From e04c1c9795ef04438186a2cf104ba2211355166c Mon Sep 17 00:00:00 2001
From: Jonathan <claude.cs@jger.nl>
Date: Tue, 14 Jul 2026 22:39:52 +0200
Subject: [PATCH] feat(discover): also seed discovery from owned-album artists
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Discovery seeded only from followed artists. Now it also seeds from artists
you own albums by (LibraryItem.artistMbid, captured since the MBID own-state
work) but don't follow — so your library shapes suggestions, not just your
follows.

- New DiscoverySeed throttle table (migration add_discovery_seed) mirroring
  WatchedArtist.lastDiscoveredAt, so each library-derived seed is swept once
  per interval (WatchedArtist seeds keep their own throttle).
- _seed_artists merges both seed sources least-recently-swept first, capped at
  chunk_size; run_discovery stamps the right throttle per seed and evicts
  contributions from seeds that are neither followed nor owned.
- Artists you already have (followed OR owned) are excluded from the suggestion
  candidates — discovery surfaces only genuinely new artists.
- discover.seedFromLibrary config (default true) + a Discovery settings toggle.

NOTE: existing LibraryItem rows have null artistMbid (populated on new
import/scan), so a re-scan is needed before old albums seed discovery.
worker 236, web 187 tests.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../migration.sql                             |  7 ++
 web/prisma/schema.prisma                      |  8 ++
 web/src/app/api/discover/config/route.ts      |  1 +
 web/src/app/settings/settings-form.tsx        | 10 +++
 web/src/test/setup.ts                         |  1 +
 worker/lyra_worker/discovery.py               | 79 ++++++++++++++-----
 worker/tests/conftest.py                      |  2 +
 worker/tests/test_discovery.py                | 42 ++++++++++
 8 files changed, 132 insertions(+), 18 deletions(-)
 create mode 100644 web/prisma/migrations/20260714203518_add_discovery_seed/migration.sql

diff --git a/web/prisma/migrations/20260714203518_add_discovery_seed/migration.sql b/web/prisma/migrations/20260714203518_add_discovery_seed/migration.sql
new file mode 100644
index 0000000..869551d
--- /dev/null
+++ b/web/prisma/migrations/20260714203518_add_discovery_seed/migration.sql
@@ -0,0 +1,7 @@
+-- CreateTable
+CREATE TABLE "DiscoverySeed" (
+    "mbid" TEXT NOT NULL,
+    "lastDiscoveredAt" TIMESTAMP(3),
+
+    CONSTRAINT "DiscoverySeed_pkey" PRIMARY KEY ("mbid")
+);
diff --git a/web/prisma/schema.prisma b/web/prisma/schema.prisma
index ee1ef0c..37dea2d 100644
--- a/web/prisma/schema.prisma
+++ b/web/prisma/schema.prisma
@@ -191,6 +191,14 @@ model DiscoverySeedContribution {
   @@index([seedMbid])
 }
 
