watchlist: don't fuse different editions as the same album (Sokhi: Expedition 33)

_normalize_album_for_match stripped ANY trailing '- clause', so a real distinguishing
subtitle ('Clair Obscur: Expedition 33 - Nos vies en Lumière (Bonus Edition)') collapsed to
the same name as the OST → _albums_likely_match treated them as one album → the watchlist
marked unowned tracks of one edition as owned via the other and under-wishlisted.

- strip a trailing '- ...' clause ONLY when every token in it is an edition/format
  qualifier (+ connectors / year-ordinal): '- Single', '- Acoustic Version', '- 2011
  Remaster' still collapse, but real subtitles ('- Nos vies en Lumière', '- Volume 2',
  '- Live in Berlin') are kept. Avoids the inverse regression (a same-album pair splitting
  into a redownload loop), which a naive narrow strip would have caused.
- drop the loose substring shortcut + raise the fuzzy floor 0.6→0.85; genuine drift already
  collapses to an EXACT match, so the looseness only ever produced false fuses.

blast radius: _albums_likely_match has exactly one caller (the allow-duplicates skip).
48 album-match tests pass (qualifier-suffix merges + edition-subtitle splits) + 219 watchlist.

core/watchlist_scanner.py

@@ -328,6 +328,22 @@ _ALBUM_QUALIFIER_RE = re.compile(
     re.IGNORECASE,
 )
 
+# A trailing "- ..." clause is stripped ONLY when EVERY token in it is an edition/format
+# qualifier (+ connectors / a year-ordinal). So "- Single", "- Acoustic Version", "- 2011
+# Remaster" collapse to the base name, but a real distinguishing subtitle ("- Nos vies en
+# Lumière", "- Live in Berlin") is kept — the bug was a blanket "- anything$" strip that
+# erased subtitles and fused different editions (Sokhi: Expedition 33 OST vs Bonus Edition).
+_DASH_QUALIFIER_WORD = (
+    r'live|acoustic|electric|instrumental|unplugged|mono|stereo|demos?|reissue|'
+    r'remix(?:es)?|edit(?:ed)?|radio|single|ep|lp|version|mix(?:es)?|sessions?|bootleg|'
+    r'covers?|original|redux|deluxe|expanded|remaster(?:ed)?|anniversary|special|'
+    r'edition|bonus|extended|explicit|clean|soundtrack|ost|score'
+)
+_TRAILING_DASH_QUALIFIER_RE = re.compile(
+    r'\s*-\s*(?:(?:' + _DASH_QUALIFIER_WORD + r'|the|a|and|&|\+|\d+(?:st|nd|rd|th)?)\b[\s\-]*)+$',
+    re.IGNORECASE,
+)
+
 
 def _normalize_album_for_match(name: str) -> str:
     """Return a canonical form of an album name suitable for fuzzy comparison.
@@ -347,8 +363,9 @@ def _normalize_album_for_match(name: str) -> str:
     # they're almost always edition or commentary noise, not part of the
     # album's identifying name.
     cleaned = re.sub(r'\s*[\(\[][^\)\]]*[\)\]]\s*', ' ', cleaned)
-    # Trailing dash-clauses ("Album - Remastered", "Album - Live")
+    # Trailing dash-clause, but ONLY when it's entirely edition/format qualifiers — a real
-    cleaned = re.sub(r'\s*-\s*[^-]+$', '', cleaned)
+    # subtitle is preserved (see _TRAILING_DASH_QUALIFIER_RE).
+    cleaned = _TRAILING_DASH_QUALIFIER_RE.sub(' ', cleaned)
     cleaned = re.sub(r'[^a-z0-9 ]+', ' ', cleaned.lower())
     cleaned = re.sub(r'\s+', ' ', cleaned).strip()
     return cleaned
@@ -382,7 +399,7 @@ def _extract_volume_marker(normalized_name: str):
     return last.group(1) or last.group(2)
 
 
-def _albums_likely_match(spotify_album: str, lib_album: str, threshold: float = 0.6) -> bool:
+def _albums_likely_match(spotify_album: str, lib_album: str, threshold: float = 0.85) -> bool:
     """Return True when two album names plausibly identify the same release.
 
