v0.16.0: the file already told you

Two real failures from the first hour of round 45, both fixed by treating
the file itself as evidence.

"B. Clem - Zuuso [1025657891]" fingerprinted fine and matched nothing.
Checked before building: AcoustID returns zero results, a MusicBrainz text
search for "Zuuso" returns zero, iTunes returns zero. The track is in no
metadata service on earth, so no smarter lookup could have answered it —
but its filename said exactly what it was. New filename_tags.py reads that,
validated against all 474 files in the untagged folder rather than invented
examples: it strips yt-dlp ids, drops "(Official Video)" noise, splits
Artist - Title, reads a leading 1-04, and falls back to the iTunes tree.
It is careful about what it removes — an 11-char trailing token counts as a
YouTube id only if it carries a digit, an underscore or flipping case, or
"Cold Draft-Underground" loses a word, and a hyphen only splits when it has
spaces around it so Jay-Z survives.

A guess says it is one: candidates carry a source, and the dialog quotes a
confidence only for real fingerprint matches.

The file now ranks lookup results too, which fixes the Alhambra case: token
overlap (words double, bare numbers single), a duration bucket against the
recording lengths AcoustID returns, and a filename year that predates what
the database knows. Live also stops counting as a demerit — it was lumped in
with Compilation and buried a live album for a file whose name said "Live At
The Alhambra". That track now proposes the right album with year 1961
instead of a compilation with 2002.

The year genuinely needed the filename: MusicBrainz's own first-release-date
for "Ahmad Jamal's Alhambra" is 2002, because its three original 1961
pressings are stored undated. The obvious fix of a second MusicBrainz call
returns the same wrong answer.

Not done on purpose: trav rejected "never propose a shorter title" —
truncating is wanted, since the garbage is usually the part being dropped.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NsiFHdyVg1UhBSxDJfTNRm

CLAUDE.md

@@ -217,6 +217,34 @@ persistence) → GUI (Qt widgets that read the manager and connect to its signal
   it never belongs in git). Audio under ~3 s has no fingerprint at all
   ("Empty fingerprint"), which is a reported failure, not a crash.
 
+- **`lintunes/filename_tags.py`** — the offline half of Identify Track, and the
+  answer to its biggest limitation: **AcoustID only knows music somebody
+  submitted**, so an underground/SoundCloud rip fingerprints perfectly and
+  matches *nothing* (verified: "B. Clem — Zuuso" returns zero results from
+  AcoustID, zero from a MusicBrainz text search, zero from iTunes). Its
+  filename, though, says exactly what it is. Pure string work, no network, no
+  Qt: strips yt-dlp's trailing id (`[1025657891]`, `-WC7gK2kgyTQ` — an 11-char
+  token is only treated as a YouTube id if it carries a digit, an underscore,
+  or repeatedly flipping case, or "Underground" would be eaten), drops
+  `(Official Video)`-style noise, splits `Artist - Title` on a *spaced* hyphen
+  only (so "Jay-Z" and "350-440-DialTone" survive), collapses a doubled
+  uploader, reads a leading `1-04`, restores `_s`→`'s`, and falls back to the
+  iTunes tree (`<root>/Artist/Album/NN Title.ext`), ignoring placeholder dirs
+  like "Unknown Artist". `fingerprint.candidate_from_filename` wraps it as a
+  candidate with `source="filename"`, appended to every lookup and standing
+  alone when there are no matches — the dialog then says where the guess came
+  from instead of quoting a fabricated confidence. The same filename also
+  feeds `hint_for(track)`, which ranks real lookup results: token overlap
+  (words worth double, bare numbers single — "alhambra" identifies a release,
+  "1961" appears in every compilation spanning it), a coarse **duration**
+  bucket off the recording lengths AcoustID returns (which is what separates a
+  150 s studio take from a 155 s live one), and a year named in the filename
+  that predates anything the database knows. That last one is not a nicety:
+  MusicBrainz's own `first-release-date` for "Ahmad Jamal's Alhambra" is
+  **2002**, because its three original 1961 pressings are in the database
+  undated — so a second MusicBrainz call would return the same wrong year, and
+  the file's own name is the only place 1961 exists.
+
 - **`lintunes/mpris.py`** — registers `org.mpris.MediaPlayer2.lintunes` over D-Bus
   so the desktop's media keys / now-playing popup control playback. Spacebar and
   arrow keys are handled locally via `MainWindow.eventFilter`.

TASKS.md

@@ -9,6 +9,32 @@ When a round closes, move its finished items to `tasks-done.md`.
 
 - [ ] archive the done tasks in here to another file, this is crufty....
 
