fix: address code review findings

- Fix pagination to not trust MusicBrainz count for cutoff (prevents
  silent discography truncation / hidden missing releases)
- Add idx_external_releases_artist_id index for hot per-artist queries
- Dedupe cached_at layout constant in external_releases.go

internal/database/database.go

@@ -121,6 +121,10 @@ func (db *DB) migrate() error {
 			name: "006_add_secondary_types_to_external_releases",
 			sql:  `ALTER TABLE external_releases ADD COLUMN secondary_types TEXT;`,
 		},
+		{
+			name: "007_index_external_releases_artist_id",
+			sql:  `CREATE INDEX IF NOT EXISTS idx_external_releases_artist_id ON external_releases(artist_id);`,
+		},
 	}
 
 	for _, m := range migrations {

internal/database/database_test.go

@@ -205,9 +205,10 @@ func TestMigrationTracking(t *testing.T) {
 		t.Fatalf("query migrations count: %v", err)
 	}
 
-	// We have 6 recorded migrations: artist_settings, external_releases,
+	// We have 7 recorded migrations: artist_settings, external_releases,
-	// local_albums, notifications_sent, cached_at column, secondary_types column.
+	// local_albums, notifications_sent, cached_at column, secondary_types
-	if count != 6 {
+	// column, and the external_releases.artist_id index.
-		t.Errorf("expected 6 applied migrations, got %d", count)
+	if count != 7 {
+		t.Errorf("expected 7 applied migrations, got %d", count)
 	}
 }

internal/database/external_releases.go

@@ -175,11 +175,10 @@ func GetExternalReleasesByArtistWithCache(db *DB, artistID string, ttl time.Dura
 	if ttl <= 0 {
 		return nil, nil
 	}
-	// cached_at is stored in the "2006-01-02 15:04:05" UTC layout via
+	// cached_at is stored in utcLayout via FormatCachedAt. Compare against an
-	// FormatCachedAt. Compare against an explicitly formatted cutoff string in
+	// explicitly formatted cutoff string in the same layout so the
-	// the same layout so the lexicographic comparison is a valid time ordering.
+	// lexicographic comparison is a valid time ordering.
-	const layout = "2006-01-02 15:04:05"
+	cutoff := time.Now().UTC().Add(-ttl).Format(utcLayout)
-	cutoff := time.Now().UTC().Add(-ttl).Format(layout)
 	rows, err := db.Conn().Query(
 		"SELECT rgid, artist_id, title, type, release_date, is_ignored, cached_at, secondary_types FROM external_releases WHERE artist_id = ? AND cached_at >= ?",
 		artistID, cutoff,

internal/musicbrainz/api.go

@@ -46,10 +46,14 @@ func (c *MusicBrainzClient) GetArtistReleaseGroups(ctx context.Context, artistMB
 
 		allGroups = append(allGroups, parsed.ReleaseGroups...)
 
-		// If we've fetched all results, we've reached the end.
+		// Stop when a page is empty (no more results) or when the page
-		// Also break on empty page to prevent infinite loop if API
+		// returned fewer items than the request limit — a reliable end-of-data
-		// returns fewer items than advertised by count.
+		// signal. We intentionally do NOT trust parsed.Count for the cutoff:
-		if len(parsed.ReleaseGroups) == 0 || offset+len(parsed.ReleaseGroups) >= parsed.Count {
+		// MusicBrainz occasionally reports an inaccurate count, which would
+		// prematurely truncate an artist's discography and hide missing
+		// releases. The empty-page check also prevents an infinite loop if the
+		// API keeps returning a non-empty page past the reported count.
+		if len(parsed.ReleaseGroups) == 0 || len(parsed.ReleaseGroups) < limit {
 			break
 		}