6.5). ML is optional — see Recommender port."""

app/infrastructure/ml/recommender.py

@@ -0,0 +1,110 @@
+"""Recommender adapters (plan §6.5).
+
+``NullRecommender`` is the default: no ML service, so every method reports
+unavailable and the ``RecommendationService`` uses its metadata fallback. When
+an embedding service exists, wire ``RemoteRecommender`` (skeleton below) to
+``ML_SERVICE_URL`` — its exact request/response contract is TODO pending that
+service. Both keep the invariant: ML is optional, never a hard dependency.
+"""
+
+import uuid
+
+import httpx
+
+from app.core.logging import get_logger
+
+log = get_logger(__name__)
+
+
+class NullRecommender:
+    """No ML configured — always unavailable, always ``None`` (→ fallback)."""
+
+    def is_available(self) -> bool:
+        return False
+
+    async def similar_track_ids(
+        self, track_id: uuid.UUID, *, limit: int, exclude_ids: list[uuid.UUID]
+    ) -> list[uuid.UUID] | None:
+        return None
+
+    async def similar_artist_ids(
+        self, artist_id: uuid.UUID, *, limit: int
+    ) -> list[uuid.UUID] | None:
+        return None
+
+    async def radio_track_ids(
+        self,
+        *,
+        seed_track_id: uuid.UUID | None,
+        exploration: float,
+        limit: int,
+        exclude_ids: list[uuid.UUID],
+    ) -> list[uuid.UUID] | None:
+        return None
+
+
+_TIMEOUT_SECONDS = 5.0
+
+
+class RemoteRecommender:
+    """HTTP client for an external embedding/recommender service.
+
+    TODO: the request/response schema below is a placeholder — align it with the
+    real ML service once its contract is known. Until then this stays unused
+    (``deps`` wires ``NullRecommender``). Every call is defensive: any error or a
+    malformed body returns ``None`` so the service degrades to metadata, matching
+    the graceful-degradation invariant.
+    """
+
+    def __init__(self, base_url: str) -> None:
+        self._base_url = base_url.rstrip("/")
+
+    def is_available(self) -> bool:
+        return True
+
+    async def _post_ids(self, path: str, payload: dict[str, object]) -> list[uuid.UUID] | None:
+        try:
+            async with httpx.AsyncClient(timeout=_TIMEOUT_SECONDS) as client:
+                resp = await client.post(f"{self._base_url}{path}", json=payload)
+                resp.raise_for_status()
+                data = resp.json()
+            ids = data.get("track_ids") if isinstance(data, dict) else None
+            if not isinstance(ids, list):
+                return None
+            return [uuid.UUID(str(i)) for i in ids]
+        except (httpx.HTTPError, ValueError, KeyError) as exc:
+            log.warning("recommender.remote_failed", path=path, error=str(exc))
+            return None
+
+    async def similar_track_ids(
+        self, track_id: uuid.UUID, *, limit: int, exclude_ids: list[uuid.UUID]
+    ) -> list[uuid.UUID] | None:
+        return await self._post_ids(
+            "/similar/tracks",
+            {"track_id": str(track_id), "limit": limit,
+             "exclude": [str(i) for i in exclude_ids]},
+        )
+
+    async def similar_artist_ids(
+        self, artist_id: uuid.UUID, *, limit: int
+    ) -> list[uuid.UUID] | None:
+        # Artist recommendations aren't part of the placeholder track contract.
+        return None
+
+    async def radio_track_ids(
+        self,
+        *,
+        seed_track_id: uuid.UUID | None,
+        exploration: float,
+        limit: int,
+        exclude_ids: list[uuid.UUID],
+    ) -> list[uuid.UUID] | None:
+        return await self._post_ids(
+            "/radio",
+            {
+                "seed_track_id": str(seed_track_id) if seed_track_id else None,
+                "exploration": exploration,
+                "limit": limit,
+                "exclude": [str(i) for i in exclude_ids],
+            },
+        )

app/infrastructure/sources/__init__.py

@@ -0,0 +1 @@
+"""Source backends — driven adapters that discover/fetch tracks."""