+## Round 46 (2026-09-02) — The file already told you: done, see tasks-done.md
+
+Filename-derived proposals when no database knows the song, plus using the
+file's own tags/name/duration to rank real lookup results. Left on the table:
+
+- [ ] **Submit fingerprints back to AcoustID.** The Zuuso case is unfixable by
+      lookup because nobody ever submitted it — and LinTunes is holding a good
+      fingerprint plus (after an identify) confirmed tags. `/v2/submit` takes
+      exactly that. It would make the library better for everyone and fix
+      trav's own second machine. Needs a user API key (not just the app key)
+      and must be explicitly opt-in — never submit tags the user hasn't
+      confirmed.
+- [ ] **An LLM pass for the genuinely ambiguous ones.** Raised by trav
+      2026-09-02. Worth being precise about where it helps: *not* the Zuuso
+      case (no data exists to reason over) and *not* the Alhambra year (1961
+      isn't in the response either). It helps where there are many plausible
+      candidates and the signal is semantic — an odd filename convention, or
+      deciding whether a number is a performance year or a release year. Would
+      be the app's first LLM dependency; keep it behind the same
+      "nothing is written until accepted" gate.
+- [ ] **A "clean up this filename" action independent of identify.** Half of
+      what round 46 built is useful with no network at all: 474 files in
+      Unknown Artist/Unknown Album could have artist/title/track filled from
+      their names in one pass. Wants the worker+progress split, not a dialog
+      per track.
+
 ## Round 45 (2026-09-02) — Ask the song what it is: done, see tasks-done.md
 
 Right-click → Identify Track…: Chromaprint fingerprint + AcoustID lookup, a

lintunes/__init__.py

@@ -1,3 +1,3 @@
 """LinTunes — iTunes-style music library manager and player for Linux."""
 
-__version__ = "0.15.0"
+__version__ = "0.16.0"

lintunes/filename_tags.py

@@ -0,0 +1,164 @@
+"""Reading tags out of a file's own name, for when no database knows the song.
+
+Fingerprinting answers "who is this?" only for music somebody has already
+submitted to AcoustID. A SoundCloud rip of an underground artist isn't in
+AcoustID, isn't in MusicBrainz and isn't in iTunes — but its filename usually
+says exactly what it is ("B. Clem - Zuuso [1025657891]"), because that's how
+yt-dlp and friends name what they download.
+
+So this module is the offline half of Identify Track: pure string work, no
+network, no Qt. It strips the download id and the "(Official Video)" noise,
+splits "Artist - Title", and falls back to the iTunes-style folder tree
+(<root>/Artist/Album/NN Title.ext) for anything the name itself doesn't say.
+
+Everything here is a *guess* and is presented as one — the dialog labels these
+proposals "from filename" rather than quoting a match confidence.
+"""
+import re
+from pathlib import Path
+
+
+# yt-dlp appends the source's id: "[1025657891]" (SoundCloud, numeric) or
+# "-WC7gK2kgyTQ" / "[EtNZnhxWLHo]" (YouTube, 11 chars of base64url).
+_YT_ID_CHARS = re.compile(r"^[A-Za-z0-9_-]{11}$")
+_NUMERIC_ID = re.compile(r"[\[\(]\s*\d{6,}\s*[\]\)]\s*$|[-_]\d{6,}\s*$")
+_BRACKET_ID = re.compile(r"[\[\(]\s*([A-Za-z0-9_-]{11})\s*[\]\)]\s*$")
+_TRAILING_ID = re.compile(r"-([A-Za-z0-9_-]{11})\s*$")
+
+# "(Official Video)", "[HD]", "(mp3)" and friends — noise, never a title.
+_NOISE = re.compile(
+    r"\s*[\(\[]\s*(?:"
+    # an optional "official", an optional qualifier, then video/audio —
+    # covers "Official Video", "Official Lyric Video", "Music Video", "Audio"
+    r"(?:official\s*)?(?:music\s*|lyrics?\s*|full\s*)?(?:video|audio)|"
+    r"official|lyrics?|visuali[sz]er|"
+    r"hd|hq|4k|full\s*album|free\s*download|out\s*now|mp3|m4a|flac"
+    r")\s*[\)\]]", re.I)
+
+# "02 Title", "1-01 Title", "03. Title"
+_LEADING_NUMBER = re.compile(r"^\s*(?:(\d{1,2})[-_ ]+)?(\d{1,3})[\.\)]?\s+(?=\S)")
+
+# A bare four-digit year. Only trusted when the text holds exactly one, so a
+# range ("1958-62", "Tour 1958-1961") is never mistaken for a release year.
+_YEAR = re.compile(r"(?<!\d)(19\d{2}|20\d{2})(?!\d)")
+
+# Filesystem-safe stand-ins for an apostrophe: "That_s So Sick!"
+_APOSTROPHE = re.compile(r"_(s|t|re|ve|ll|d|m)\b", re.I)
+
+_UNKNOWN_DIRS = {"unknown artist", "unknown album", "various artists",
+                 "music", "downloads", "unknown", ""}
+
+
+def _looks_like_a_youtube_id(token: str) -> bool:
+    """An 11-char base64url token that reads as random rather than as a word.
+
+    A real word can be 11 characters ("Underground"), so length alone would
+    eat legitimate titles. Random ids carry a digit, an underscore, or case
+    that flips repeatedly — an English word does none of those.
+    """
+    if not _YT_ID_CHARS.match(token):
+        return False
+    if any(c.isdigit() for c in token) or "_" in token:
+        return True
+    transitions = sum(
+        1 for a, b in zip(token, token[1:])
+        if a.isalpha() and b.isalpha() and a.isupper() != b.isupper())
+    return transitions >= 3
+
+
+def strip_download_id(stem: str) -> str:
+    """Remove a trailing yt-dlp source id, in brackets or hyphen-attached."""
+    text = _NUMERIC_ID.sub("", stem).strip()
+    match = _BRACKET_ID.search(text)
+    if match and _looks_like_a_youtube_id(match.group(1)):
+        return text[:match.start()].strip()
+    match = _TRAILING_ID.search(text)
+    if match and _looks_like_a_youtube_id(match.group(1)):
+        return text[:match.start()].strip()
+    return text
+
+
+def strip_noise(text: str) -> str:
+    """Drop "(Official Video)"-style decorations."""
+    previous = None
+    while previous != text:
+        previous = text
+        text = _NOISE.sub("", text).strip()
+    return text.strip(" -_·—–").strip()
+
+
+def restore_apostrophes(text: str) -> str:
+    return _APOSTROPHE.sub(r"'\1", text)
+
+
+def leading_track_number(text: str) -> tuple[str, int | None, int | None]:
+    """Split "1-01 Title" into (title, track number, disc number)."""
+    match = _LEADING_NUMBER.match(text)
+    if not match:
+        return text, None, None
+    disc, number = match.group(1), match.group(2)
+    rest = text[match.end():].strip()
+    if not rest:
+        return text, None, None
+    return (rest, int(number), int(disc) if disc else None)
+
+
+def year_in(text: str) -> int | None:
+    """The year a string names, when it names exactly one."""
+    years = set(_YEAR.findall(text or ""))
+    if len(years) != 1:
+        return None
+    return int(next(iter(years)))
+
+
+def split_artist_title(text: str) -> tuple[str | None, str]:
+    """Split "Artist - Title" on the first hyphen that has space around it.
+
+    A hyphen inside a word ("Jay-Z", "re-recorded") is never a separator, and
+    a doubled artist ("DJ X - DJ X - Track") collapses to one.
+    """
+    parts = [p.strip() for p in re.split(r"\s+[-–—]\s+", text) if p.strip()]
+    if len(parts) < 2:
+        return None, text.strip()
+    # yt-dlp sometimes repeats the uploader: keep one copy.
+    while len(parts) > 2 and parts[0].casefold() == parts[1].casefold():
+        parts.pop(0)
+    artist = parts[0]
+    title = " - ".join(parts[1:]) if len(parts) > 2 else parts[1]
+    return artist, title
+
+
+def parse_filename(location: str) -> dict:
+    """Best-effort tags from a path. Keys are LinTunes field names; absent
+    keys mean "the name doesn't say"."""
+    path = Path(location)
+    stem = strip_download_id(path.stem)
+    stem = strip_noise(stem)
+    stem, number, disc = leading_track_number(stem)
+    artist, title = split_artist_title(stem)
+    title = restore_apostrophes(strip_noise(title)).strip()
+
+    fields: dict = {}
+    if title:
+        fields["name"] = title
+    if artist:
+        fields["artist"] = restore_apostrophes(artist).strip()
+    if number is not None:
+        fields["track_number"] = number
+    if disc is not None:
+        fields["disc_number"] = disc
+
+    # The iTunes-style tree fills in what the name itself doesn't say:
+    # <root>/Artist/Album/NN Title.ext
+    parents = [p.name for p in path.parents]
+    if len(parents) >= 2:
+        album_dir, artist_dir = parents[0], parents[1]
+        if album_dir.casefold() not in _UNKNOWN_DIRS:
+            fields.setdefault("album", album_dir)
+        if artist_dir.casefold() not in _UNKNOWN_DIRS:
+            fields.setdefault("artist", artist_dir)
+
+    year = year_in(path.stem)
+    if year is not None:
+        fields["year"] = year
+    return fields