     Designed to swallow naming drift between metadata sources and the
@@ -406,11 +423,11 @@ def _albums_likely_match(spotify_album: str, lib_album: str, threshold: float =
         return False
     if norm_a == norm_b:
         return True
-    # After normalization the shorter name often becomes a prefix /
+    # No loose substring shortcut: after qualifier-stripping, a short name being a
-    # substring of the longer one ("napoleon dynamite" ⊂ "napoleon
+    # prefix of a longer one is usually a DIFFERENT edition carrying a real subtitle
-    # dynamite music from the motion picture" before stripping).
+    # ("clair obscur expedition 33" ⊂ "clair obscur expedition 33 nos vies en lumiere"),
-    if norm_a in norm_b or norm_b in norm_a:
+    # not naming drift. Genuine drift collapses to an EXACT match above; everything else
-        return True
+    # must clear a high overall-similarity bar.
     return SequenceMatcher(None, norm_a, norm_b).ratio() >= threshold
 
 

tests/test_watchlist_album_match.py

@@ -156,6 +156,13 @@ def test_compilation_score_explanation() -> None:
         ("Inception (Music From The Motion Picture)", "Inception Soundtrack"),
         # Substring containment
         ("Random Access Memories", "Random Access Memories (Bonus Edition)"),
+        # Dash-suffixed qualifiers must still collapse — these are the SAME album, so
+        # treating them as different would re-wishlist/redownload forever (the failure the
+        # original blanket strip guarded against; the narrowed strip must keep covering it).
+        ("Album Name", "Album Name - Single"),
+        ("Album Name", "Album Name - Acoustic Version"),
+        ("Hotel California", "Hotel California - 2013 Remaster"),
+        ("Album", "Album - The Remixes"),
     ],
 )
 def test_likely_match_positive(spotify_name, lib_name) -> None:
@@ -176,12 +183,29 @@ def test_likely_match_positive(spotify_name, lib_name) -> None:
         ("Abbey Road", "Sgt. Pepper's Lonely Hearts Club Band"),
         # Same word in title but different album
         ("Greatest Hits Volume 1", "Greatest Hits Volume 2"),
+        # Sokhi: distinct editions of the same franchise — the OST vs a bonus edition
+        # with a real subtitle. USED to collapse to the same normalized name (the blanket
+        # trailing-dash strip removed "- Nos vies en Lumière"), so the watchlist marked
+        # unowned OST tracks as owned via the bonus edition. Must be DIFFERENT albums.
+        ("Clair Obscur: Expedition 33: Original Soundtrack",
+         "Clair Obscur: Expedition 33 - Nos vies en Lumière (Bonus Edition)"),
+        # A real subtitle after a dash must not be stripped down to the base name.
+        ("The Album", "The Album - A Whole Different Subtitle"),
     ],
 )
 def test_likely_match_negative(spotify_name, lib_name) -> None:
     assert not _albums_likely_match(spotify_name, lib_name)
 
 
+def test_real_subtitle_after_dash_is_preserved() -> None:
+    # the regression's root cause: a meaningful subtitle must survive normalization,
+    # while a recognized qualifier after a dash ("- Live", "- 2011") still collapses.
+    assert _normalize_album_for_match("Clair Obscur: Expedition 33 - Nos vies en Lumière") \
+        != _normalize_album_for_match("Clair Obscur: Expedition 33")
+    assert _normalize_album_for_match("Some Album - Live") == _normalize_album_for_match("Some Album")
+    assert _normalize_album_for_match("Some Album - 2011") == _normalize_album_for_match("Some Album")
+
+
 # ---------------------------------------------------------------------------
 # _albums_likely_match — defensive cases
 # ---------------------------------------------------------------------------