app/infrastructure/sources/local_folder.py

@@ -0,0 +1,60 @@
+"""``local`` source — indexes audio files from a mounted folder.
+
+Walks a configured root directory and yields each audio file as a
+:class:`SourceFile`. It does **not** parse tags or resolve artist/album — that's
+enrichment's job (plan §6.2); this stays a thin discovery layer. ``source_id``
+is the path relative to the root, so re-scans are idempotent.
+"""
+
+import os
+from collections.abc import Iterator
+from pathlib import Path
+
+from app.domain.sources import SourceFile, SourceInfo
+from app.infrastructure.db.models.enums import TrackSource
+
+# Extensions we treat as audio. Mirrors the formats StreamingService serves.
+_AUDIO_EXTENSIONS = frozenset(
+    {"mp3", "flac", "m4a", "aac", "ogg", "opus", "wav", "wma", "aiff", "aif", "alac"}
+)
+
+
+class LocalFolderSource:
+    """Implements :class:`app.domain.ports.IndexableSource`."""
+
+    name = TrackSource.LOCAL.value
+
+    def __init__(self, root: Path) -> None:
+        self._root = root
+
+    def info(self) -> SourceInfo:
+        return SourceInfo(
+            name=self.name,
+            label="Local folder",
+            kind="indexable",
+            available=self.is_available(),
+        )
+
+    def is_available(self) -> bool:
+        return self._root.is_dir()
+
+    def scan(self) -> Iterator[SourceFile]:
+        if not self.is_available():
+            return
+        for dirpath, _dirnames, filenames in os.walk(self._root):
+            for filename in sorted(filenames):
+                ext = Path(filename).suffix.lower().lstrip(".")
+                if ext not in _AUDIO_EXTENSIONS:
+                    continue
+                path = Path(dirpath) / filename
+                try:
+                    size = path.stat().st_size
+                except OSError:
+                    continue  # vanished/unreadable between walk and stat → skip
+                yield SourceFile(
+                    source_id=path.relative_to(self._root).as_posix(),
+                    path=path,
+                    suggested_title=path.stem or "Unknown",
+                    file_format=ext,
+                    file_size=size,
+                )

app/infrastructure/sources/registry.py

@@ -0,0 +1,70 @@
+"""Source registry — selection + enumeration of configured backends.
+
+Built from settings at the composition root. Only sources that are configured
+are registered (e.g. ``local`` appears only when ``LOCAL_MEDIA_IMPORT_PATH`` is
+set; ``youtube`` only when ``YOUTUBE_ENABLED``), so enumeration reflects what the
+instance can actually use.
+"""
+
+from typing import cast
+
+from app.core.config import Settings
+from app.domain.errors import NotFoundError, ValidationError
+from app.domain.ports import FetchableSource, IndexableSource, SearchableSource, SourceBackend
+from app.domain.sources import SourceInfo
+from app.infrastructure.sources.local_folder import LocalFolderSource
+from app.infrastructure.sources.youtube import YouTubeMusicSource
+
+
+class SourceRegistry:
+    def __init__(self, backends: list[SourceBackend]) -> None:
+        self._by_name = {backend.name: backend for backend in backends}
+
+    def get(self, name: str) -> SourceBackend:
+        backend = self._by_name.get(name)
+        if backend is None:
+            raise NotFoundError(f"Source {name!r} is not configured.")
+        return backend
+
+    def indexable(self, name: str) -> IndexableSource:
+        backend = self.get(name)
+        if not hasattr(backend, "scan"):
+            raise ValidationError(f"Source {name!r} cannot be indexed.")
+        return cast(IndexableSource, backend)
+
+    def searchable(self, name: str) -> SearchableSource:
+        backend = self.get(name)
+        if not hasattr(backend, "search"):
+            raise ValidationError(f"Source {name!r} cannot be searched.")
+        return cast(SearchableSource, backend)
+
+    def fetchable(self, name: str) -> FetchableSource:
+        backend = self.get(name)
+        if not hasattr(backend, "fetch"):
+            raise ValidationError(f"Source {name!r} cannot download.")
+        return cast(FetchableSource, backend)
+
+    def searchables(self) -> list[SearchableSource]:
+        """Every registered source that supports search (for cross-source search)."""
+        return [cast(SearchableSource, b) for b in self._by_name.values() if hasattr(b, "search")]
+
+    def indexables(self) -> list[IndexableSource]:
+        """Every registered source that can be indexed (for a full re-scan)."""
+        return [cast(IndexableSource, b) for b in self._by_name.values() if hasattr(b, "scan")]
+
+    def infos(self) -> list[SourceInfo]:
+        return [backend.info() for backend in self._by_name.values()]
+
+
+def build_source_registry(settings: Settings) -> SourceRegistry:
+    backends: list[SourceBackend] = []
+    if settings.local_media_import_path is not None:
+        backends.append(LocalFolderSource(settings.local_media_import_path))
+    if settings.youtube_enabled:
+        backends.append(
+            YouTubeMusicSource(
+                cookies_path=settings.youtube_cookies_path,
+                tmp_dir=settings.upload_tmp_dir,
+            )
+        )
+    return SourceRegistry(backends)