lintunes/fingerprint.py

@@ -12,13 +12,17 @@ AcoustID application key is user-supplied via Preferences (the Last.fm
 precedent); registration is free at https://acoustid.org/new-application.
 """
 import json
+import re
 import shutil
 import subprocess
 import threading
 from dataclasses import asdict, dataclass
+from pathlib import Path
 
 from PyQt6.QtCore import QObject, pyqtSignal
 
+from lintunes import filename_tags
+
 
 ACOUSTID_LOOKUP_URL = "https://api.acoustid.org/v2/lookup"
 LOOKUP_META = "recordings releasegroups releases tracks compress"
@@ -33,6 +37,10 @@ class IdentifyCandidate:
     """One proposed identification. Field names match Track/tagging so
     ``fields()`` can go straight into ``LibraryManager.edit_track_fields``."""
     score: float
+    # "acoustid" (a fingerprint match, so ``score`` is a real confidence) or
+    # "filename" (a guess read off the file's own name — no confidence to
+    # quote, and the dialog says so rather than implying one).
+    source: str = "acoustid"
     name: str | None = None
     artist: str | None = None
     album_artist: str | None = None
@@ -45,7 +53,7 @@ class IdentifyCandidate:
     def fields(self) -> dict:
         """Proposed edits keyed by LinTunes field names, Nones omitted."""
         return {k: v for k, v in asdict(self).items()
-                if k != "score" and v is not None}
+                if k not in ("score", "source") and v is not None}
 
 
 # ---- pure parsing/ranking helpers (offline-testable) ----
@@ -78,18 +86,75 @@ def _oldest_year(releasegroup: dict) -> int | None:
     return min(years) if years else None
 
