diff --git a/database/music_database.py b/database/music_database.py
index 6d95d422..17d2e41c 100644
--- a/database/music_database.py
+++ b/database/music_database.py
@@ -396,6 +396,7 @@ class MusicDatabase:
 
             # Add per-artist preferred_metadata_source column (migration)
             self._add_watchlist_preferred_metadata_source_column(cursor)
+            self._clear_deezer_ids_stored_as_itunes(cursor)
 
             # Make spotify_artist_id nullable for iTunes-only artists (migration)
             self._fix_watchlist_spotify_id_nullable(cursor)
@@ -2114,6 +2115,38 @@ class MusicDatabase:
         except Exception as e:
             logger.error(f"Error adding preferred_metadata_source column to watchlist_artists: {e}")
 
+    def _clear_deezer_ids_stored_as_itunes(self, cursor):
+        """Repair: watchlist iTunes ids that are actually Deezer ids.
+
+        The watchlist scanner's _match_to_itunes used to search via
+        MetadataService.itunes — which holds the PRIMARY source's client, not
+        iTunes — so with a Deezer primary it stored DEEZER artist ids in
+        itunes_artist_id (verified live: Taylor Swift's "iTunes" id was her
+        Deezer id 12246; real one is 159260351). The backfill only fills
+        EMPTY ids, so these wrong ids would never self-heal. The corruption
+        signature is itunes == deezer (distinct id spaces — a legit equal
+        pair is effectively impossible; worst case is a NULL that re-matches
+        correctly on the next scan). Idempotent: clearing kills the equality.
+        """
+        try:
+            cursor.execute("PRAGMA table_info(watchlist_artists)")
+            columns = [column[1] for column in cursor.fetchall()]
+            if 'itunes_artist_id' not in columns or 'deezer_artist_id' not in columns:
+                return
+            cursor.execute("""
+                UPDATE watchlist_artists SET itunes_artist_id = NULL
+                WHERE itunes_artist_id = deezer_artist_id
+                  AND deezer_artist_id IS NOT NULL AND deezer_artist_id != ''
+            """)
+            if cursor.rowcount:
+                logger.info(
+                    "Cleared %d watchlist iTunes id(s) that were actually Deezer ids "
+                    "(pre-fix _match_to_itunes searched the primary source); they "
+                    "re-match via real iTunes on the next watchlist scan",
+                    cursor.rowcount)
+        except Exception as e:
+            logger.error(f"Error clearing deezer-as-itunes watchlist ids: {e}")
+
     def _add_similar_artists_last_featured_column(self, cursor):
         """Add last_featured column to similar_artists for hero slider cycling"""
         try:
diff --git a/tests/test_watchlist_itunes_id_repair.py b/tests/test_watchlist_itunes_id_repair.py
new file mode 100644
index 00000000..2db22990
--- /dev/null
+++ b/tests/test_watchlist_itunes_id_repair.py
@@ -0,0 +1,72 @@
+"""Repair migration: watchlist iTunes ids that are actually Deezer ids.
+
+_match_to_itunes used to search the PRIMARY source's client (the misnamed
+MetadataService.itunes slot) and store that source's artist id in the
+itunes column — with a Deezer primary, Deezer ids landed as "iTunes" ids
+(verified in a live DB: 6 of 9 rows). Since the backfill only fills EMPTY
+ids, the migration must clear the corrupted ones (signature: itunes ==
+deezer) so the fixed matcher can re-fill them.
+"""
+
+from __future__ import annotations
+
+import sqlite3
+
+import database.music_database as mdb
+from database.music_database import MusicDatabase
+
+
+def _open_raw(db_path):
+    return sqlite3.connect(db_path)
+
+
+def _reinit(db_path):
+    """Schema init runs once per process per path (module memo) — clear the
+    memo so re-construction replays the migration block like a real app
+    restart (fresh process) would."""
+    mdb._database_initialized_paths.clear()
+    return MusicDatabase(db_path)
+
+
+def test_deezer_as_itunes_ids_cleared_and_legit_kept(tmp_path):
+    db_path = str(tmp_path / 'm.db')
+    MusicDatabase(db_path)  # create schema (migration runs, table empty)
+
+    conn = _open_raw(db_path)
+    c = conn.cursor()
+    c.execute("""INSERT INTO watchlist_artists (artist_name, spotify_artist_id, itunes_artist_id, deezer_artist_id)
+                 VALUES ('Taylor Swift', 'sp1', '12246', '12246')""")        # corrupted
+    c.execute("""INSERT INTO watchlist_artists (artist_name, spotify_artist_id, itunes_artist_id, deezer_artist_id)
+                 VALUES ('Eminem', 'sp2', '111051', '13')""")                # legit
+    c.execute("""INSERT INTO watchlist_artists (artist_name, spotify_artist_id, itunes_artist_id, deezer_artist_id)
+                 VALUES ('NoDeezer', 'sp3', '999', NULL)""")                 # no deezer id -> keep
+    conn.commit()
+    conn.close()
+
+    _reinit(db_path)  # like an app restart -> migration sweeps existing rows
+
+    conn = _open_raw(db_path)
+    c = conn.cursor()
+    c.execute("SELECT artist_name, itunes_artist_id FROM watchlist_artists ORDER BY artist_name")
+    got = dict(c.fetchall())
+    conn.close()
+
+    assert got['Taylor Swift'] is None          # corruption cleared
+    assert got['Eminem'] == '111051'            # real id untouched
+    assert got['NoDeezer'] == '999'             # equality needs a deezer id
+
+
+def test_migration_idempotent(tmp_path):
+    db_path = str(tmp_path / 'm.db')
+    MusicDatabase(db_path)
+    conn = _open_raw(db_path)
+    conn.execute("""INSERT INTO watchlist_artists (artist_name, spotify_artist_id, itunes_artist_id, deezer_artist_id)
+                    VALUES ('A', 'sp', '5', '5')""")
+    conn.commit()
+    conn.close()
+    _reinit(db_path)
+    _reinit(db_path)  # second run: nothing left to clear, no error
+    conn = _open_raw(db_path)
+    val = conn.execute("SELECT itunes_artist_id FROM watchlist_artists").fetchone()[0]
+    conn.close()
+    assert val is None