app/infrastructure/sources/youtube.py

@@ -0,0 +1,207 @@
+"""``youtube`` source — YouTube Music search + download (plan §5).
+
+A *fetch* source: it searches YouTube Music (via ``ytmusicapi``, which returns
+clean song/artist/album/duration rows) and downloads the chosen item with
+``yt-dlp``. The two libraries are synchronous, so every call is bounced to a
+worker thread (``anyio.to_thread``); the sync yt-dlp progress hook bridges back
+to the async progress callback via ``anyio.from_thread``.
+
+Both libraries are optional dependencies — if either is missing the source is
+simply *unavailable* (it never crashes import or the registry; graceful
+degradation per CLAUDE.md). The audio stream is stored **as-is** (YouTube serves
+lossy Opus/AAC; re-encoding would be lossy→lossy, plan §6.6).
+
+``source_id`` is the YouTube ``videoId`` — stable, so a re-download of the same
+id is idempotent and dedups against an existing track.
+"""
+
+import functools
+import tempfile
+from collections.abc import Callable
+from pathlib import Path
+from typing import Any
+
+import anyio
+
+from app.core.logging import get_logger
+from app.domain.ports import ProgressCallback
+from app.domain.sources import (
+    KIND_FETCH,
+    DownloadResult,
+    RawMetadata,
+    SearchResult,
+    SourceInfo,
+)
+from app.infrastructure.db.models.enums import TrackSource
+
+log = get_logger(__name__)
+
+# Functions a caller may inject for testing (defaults do the real library work).
+SearchFn = Callable[[str, int], list[dict[str, Any]]]
+# (video_id, tmp_dir, progress_hook, cookies_path) -> normalized download dict
+DownloadFn = Callable[[str, Path, Callable[[dict[str, Any]], None], Path | None], dict[str, Any]]
+
+
+def _libs_available() -> bool:
+    try:
+        import yt_dlp  # noqa: F401
+        import ytmusicapi  # noqa: F401
+    except ImportError:
+        return False
+    return True
+
+
+def _watch_url(video_id: str) -> str:
+    return f"https://music.youtube.com/watch?v={video_id}"
+
+
+class YouTubeMusicSource:
+    """Implements :class:`app.domain.ports.SearchableSource` and
+    :class:`~app.domain.ports.FetchableSource`."""
+
+    name = TrackSource.YOUTUBE.value
+
+    def __init__(
+        self,
+        *,
+        cookies_path: Path | None = None,
+        tmp_dir: Path | None = None,
+        search_fn: SearchFn | None = None,
+        download_fn: DownloadFn | None = None,
+    ) -> None:
+        self._cookies_path = cookies_path
+        self._tmp_dir = tmp_dir
+        self._search_fn = search_fn or _default_search
+        self._download_fn = download_fn or _default_download
+        # Only the real library path needs the deps; an injected fn is self-contained.
+        self._injected = search_fn is not None or download_fn is not None
+
+    def info(self) -> SourceInfo:
+        return SourceInfo(
+            name=self.name,
+            label="YouTube Music",
+            kind=KIND_FETCH,
+            available=self.is_available(),
+        )
+
+    def is_available(self) -> bool:
+        return True if self._injected else _libs_available()
+
+    async def search(self, query: str, *, limit: int) -> list[SearchResult]:
+        query = query.strip()
+        if not query:
+            return []
+        try:
+            rows = await anyio.to_thread.run_sync(functools.partial(self._search_fn, query, limit))
+        except Exception:
+            # No results / service down → degrade to empty (plan §5, CLAUDE.md).
+            log.warning("ytm_search_failed", query=query)
+            return []
+        return [r for r in (self._to_result(row) for row in rows) if r is not None]
+
+    async def fetch(
+        self, source_id: str, *, on_progress: ProgressCallback | None = None
+    ) -> DownloadResult:
+        tmp_dir = self._tmp_dir or Path(tempfile.gettempdir())
+
+        def hook(d: dict[str, Any]) -> None:
+            if on_progress is None or d.get("status") != "downloading":
+                return
+            total = d.get("total_bytes") or d.get("total_bytes_estimate")
+            done = d.get("downloaded_bytes")
+            if not total or done is None:
+                return
+            # Cap below 1.0 — the job only reaches 1.0 once stored + imported.
+            frac = min(done / total, 0.99)
+            # Bridge sync hook (worker thread) → async callback (event loop).
+            anyio.from_thread.run(on_progress, frac)
+
+        def _run() -> dict[str, Any]:
+            return self._download_fn(source_id, tmp_dir, hook, self._cookies_path)