 
+# Secondary types that mean "this is not the album the song came from".
+# Live is deliberately absent: a live album is a real album, and demoting it
+# buried "Ahmad Jamal's Alhambra" under a compilation for a file whose own
+# name said "Live At The Alhambra".
+_DEMOTED_SECONDARY = {"compilation", "interview", "remix", "dj-mix",
+                      "mixtape/street", "demo", "audiobook"}
+
+
 def _releasegroup_sort_key(releasegroup: dict) -> tuple:
-    """Rank a release group for the default proposal: the earliest proper
-    studio Album first, then EPs/Singles, then Compilations/Live/etc."""
+    """Rank a release group for the default proposal: a proper album first,
+    then EPs/Singles, with compilations and the like pushed down."""
     rg_type = releasegroup.get("type") or ""
     secondary = releasegroup.get("secondarytypes") or []
-    if not secondary:
-        rank = {"Album": 0, "EP": 1, "Single": 2}.get(rg_type, 4)
-    else:
-        rank = 3 if rg_type == "Album" else 4
+    rank = {"Album": 0, "EP": 1, "Single": 2}.get(rg_type, 4)
+    if any((s or "").casefold() in _DEMOTED_SECONDARY for s in secondary):
+        rank += 3
     return (rank, _oldest_year(releasegroup) or 9999)
 
 
+_TOKEN = re.compile(r"[^\W\d_]+|\d+", re.UNICODE)
+_STOPWORDS = {"the", "a", "an", "of", "and", "in", "at", "on", "live",
+              "feat", "ft", "remastered", "version", "mix", "edit"}
+
+
+def _tokens(text: str) -> set:
+    """Comparable words from a title/album/filename."""
+    return {t.casefold() for t in _TOKEN.findall(text or "")
+            if len(t) > 1 and t.casefold() not in _STOPWORDS}
+
+
+def _duration_bucket(recording: dict, duration: float | None) -> int:
+    """How well a recording's length matches the file's, coarsely.
+
+    AcoustID often returns several recordings of the same song — a studio
+    take, a live take, an edit — and their lengths are what tell them apart.
+    Buckets rather than raw difference so a second or two never outweighs the
+    other evidence, but a clearly different take sinks.
+    """
+    other = recording.get("duration")
+    if duration is None or not isinstance(other, (int, float)):
+        return 1  # unknown: behind an exact match, ahead of a bad one
+    diff = abs(other - duration)
+    if diff <= 1.5:
+        return 0
+    if diff <= 4:
+        return 1
+    if diff <= 12:
+        return 2
+    return 3
+
+
+def _hint_overlap(hint_tokens: set, candidate: IdentifyCandidate) -> int:
+    """How much a candidate echoes what the file already claims to be.
+
+    The distinctive words in "Snowfall (Live At The Alhambra_1961)" —
+    *snowfall*, *alhambra*, *1961* — are the best evidence available about
+    which of eight plausible releases this actually is, and ignoring them was
+    why a compilation could outrank the album named in the filename.
+    """
+    if not hint_tokens:
+        return 0
+    text = " ".join(p for p in (candidate.name, candidate.artist,
+                                candidate.album_artist, candidate.album) if p)
+    # A word like "alhambra" names a release; a bare number like "1961" turns
+    # up in every compilation spanning that year, so it counts for less.
+    return sum(2 if not token.isdigit() else 1
+               for token in hint_tokens & _tokens(text))
+
+
 def _pick_release(releasegroup: dict) -> dict | None:
     """Earliest dated release that carries mediums (track positions); falls
     back to the earliest dated one, then the first."""
@@ -126,7 +191,8 @@ def _track_position(release: dict | None) -> tuple[int | None, int | None,
 
 
 def parse_lookup(payload: dict, threshold: float = SCORE_THRESHOLD,
-                 limit: int = MAX_CANDIDATES) -> list[IdentifyCandidate]:
+                 limit: int = MAX_CANDIDATES, hint: str = "",
+                 duration: float | None = None) -> list[IdentifyCandidate]:
     """Turn an AcoustID lookup response into ranked candidates.
 