+// Throttle registry for discovery seeds that are NOT WatchedArtist rows — i.e. artists you own
+// albums by (LibraryItem.artistMbid) but don't follow. Mirrors WatchedArtist.lastDiscoveredAt so
+// each library-derived seed is swept once per interval, not every sweep.
+model DiscoverySeed {
+  mbid             String    @id
+  lastDiscoveredAt DateTime?
+}
+
 model ScanWorkItem {
   id     String  @id @default(cuid())
   scanId String
diff --git a/web/src/app/api/discover/config/route.ts b/web/src/app/api/discover/config/route.ts
index 81d3403..4c2b235 100644
--- a/web/src/app/api/discover/config/route.ts
+++ b/web/src/app/api/discover/config/route.ts
@@ -6,6 +6,7 @@ const DEFAULTS: Record<string, string> = {
   "discover.similarPerSeed": "20",
   "discover.albumsPerArtist": "1",
   "discover.albumsOnly": "true",
+  "discover.seedFromLibrary": "true",
   "discover.minScore": "0",
   "discover.chunkSize": "5",
   "discover.listenBrainzUrl": "",
diff --git a/web/src/app/settings/settings-form.tsx b/web/src/app/settings/settings-form.tsx
index 1922e19..8f48202 100644
--- a/web/src/app/settings/settings-form.tsx
+++ b/web/src/app/settings/settings-form.tsx
@@ -395,6 +395,8 @@ export function SettingsForm() {
               <b> Similar per seed</b> — candidates pulled per artist. <b>Albums per artist</b> —
               releases surfaced per suggested artist. <b>Min score</b> — drops weak suggestions.{" "}
               <b>Albums only</b> — suggest full-length albums only, hiding singles and EPs.{" "}
+              <b>Seed from library</b> — also find artists similar to ones you own albums by, not
+              just the artists you follow.{" "}
               <b>ListenBrainz base URL</b> — leave blank for the default endpoint.
             </HelpButton>
           </div>
@@ -414,6 +416,14 @@ export function SettingsForm() {
             />
             Albums only (hide singles &amp; EPs)
           </label>
+          <label className="toggle" style={{ margin: "6px 0 0" }}>
+            <input
+              type="checkbox"
+              checked={discoverCfg["discover.seedFromLibrary"] !== "false"}
+              onChange={(e) => setCfg("discover.seedFromLibrary", e.target.checked ? "true" : "false")}
+            />
+            Seed from library (similar to albums you own)
+          </label>
           <label className="field">
             <span>ListenBrainz base URL (blank = default)</span>
             <input value={discoverCfg["discover.listenBrainzUrl"] ?? ""}
diff --git a/web/src/test/setup.ts b/web/src/test/setup.ts
index 43e2488..cc9df86 100644
--- a/web/src/test/setup.ts
+++ b/web/src/test/setup.ts
@@ -17,6 +17,7 @@ beforeEach(async () => {
   await prisma.watchedArtist.deleteMany();
   await prisma.discoverySuggestion.deleteMany();
   await prisma.discoverySeedContribution.deleteMany();
+  await prisma.discoverySeed.deleteMany();
   await prisma.config.deleteMany();
   await prisma.apiCache.deleteMany(); // read-through cache must not leak across tests
   await prisma.unmatchedAlbum.deleteMany();
diff --git a/worker/lyra_worker/discovery.py b/worker/lyra_worker/discovery.py
index 3370224..66b625b 100644
--- a/worker/lyra_worker/discovery.py
+++ b/worker/lyra_worker/discovery.py
@@ -18,6 +18,7 @@ class DiscoveryConfig:
     albums_per_artist: int = 1
     min_score: float = 0.0
     albums_only: bool = True
+    seed_from_library: bool = True
 
     @classmethod
     def from_config(cls, config: dict) -> "DiscoveryConfig":
@@ -47,6 +48,7 @@ class DiscoveryConfig:
             albums_per_artist=_int("discover.albumsPerArtist", 1),
             min_score=_float("discover.minScore", 0.0),
             albums_only=_bool("discover.albumsOnly", True),
+            seed_from_library=_bool("discover.seedFromLibrary", True),
         )
 
 
@@ -71,15 +73,37 @@ def _healthy_sources(sources):
 
 
 def _seed_artists(conn: psycopg.Connection, cfg: DiscoveryConfig):
+    """Return up to chunk_size due seeds as (mbid, name, source) where source is 'watched'
+    (a followed artist, throttled on WatchedArtist.lastDiscoveredAt) or 'library' (an artist
+    you own albums by but don't follow, throttled on DiscoverySeed.lastDiscoveredAt). Both are
+    ordered least-recently-swept first so a sweep drains all eligible seeds across chunks."""
+    interval = cfg.interval_hours
     with conn.cursor() as cur:
         cur.execute(
-            'SELECT mbid, name FROM "WatchedArtist" '
-            'WHERE "lastDiscoveredAt" IS NULL '
-            '   OR "lastDiscoveredAt" < now() - make_interval(hours => %s) '
-            'ORDER BY "lastDiscoveredAt" ASC NULLS FIRST LIMIT %s',
-            (cfg.interval_hours, cfg.chunk_size),
+            'SELECT mbid, name, \'watched\' AS source, "lastDiscoveredAt" FROM "WatchedArtist" '
+            'WHERE "lastDiscoveredAt" IS NULL OR "lastDiscoveredAt" < now() - make_interval(hours => %s) ',
+            (interval,),
         )
-        return cur.fetchall()
+        rows = cur.fetchall()
+
+        if cfg.seed_from_library:
+            # Distinct owned-album artists (with a captured MBID) that we don't already follow,
+            # left-joined to their throttle row. A brand-new library seed (no DiscoverySeed row)
+            # is immediately eligible via NULLS FIRST.
+            cur.execute(
+                'SELECT DISTINCT ON (li."artistMbid") li."artistMbid", li.artist, \'library\', ds."lastDiscoveredAt" '
+                'FROM "LibraryItem" li '
+                'LEFT JOIN "DiscoverySeed" ds ON ds.mbid = li."artistMbid" '
+                'WHERE li."artistMbid" IS NOT NULL '
+                '  AND li."artistMbid" NOT IN (SELECT mbid FROM "WatchedArtist") '
+                '  AND (ds."lastDiscoveredAt" IS NULL OR ds."lastDiscoveredAt" < now() - make_interval(hours => %s))',
+                (interval,),
+            )
+            rows += cur.fetchall()
+
+    # Merge both sources, least-recently-swept first (NULL = never), cap at chunk_size.
+    rows.sort(key=lambda r: (r[3] is not None, r[3]))
+    return [(mbid, name, source) for mbid, name, source, _last in rows[: cfg.chunk_size]]
 