+
+        info = await anyio.to_thread.run_sync(_run)
+        path = Path(info["filepath"])
+        stat = await anyio.Path(path).stat()
+        return DownloadResult(
+            source_id=source_id,
+            path=path,
+            file_format=info["file_format"],
+            file_size=stat.st_size,
+            bitrate=info.get("bitrate"),
+            suggested_title=info.get("title") or source_id,
+        )
+
+    async def get_metadata(self, source_id: str) -> RawMetadata | None:
+        # The search result already carries a usable title/artist, and the
+        # canonical metadata comes from enrichment (§6.2). A dedicated lookup is
+        # an optional refinement — skipped for now (returns None gracefully).
+        return None
+
+    def _to_result(self, row: dict[str, Any]) -> SearchResult | None:
+        video_id = row.get("videoId")
+        if not video_id:
+            return None  # non-playable row (e.g. a video without audio id)
+        artists = row.get("artists") or []
+        artist = ", ".join(a["name"] for a in artists if a.get("name")) or None
+        album = (row.get("album") or {}).get("name") if isinstance(row.get("album"), dict) else None
+        thumbnails = row.get("thumbnails") or []
+        thumbnail = thumbnails[-1].get("url") if thumbnails else None
+        return SearchResult(
+            source=self.name,
+            source_id=str(video_id),
+            title=row.get("title") or "Unknown",
+            artist=artist,
+            album=album,
+            duration_seconds=row.get("duration_seconds"),
+            thumbnail_url=thumbnail,
+            raw=row,
+        )
+
+
+def _default_search(query: str, limit: int) -> list[dict[str, Any]]:
+    """Real ytmusicapi search (songs only). Runs in a worker thread."""
+    from ytmusicapi import YTMusic
+
+    yt = YTMusic()  # unauthenticated: public search needs no login
+    results: list[dict[str, Any]] = yt.search(query, filter="songs", limit=limit)
+    return results[:limit]
+
+
+def _default_download(
+    video_id: str,
+    tmp_dir: Path,
+    progress_hook: Callable[[dict[str, Any]], None],
+    cookies_path: Path | None,
+) -> dict[str, Any]:
+    """Real yt-dlp download of the best audio stream. Runs in a worker thread.
+
+    Stores the original stream (no transcode — plan §6.3/§6.6). Returns a
+    normalized dict the adapter maps to :class:`DownloadResult`.
+    """
+    from yt_dlp import YoutubeDL
+
+    opts: dict[str, Any] = {
+        "format": "bestaudio/best",
+        "outtmpl": str(tmp_dir / "%(id)s.%(ext)s"),
+        "quiet": True,
+        "no_warnings": True,
+        "noprogress": True,
+        "progress_hooks": [progress_hook],
+    }
+    # Use cookies only when the file is actually present: the path can be set
+    # unconditionally (e.g. a mounted volume that may be empty) and downloads
+    # still work without it — cookies just unlock age/region-restricted items.
+    if cookies_path is not None and cookies_path.is_file():
+        opts["cookiefile"] = str(cookies_path)
+
+    with YoutubeDL(opts) as ydl:
+        info = ydl.extract_info(_watch_url(video_id), download=True)
+        filepath = Path(ydl.prepare_filename(info))
+
+    abr = info.get("abr")
+    return {
+        "filepath": filepath,
+        "file_format": filepath.suffix.lstrip(".").lower() or "m4a",
+        "bitrate": int(abr) if abr else None,
+        "title": info.get("title"),
+    }

app/infrastructure/storage/local.py

@@ -8,7 +8,7 @@ from pathlib import Path
 
 import anyio
 
-from app.domain.entities.storage import ObjectStat
+from app.domain.entities.storage import DiskUsage, ObjectStat
 from app.domain.errors import StorageError
 
 _EXT_CONTENT_TYPE: dict[str, str] = {
@@ -78,6 +78,15 @@ class LocalFileStorage:
     async def delete(self, key: str) -> None:
         (self._media_path / key).unlink(missing_ok=True)
 
+    async def disk_usage(self) -> DiskUsage | None:
+        # The media root may not exist yet on a fresh instance — walk up to the
+        # nearest existing ancestor so we still report the underlying volume.
+        path = self._media_path
+        while not path.exists() and path != path.parent:
+            path = path.parent
+        usage = await anyio.to_thread.run_sync(shutil.disk_usage, str(path))
+        return DiskUsage(total=usage.total, used=usage.used, free=usage.free)
+
     def as_local_path(self, key: str) -> AbstractAsyncContextManager[Path]:
         return self._as_local_path_cm(key)