     One candidate per release group of each matched recording, so the
@@ -135,16 +201,23 @@ def parse_lookup(payload: dict, threshold: float = SCORE_THRESHOLD,
     the minimum across all its release groups' releases — even when the
     proposed album is a later compilation: years sort by when the song first
     came out, not by which pressing this file happens to be from.
+
+    ``hint`` is whatever the file already claims to be (its tags and its
+    filename). It never invents a candidate, but it breaks ties: a release
+    whose words echo the hint outranks one that doesn't, and a year named in
+    the hint wins when it predates anything the database knows — MusicBrainz
+    often holds only a reissue's date, with the original pressing undated.
     """
-    candidates = []
+    hint_tokens = _tokens(hint)
+    hint_year = _single_year(hint)
+    scored = []
     seen = set()
-    results = sorted(payload.get("results", []),
-                     key=lambda r: r.get("score", 0), reverse=True)
-    for result in results:
+    for result in payload.get("results", []):
         score = result.get("score", 0)
         if score < threshold:
             continue
         for recording in result.get("recordings", []):
+            length = _duration_bucket(recording, duration)
             title = recording.get("title")
             artist = _join_artists(recording.get("artists"))
             groups = sorted(recording.get("releasegroups") or [],
@@ -152,10 +225,16 @@ def parse_lookup(payload: dict, threshold: float = SCORE_THRESHOLD,
             original_years = [y for rg in groups
                               if (y := _oldest_year(rg)) is not None]
             original_year = min(original_years) if original_years else None
+            if hint_year is not None and (original_year is None
+                                          or hint_year < original_year):
+                original_year = hint_year
             if not groups:
                 if title:
-                    candidates.append(IdentifyCandidate(
-                        score=score, name=title, artist=artist))
+                    bare = IdentifyCandidate(score=score, name=title,
+                                             artist=artist, year=hint_year)
+                    scored.append(((-round(score, 2), length,
+                                    -_hint_overlap(hint_tokens, bare),
+                                    4, 9999), bare))
                 continue
             for releasegroup in groups:
                 album = releasegroup.get("title")
@@ -165,7 +244,7 @@ def parse_lookup(payload: dict, threshold: float = SCORE_THRESHOLD,
                 seen.add(key)
                 number, count, disc = _track_position(
                     _pick_release(releasegroup))
-                candidates.append(IdentifyCandidate(
+                candidate = IdentifyCandidate(
                     score=score,
                     name=title,
                     artist=artist,
@@ -175,8 +254,47 @@ def parse_lookup(payload: dict, threshold: float = SCORE_THRESHOLD,
                     track_number=number,
                     track_count=count,
                     disc_number=disc,
-                ))
-    return candidates[:limit]
+                )
+                rank, rg_year = _releasegroup_sort_key(releasegroup)
+                scored.append(((-round(score, 2), length,
+                                -_hint_overlap(hint_tokens, candidate),
+                                rank, rg_year), candidate))
+    scored.sort(key=lambda pair: pair[0])
+    return [candidate for _key, candidate in scored][:limit]
+
+
+def _single_year(text: str) -> int | None:
+    """The year a string names, when it names exactly one (so a range like
+    "1958-62" is never mistaken for a release year)."""
+    return filename_tags.year_in(text or "")
+
+
+def hint_for(track) -> str:
+    """What a track already claims to be — its tags plus its filename. Used
+    to rank lookup results, and to fall back on when nothing matches."""
+    parts = [track.name or "", track.artist or "", track.album or ""]
+    if track.location:
+        parts.append(Path(track.location).stem)
+    return " ".join(p for p in parts if p)
+
+
+def candidate_from_filename(track) -> IdentifyCandidate | None:
+    """A proposal read off the file's own name, for the very common case that
+    no database has ever heard of the song.
+
+    Underground and self-released music simply isn't in AcoustID — the
+    fingerprint is fine, there is just nothing to match it against — but the
+    download named the file "Artist - Title [id]", which is how a human reads
+    it too. Returns None when the name yields nothing usable.
+    """
+    if not track.location:
+        return None
+    fields = filename_tags.parse_filename(track.location)
+    candidate = IdentifyCandidate(score=0.0, source="filename")
+    for key, value in fields.items():
+        if hasattr(candidate, key):
+            setattr(candidate, key, value)
+    return candidate if candidate.fields() else None
 
 
 # ---- subprocess + network ----
@@ -209,8 +327,8 @@ def fingerprint_file(location: str) -> tuple[int, str]:
         raise OSError(f"couldn't read fpcalc output for {name}") from e
 
 
-def lookup_fingerprint(api_key: str, duration: int,
-                       fingerprint: str) -> list[IdentifyCandidate]:
+def lookup_fingerprint(api_key: str, duration: int, fingerprint: str,
+                       hint: str = "") -> list[IdentifyCandidate]:
     import requests
     response = requests.post(
         ACOUSTID_LOOKUP_URL,
@@ -224,25 +342,41 @@ def lookup_fingerprint(api_key: str, duration: int,
     if payload.get("status") != "ok":
         message = (payload.get("error") or {}).get("message", "lookup failed")
         raise RuntimeError(message)
-    return parse_lookup(payload)
+    return parse_lookup(payload, hint=hint, duration=duration)
 
 
 class TrackIdentifier(QObject):
     """One fingerprint+lookup running off the GUI thread.
 
     ``finished`` carries {"track_id": int, "candidates": [IdentifyCandidate]}
-    or {"track_id": int, "error": str}.
+    or {"track_id": int, "error": str}. A filename guess is appended to the
+    fingerprint's matches, and stands alone when there are none — a lookup
+    that found nothing is not the same as nothing to propose.
     """
 
     finished = pyqtSignal(object)
 
-    def identify(self, track_id: int, location: str, api_key: str):
+    def identify(self, track, api_key: str):
+        track_id = track.track_id
+        location = track.location
+        hint = hint_for(track)
+        guess = candidate_from_filename(track)
+
         def work():
             try:
                 duration, fp = fingerprint_file(location)
-                candidates = lookup_fingerprint(api_key, duration, fp)
-                self.finished.emit(
-                    {"track_id": track_id, "candidates": candidates})
+                candidates = lookup_fingerprint(api_key, duration, fp, hint)
             except Exception as e:
-                self.finished.emit({"track_id": track_id, "error": str(e)})
+                # A file too short or too odd to fingerprint still has a name.
+                if guess is not None:
+                    self.finished.emit({"track_id": track_id,
+                                        "candidates": [guess],
+                                        "note": str(e)})
+                else:
+                    self.finished.emit({"track_id": track_id, "error": str(e)})
+                return
+            if guess is not None:
+                candidates = candidates + [guess]
+            self.finished.emit(
+                {"track_id": track_id, "candidates": candidates})
         threading.Thread(target=work, daemon=True).start()

lintunes/gui/identify_dialog.py

@@ -35,19 +35,20 @@ class IdentifyDialog(QDialog):
         self.cancel_all = False
 
         layout = QVBoxLayout(self)
-        best = candidates[0]
-        header = QLabel(f"Identified “{track.name or '(untitled)'}” — "
-                        f"match confidence {round(best.score * 100)}%")
-        header.setWordWrap(True)
-        layout.addWidget(header)
+        self._header = QLabel()
+        self._header.setWordWrap(True)
+        layout.addWidget(self._header)
 