 
 def _followed_mbids(conn: psycopg.Connection) -> set[str]:
@@ -88,6 +112,14 @@ def _followed_mbids(conn: psycopg.Connection) -> set[str]:
         return {r[0] for r in cur.fetchall()}
 
 
+def _owned_artist_mbids(conn: psycopg.Connection) -> set[str]:
+    """MBIDs of artists you own albums by (captured on LibraryItem). Used both as extra
+    discovery seeds and to keep already-owned artists out of the suggestion candidates."""
+    with conn.cursor() as cur:
+        cur.execute('SELECT DISTINCT "artistMbid" FROM "LibraryItem" WHERE "artistMbid" IS NOT NULL')
+        return {r[0] for r in cur.fetchall()}
+
+
 def _upsert_artist(conn: psycopg.Connection, cand: dict) -> int:
     dedupe = f"artist:{cand['mbid']}:"
     with conn.cursor() as cur:
@@ -250,28 +282,39 @@ def run_discovery(conn: psycopg.Connection, sources: list[SimilaritySource],
     every seed's latest contribution, independent of chunk and sweep boundaries."""
     seeds = _seed_artists(conn, cfg)
     followed = _followed_mbids(conn)
+    owned = _owned_artist_mbids(conn) if cfg.seed_from_library else set()
+    # Artists you already have (followed or owned) are seeds/inputs, never suggested as
+    # candidates — discovery only surfaces genuinely new artists.
+    known = followed | owned
     live = _healthy_sources(sources)
 
-    # Evict contributions from seeds no longer followed — unfollow/removal has no
-    # FK cascade, so without this a removed seed's score would linger and inflate
-    # the aggregate forever. Seed `affected` with the touched candidates so their
-    # suggestion scores are recomputed from the remaining live seeds this sweep.
+    # Evict contributions whose seed is no longer valid — an unfollowed artist AND (when
+    # library seeding is on) an artist whose last owned album was deleted. Neither has an FK
+    # cascade, so without this a stale seed's score would linger and inflate the aggregate.
+    # Seed `affected` with the touched candidates so their scores recompute from live seeds.
+    valid_seeds = list(known)
     with conn.cursor() as cur:
         cur.execute('SELECT DISTINCT "candidateMbid" FROM "DiscoverySeedContribution" '
-                    'WHERE "seedMbid" NOT IN (SELECT mbid FROM "WatchedArtist")')
+                    'WHERE NOT ("seedMbid" = ANY(%s))', (valid_seeds,))
         affected = {r[0] for r in cur.fetchall()}
-        cur.execute('DELETE FROM "DiscoverySeedContribution" '
-                    'WHERE "seedMbid" NOT IN (SELECT mbid FROM "WatchedArtist")')
+        cur.execute('DELETE FROM "DiscoverySeedContribution" WHERE NOT ("seedMbid" = ANY(%s))',
+                    (valid_seeds,))
 
-    for seed_mbid, _seed_name in seeds:
-        affected |= _record_contributions(conn, seed_mbid, live, followed, cfg)
+    for seed_mbid, _seed_name, _source in seeds:
+        affected |= _record_contributions(conn, seed_mbid, live, known, cfg)
 
     surfaced = _recompute_suggestions(conn, affected, cfg)
     albums = _derive_albums(conn, browser, surfaced, cfg)
 
