feat: add confidence scoring

worker/lyra_worker/confidence.py

@@ -0,0 +1,21 @@
+from difflib import SequenceMatcher
+
+from lyra_worker.types import Candidate, MBTarget
+
+
+def _sim(a: str, b: str) -> float:
+    return SequenceMatcher(None, a.strip().casefold(), b.strip().casefold()).ratio()
+
+
+def score_confidence(target: MBTarget, c: Candidate) -> float:
+    """0..1 confidence that candidate c is the requested target."""
+    artist_sim = _sim(target.artist, c.matched_artist)
+    album_sim = _sim(target.album, c.matched_album)
+
+    if target.track_count is None or target.track_count == c.track_count:
+        track_factor = 1.0
+    else:
+        diff = abs(target.track_count - c.track_count)
+        track_factor = max(0.0, 1.0 - diff / max(target.track_count, 1))
+
+    return 0.5 * artist_sim + 0.4 * album_sim + 0.1 * track_factor

worker/tests/test_confidence.py

@@ -0,0 +1,37 @@
+from lyra_worker.confidence import score_confidence
+from lyra_worker.types import Candidate, MBTarget, Quality
+
+Q = Quality(fmt="FLAC", lossless=True, bit_depth=16, sample_rate=44100)
+
+
+def _cand(artist, album, tracks=10):
+    return Candidate(source="qobuz", source_ref="x", matched_artist=artist,
+                     matched_album=album, quality=Q, track_count=tracks)
+
+
+def test_exact_match_scores_high():
+    target = MBTarget(artist="Radiohead", album="In Rainbows", track_count=10)
+    assert score_confidence(target, _cand("Radiohead", "In Rainbows", 10)) > 0.95
+
+
+def test_case_and_whitespace_insensitive():
+    target = MBTarget(artist="Radiohead", album="In Rainbows")
+    assert score_confidence(target, _cand("  radiohead ", "IN RAINBOWS")) > 0.95
+
+
+def test_wrong_album_scores_low():
+    target = MBTarget(artist="Radiohead", album="In Rainbows", track_count=10)
+    assert score_confidence(target, _cand("Radiohead", "OK Computer", 12)) < 0.7
+
+
+def test_unknown_track_count_does_not_penalize():
+    target = MBTarget(artist="Radiohead", album="In Rainbows", track_count=None)
+    # track_count unknown -> track_factor neutral; name match still drives a high score
+    assert score_confidence(target, _cand("Radiohead", "In Rainbows", 99)) > 0.95
+
+
+def test_track_count_mismatch_lowers_score():
+    target = MBTarget(artist="Radiohead", album="In Rainbows", track_count=10)
+    good = score_confidence(target, _cand("Radiohead", "In Rainbows", 10))
+    bad = score_confidence(target, _cand("Radiohead", "In Rainbows", 4))
+    assert bad < good