         # All alternates at a glance — a candidate summarizes to one line, so
         # a dropdown beats AlbumArtDialog's blind Next-clicking.
         self._combo = QComboBox()
         for c in candidates:
             year = f" ({c.year})" if c.year else ""
-            self._combo.addItem(" — ".join(
-                p for p in (c.name, c.artist, c.album) if p) + year)
+            summary = " — ".join(
+                p for p in (c.name, c.artist, c.album) if p) + year
+            if c.source == "filename":
+                summary = f"From the filename: {summary}"
+            self._combo.addItem(summary)
         layout.addWidget(self._combo)
 
         grid = QGridLayout()
@@ -105,6 +106,17 @@ class IdentifyDialog(QDialog):
         iff the candidate proposes a value that differs from the current one;
         rows with nothing proposed are unchecked and disabled."""
         candidate = self._candidates[index]
+        name = self._track.name or "(untitled)"
+        if candidate.source == "filename":
+            # No database matched this, so there is no confidence to quote —
+            # say where the proposal actually came from instead of implying
+            # something recognized the audio.
+            self._header.setText(
+                f"No database knows “{name}”. Read from the file's name:")
+        else:
+            self._header.setText(
+                f"Identified “{name}” — match confidence "
+                f"{round(candidate.score * 100)}%")
         for field, _label, is_number, _max in _FIELDS:
             current = self._current_value(field)
             proposed = getattr(candidate, field)

lintunes/gui/main_window.py

@@ -1138,7 +1138,7 @@ class MainWindow(QMainWindow):
         self._identifier = TrackIdentifier()
         self._identifier.finished.connect(
             lambda result, t=track, k=api_key: self._on_identified(t, k, result))
-        self._identifier.identify(track.track_id, track.location, api_key)
+        self._identifier.identify(track, api_key)
 
     def _on_identified(self, track, api_key: str, result: dict):
         if "error" in result:
@@ -1146,7 +1146,8 @@ class MainWindow(QMainWindow):
                 f"Couldn't identify “{track.name}”: {result['error']}", 6000)
         elif not result["candidates"]:
             self.statusBar().showMessage(
-                f"No match found for “{track.name}”", 6000)
+                f"Nothing to propose for “{track.name}” — no database match "
+                f"and the filename says nothing", 6000)
         else:
             self.statusBar().clearMessage()
             dialog = IdentifyDialog(track, result["candidates"],

tasks-done.md

@@ -1,5 +1,51 @@
 ## Done
 
+### Round 46 (2026-09-02) — The file already told you (v0.16.0)
+
+Round 45 shipped and trav broke it in two places within the hour, both real.
+"B. Clem - Zuuso [1025657891]" fingerprinted fine and matched **nothing**, and
+"Snowfall (Live At The Alhambra_1961)" matched but proposed a 2002 reissue year
+and ranked a compilation above the album its own filename named.
+
+- [x] **The limit is the database, not the fingerprint.** Verified before
+      building anything: that Zuuso fingerprint returns zero AcoustID results,
+      a MusicBrainz text search for "Zuuso" returns zero, and iTunes returns
+      zero (and irrelevant fuzz for "B. Clem"). The track exists in no metadata
+      service on earth. So no smarter lookup — LLM or otherwise — could have
+      answered it, because there was nothing to answer with.
+- [x] **But the filename knew.** New `lintunes/filename_tags.py` reads tags out
+      of the name yt-dlp gave the file. "B. Clem - Zuuso [1025657891]" → artist
+      "B. Clem", name "Zuuso", which is exactly what trav said it was. Checked
+      against all 474 files in the untagged folder, not against invented
+      examples.
+- [x] **Careful about what it strips.** An 11-character trailing token is only
+      a YouTube id if it carries a digit, an underscore, or case that flips
+      repeatedly — otherwise "Cold Draft-Underground" loses a word. A hyphen
+      only splits artist from title when it has spaces around it, so "Jay-Z"
+      and "350-440-DialTone" survive. A 4-digit year is never a numeric id.
+- [x] **A guess is labelled a guess.** `IdentifyCandidate.source` is
+      "acoustid" or "filename"; the dialog quotes a match confidence only for
+      the former and says "No database knows this — read from the file's name"
+      for the latter, rather than inventing a percentage.
+- [x] **The file now ranks the results too.** `hint_for(track)` feeds tag +
+      filename tokens into `parse_lookup`: overlap scoring (words count double,
+      bare numbers single — "alhambra" identifies a release, "1961" appears in
+      every compilation spanning it), a coarse duration bucket against the
+      recording lengths AcoustID returns, and a filename year that predates
+      what the database knows. Alhambra now ranks the right album first with
+      year 1961 instead of a compilation with 2002.
+- [x] **Live is no longer a demerit.** It was lumped in with Compilation, which
+      buried a live album for a file whose name said "Live At The Alhambra".
+      Only compilations, remixes, interviews and the like are demoted now.
+- [x] **Why the year needed the filename at all:** MusicBrainz's own
+      `first-release-date` for "Ahmad Jamal's Alhambra" is *2002* — its three
+      original 1961 pressings are in the database undated. A second MusicBrainz
+      call, the obvious fix, returns the same wrong answer. 1961 exists only in
+      trav's filename.
+- [x] Not done, deliberately: trav rejected "never propose a shorter title" —
+      truncating is *wanted*, because the garbage in a title is usually the
+      part being dropped.
+
 ### Round 45 (2026-09-02) — Ask the song what it is (v0.15.0)
 