     with conn.cursor() as cur:
-        for seed_mbid, _seed_name in seeds:
-            cur.execute('UPDATE "WatchedArtist" SET "lastDiscoveredAt" = now() WHERE mbid = %s',
-                        (seed_mbid,))
+        for seed_mbid, _seed_name, source in seeds:
+            if source == "library":
+                cur.execute(
+                    'INSERT INTO "DiscoverySeed" (mbid, "lastDiscoveredAt") VALUES (%s, now()) '
+                    'ON CONFLICT (mbid) DO UPDATE SET "lastDiscoveredAt" = now()',
+                    (seed_mbid,))
+            else:
+                cur.execute('UPDATE "WatchedArtist" SET "lastDiscoveredAt" = now() WHERE mbid = %s',
+                            (seed_mbid,))
     conn.commit()
     return DiscoveryResult(artists=len(surfaced), albums=albums, seeds=len(seeds))
diff --git a/worker/tests/conftest.py b/worker/tests/conftest.py
index f70cf0c..d571424 100644
--- a/worker/tests/conftest.py
+++ b/worker/tests/conftest.py
@@ -29,6 +29,7 @@ def conn():
         cur.execute('DELETE FROM "WatchedArtist"')
         cur.execute('DELETE FROM "DiscoverySuggestion"')
         cur.execute('DELETE FROM "DiscoverySeedContribution"')
+        cur.execute('DELETE FROM "DiscoverySeed"')
         cur.execute('DELETE FROM "ScanWorkItem"')
         cur.execute('DELETE FROM "UnmatchedAlbum"')
         cur.execute('DELETE FROM "ApiCache"')
@@ -48,6 +49,7 @@ def conn():
         cur.execute('DELETE FROM "WatchedArtist"')
         cur.execute('DELETE FROM "DiscoverySuggestion"')
         cur.execute('DELETE FROM "DiscoverySeedContribution"')
+        cur.execute('DELETE FROM "DiscoverySeed"')
         cur.execute('DELETE FROM "ScanWorkItem"')
         cur.execute('DELETE FROM "UnmatchedAlbum"')
         cur.execute('DELETE FROM "ApiCache"')
diff --git a/worker/tests/test_discovery.py b/worker/tests/test_discovery.py
index 445e5e2..c473ceb 100644
--- a/worker/tests/test_discovery.py
+++ b/worker/tests/test_discovery.py
@@ -11,6 +11,48 @@ def _artist_rows(conn):
         return cur.fetchall()
 
 
+def _insert_library_item(conn, artist, artist_mbid, album="Owned Album"):
+    with conn.cursor() as cur:
+        cur.execute(
+            'INSERT INTO "LibraryItem" (id, artist, album, path, source, format, "qualityClass", '
+            '"artistMbid", "importedAt") '
+            "VALUES (gen_random_uuid()::text, %s, %s, %s, 'scan', 'FLAC', 2, %s, now())",
+            (artist, album, f"/music/{artist}/{album}", artist_mbid),
+        )
+    conn.commit()
+
+
+def test_seeds_discovery_from_owned_album_artists(conn):
+    # An artist you own an album by but don't follow becomes a discovery seed.
+    _insert_library_item(conn, "Owned Artist", "lib-seed")
+    src = FakeSimilaritySource(similar={"lib-seed": [SimilarArtist("c1", "New Find", 0.9)]})
+    result = run_discovery(conn, [src], FakeMbBrowser(), DiscoveryConfig())
+    assert result.seeds == 1
+    assert [r[0] for r in _artist_rows(conn)] == ["c1"]  # surfaced from the library seed
+    with conn.cursor() as cur:  # library seed throttled via DiscoverySeed, not WatchedArtist
+        cur.execute('SELECT "lastDiscoveredAt" IS NOT NULL FROM "DiscoverySeed" WHERE mbid = %s', ("lib-seed",))
+        assert cur.fetchone()[0] is True
+
+
+def test_library_seeding_can_be_disabled(conn):
+    _insert_library_item(conn, "Owned Artist", "lib-seed")
+    src = FakeSimilaritySource(similar={"lib-seed": [SimilarArtist("c1", "New Find", 0.9)]})
+    result = run_discovery(conn, [src], FakeMbBrowser(), DiscoveryConfig(seed_from_library=False))
+    assert result.seeds == 0 and _artist_rows(conn) == []  # library not seeded
+
+
+def test_owned_artist_is_not_suggested_as_a_candidate(conn):
+    # You own an album by "c1"; a followed seed finds c1 similar. c1 must not be suggested back.
+    insert_watched_artist(conn, mbid="s1", name="Seed")
+    _insert_library_item(conn, "Owned", "c1")
+    src = FakeSimilaritySource(similar={
+        "s1": [SimilarArtist("c1", "Owned", 0.9), SimilarArtist("c2", "New", 0.8)],
+        "c1": [],  # the library seed itself finds nothing new here
+    })
+    run_discovery(conn, [src], FakeMbBrowser(), DiscoveryConfig())
+    assert [r[0] for r in _artist_rows(conn)] == ["c2"]  # c1 excluded (already owned)
+
+
 def test_aggregates_scores_across_seeds(conn):
     insert_watched_artist(conn, mbid="s1", name="Seed One")
     insert_watched_artist(conn, mbid="s2", name="Seed Two")