 Some files arrive with the title right and everything else missing or wrong,

tests/test_round46.py

@@ -0,0 +1,227 @@
+"""Round 46 — Identify Track stops giving up when no database knows the song.
+
+Two real failures from trav's library drove this. "B. Clem - Zuuso
+[1025657891]" fingerprinted fine and returned *zero* AcoustID results — the
+track is underground, so it is in no database at all, and no amount of
+cleverness can look up what does not exist. But its filename said exactly what
+it was, which is how trav knew. And "Snowfall (Live At The Alhambra_1961)"
+matched, yet proposed a 2002 reissue year and ranked a compilation above the
+album the filename named, because nothing compared the results against what
+the file already said.
+
+So the file itself is now evidence: it seeds a proposal when the lookup finds
+nothing, and it ranks the lookup's results when it finds several.
+"""
+import pytest
+
+from lintunes.fingerprint import (
+    IdentifyCandidate, candidate_from_filename, hint_for, parse_lookup,
+)
+from lintunes.filename_tags import (
+    leading_track_number, parse_filename, split_artist_title,
+    strip_download_id, strip_noise, year_in,
+)
+
+
+class _Track:
+    def __init__(self, name="", artist="", album="", location=""):
+        self.track_id = 1
+        self.name = name
+        self.artist = artist
+        self.album = album
+        self.location = location
+
+
+UNKNOWN = "/m/Unknown Artist/Unknown Album/"
+
+
+class TestStrippingDownloadIds:
+    @pytest.mark.parametrize("stem, expected", [
+        ("B. Clem - Zuuso [1025657891]", "B. Clem - Zuuso"),
+        ("Stardust-321895497", "Stardust"),
+        ("Cold Draft-WC7gK2kgyTQ", "Cold Draft"),
+        ("Self Esteem [EtNZnhxWLHo]", "Self Esteem"),
+        ("Caught Up in the Rapture [Oz-b86LZ21c]", "Caught Up in the Rapture"),
+    ])
+    def test_yt_dlp_ids_are_removed(self, stem, expected):
+        assert strip_download_id(stem) == expected
+
+    def test_a_real_word_is_not_mistaken_for_a_youtube_id(self):
+        """"Underground" is 11 characters too — length alone can't decide."""
+        assert strip_download_id("Cold Draft-Underground") == \
+            "Cold Draft-Underground"
+
+    def test_a_year_is_not_mistaken_for_a_numeric_id(self):
+        assert strip_download_id("Cross Country Tour 1958-1961") == \
+            "Cross Country Tour 1958-1961"
+
+
+class TestStrippingNoise:
+    @pytest.mark.parametrize("text", [
+        "Dance II (Official Video)",
+        "Dance II (Official Music Video)",
+        "Dance II (Official Lyric Video)",
+        "Dance II [HD]",
+        "Dance II (Audio)",
+    ])
+    def test_decorations_go(self, text):
+        assert strip_noise(text) == "Dance II"
+
+    def test_a_real_parenthetical_stays(self):
+        assert strip_noise("Take Off (ASOT 1159)") == "Take Off (ASOT 1159)"
+
+
+class TestSplittingAndNumbering:
+    def test_artist_title_split(self):
+        assert split_artist_title("B. Clem - Zuuso") == ("B. Clem", "Zuuso")
+
+    def test_hyphen_inside_a_word_is_not_a_separator(self):
+        assert split_artist_title("350-440-DialTone") == \
+            (None, "350-440-DialTone")
+
+    def test_a_repeated_uploader_collapses(self):
+        assert split_artist_title("Sean Dream - Sean Dream - tomorrow111") == \
+            ("Sean Dream", "tomorrow111")
+
+    def test_leading_track_and_disc_numbers(self):
+        assert leading_track_number("1-04 Sitting Back") == \
+            ("Sitting Back", 4, 1)
+        assert leading_track_number("02 nashville cats") == \
+            ("nashville cats", 2, None)
+
+    def test_a_number_that_is_the_whole_title_is_kept(self):
+        assert leading_track_number("1979") == ("1979", None, None)
+
+
+class TestYearMining:
+    def test_a_single_year_is_read(self):
+        assert year_in("Snowfall (Live At The Alhambra_1961)") == 1961
+
+    def test_a_range_is_refused(self):
+        """"1958-62" and "1958-1961" name a span, not a release year."""
+        assert year_in("Cross Country Tour 1958-1961") is None
+
+    def test_no_year_is_none(self):
+        assert year_in("Zuuso") is None
+
+
+class TestFilenameProposal:
+    def test_the_zuuso_case(self):
+        """The track that started this: fingerprinted fine, matched nothing,
+        and its name said exactly what trav already knew it was."""
+        track = _Track(name="B. Clem - Zuuso [1025657891]",
+                       location=UNKNOWN + "B. Clem - Zuuso [1025657891].mp3")
+        candidate = candidate_from_filename(track)
+        assert candidate.source == "filename"
+        assert candidate.fields() == {"name": "Zuuso", "artist": "B. Clem"}
+
+    def test_apostrophes_come_back(self):
+        track = _Track(location=UNKNOWN + "cehryl - That_s So Sick! [738273660].mp3")
+        assert candidate_from_filename(track).name == "That's So Sick!"
+
+    def test_the_folder_tree_fills_what_the_name_omits(self):
+        track = _Track(location="/m/Onra/Long Distance/1-04 Sitting Back.mp3")
+        fields = candidate_from_filename(track).fields()
+        assert fields["artist"] == "Onra"
+        assert fields["album"] == "Long Distance"
+        assert fields["name"] == "Sitting Back"
+        assert fields["track_number"] == 4
+
+    def test_placeholder_folders_are_not_treated_as_tags(self):
+        track = _Track(location=UNKNOWN + "blurf.mp3")
+        fields = candidate_from_filename(track).fields()
+        assert "artist" not in fields and "album" not in fields
+
+    def test_a_track_with_no_file_proposes_nothing(self):
+        assert candidate_from_filename(_Track()) is None
+
+
+def _payload(*recordings, score=0.9):
+    return {"status": "ok",
+            "results": [{"score": score, "recordings": list(recordings)}]}
+
+
+def _recording(title, duration=None, groups=()):
+    rec = {"title": title, "artists": [{"name": "The Band"}],
+           "releasegroups": list(groups)}
+    if duration is not None:
+        rec["duration"] = duration
+    return rec
+
+
+def _group(title, year, type_="Album", secondary=None):
+    return {"title": title, "type": type_,
+            "secondarytypes": list(secondary or []),
+            "releases": [{"date": {"year": year} if year else {}}]}
+
+
+class TestRankingAgainstTheFile:
+    def test_the_album_the_filename_names_wins(self):
+        payload = _payload(_recording("Snowfall", groups=[
+            _group("Some Other Record", 1970),
+            _group("Ahmad Jamal's Alhambra", 2002, secondary=["Live"]),
+        ]))
+        hint = "Snowfall (Live At The Alhambra_1961)"
+        assert parse_lookup(payload, hint=hint)[0].album == \
+            "Ahmad Jamal's Alhambra"
+
+    def test_a_live_album_is_no_longer_demoted(self):
+        """Live is what a live album *is*; only compilations get pushed down."""
+        payload = _payload(_recording("Snowfall", groups=[
+            _group("Greatest Hits", 1990, secondary=["Compilation"]),
+            _group("The Live Album", 1995, secondary=["Live"]),
+        ]))
+        assert parse_lookup(payload)[0].album == "The Live Album"
+
+    def test_duration_picks_the_right_take(self):
+        """Two recordings of one song; only the length tells them apart."""
+        payload = _payload(
+            _recording("Snowfall", duration=155.0,
+                       groups=[_group("The Long Take", 2000)]),
+            _recording("Snowfall", duration=150.3,
+                       groups=[_group("The Short Take", 2000)]),
+        )
+        assert parse_lookup(payload, duration=150)[0].album == "The Short Take"
+
+    def test_a_year_in_the_filename_beats_a_reissue_date(self):
+        """MusicBrainz often has only the CD reissue dated, with the original
+        pressing undated — so the file's own 1961 is the better evidence."""
+        payload = _payload(_recording("Snowfall", groups=[
+            _group("Ahmad Jamal's Alhambra", 2002)]))
+        hint = "Snowfall (Live At The Alhambra_1961)"
+        assert parse_lookup(payload, hint=hint)[0].year == 1961
+
+    def test_a_later_filename_year_never_overrides_an_earlier_one(self):
+        payload = _payload(_recording("Song", groups=[_group("Album", 1967)]))
+        assert parse_lookup(payload, hint="Song (remaster 1999)")[0].year == 1967
+
+    def test_hint_never_invents_a_candidate(self):
+        assert parse_lookup({"status": "ok", "results": []},
+                            hint="anything at all") == []
+
+
+class TestHint:
+    def test_hint_gathers_tags_and_filename(self):
+        track = _Track(name="Snowfall", artist="Ahmad Jamal",
+                       location=UNKNOWN + "Snowfall (Live At The Alhambra_1961).mp3")
+        hint = hint_for(track)
+        assert "Snowfall" in hint and "Ahmad Jamal" in hint and "1961" in hint
+
+
+class TestDialogLabelsAGuessHonestly:
+    def test_filename_candidates_say_where_they_came_from(self, qapp):
+        from lintunes.gui.identify_dialog import IdentifyDialog
+        track = _Track(name="B. Clem - Zuuso [1025657891]")
+        guess = IdentifyCandidate(score=0.0, source="filename",
+                                  name="Zuuso", artist="B. Clem")
+        dialog = IdentifyDialog(track, [guess])
+        assert "file's name" in dialog._header.text().lower()
+        # No fabricated confidence percentage for something nothing matched.
+        assert "%" not in dialog._header.text()
+        assert "From the filename" in dialog._combo.itemText(0)
+
+    def test_acoustid_candidates_still_quote_confidence(self, qapp):
+        from lintunes.gui.identify_dialog import IdentifyDialog
+        candidate = IdentifyCandidate(score=0.98, name="Snowfall")
+        dialog = IdentifyDialog(_Track(name="x"), [candidate])
+        assert "98%" in dialog._header.text()