@hyperspell/openclaw-hyperspell 0.7.2 → 0.8.0

package/sommeliagent/scripts/recommend.py

@@ -1,777 +0,0 @@
-# /// script
-# requires-python = ">=3.11"
-# dependencies = ["httpx"]
-# ///
-"""
-SommeliAgent — Wine recommendation engine.
-
-Fetches Spotify listening data, maps music features to wine dimensions,
-and outputs personalized recommendations with cross-domain explanations.
-"""
-
-import argparse
-import json
-import stat
-import sys
-import time
-from dataclasses import dataclass, field, asdict
-from pathlib import Path
-
-import httpx
-
-from wine_db import WineProfile, Wine, WINE_DB  # noqa: E402
-
-CONFIG_DIR = Path.home() / ".sommeliagent"
-TOKEN_FILE = CONFIG_DIR / "token.json"
-RATINGS_FILE = CONFIG_DIR / "ratings.json"
-CACHE_FILE = CONFIG_DIR / "profile_cache.json"
-SPOTIFY_API = "https://api.spotify.com/v1"
-CACHE_TTL_SECONDS = 3600  # 1 hour
-
-
-# ──────────────────────────────────────────────
-# Types
-# ──────────────────────────────────────────────
-
-@dataclass
-class MusicProfile:
-    avg_valence: float = 0.0
-    avg_energy: float = 0.0
-    avg_danceability: float = 0.0
-    avg_acousticness: float = 0.0
-    avg_tempo: float = 0.0
-    avg_complexity: float = 0.0
-    obscurity_score: float = 0.0
-    genre_distribution: dict[str, float] = field(default_factory=dict)
-    mood_label: str = ""
-    top_artists: list[str] = field(default_factory=list)
-    top_tracks: list[str] = field(default_factory=list)
-    has_audio_features: bool = True
-
-
-@dataclass
-class MusicWineConnection:
-    music_signal: str
-    wine_signal: str
-    explanation: str
-    strength: float
-
-
-@dataclass
-class Recommendation:
-    wine: Wine
-    score: float
-    reasoning: str
-    connections: list[MusicWineConnection]
-
-
-# ──────────────────────────────────────────────
-# Spotify API (with retry + caching)
-# ──────────────────────────────────────────────
-
-def get_access_token() -> str:
-    """Get a valid Spotify access token. Delegates refresh to auth module."""
-    # Import auth module from same directory
-    sys.path.insert(0, str(Path(__file__).parent))
-    from auth import get_access_token as _refresh_token, load_token
-
-    token = _refresh_token()
-    if token:
-        return token
-
-    # Fallback: use stored access token directly (may be expired)
-    token_data = load_token()
-    if token_data and token_data.get("access_token"):
-        return token_data["access_token"]
-
-    print("ERROR: Not authenticated. Run the auth script first.", file=sys.stderr)
-    sys.exit(1)
-
-
-def spotify_get(token: str, endpoint: str, params: dict | None = None, max_retries: int = 2) -> dict:
-    """GET with retry on 429 (rate limit) and transient errors."""
-    for attempt in range(max_retries + 1):
-        resp = httpx.get(
-            f"{SPOTIFY_API}{endpoint}",
-            headers={"Authorization": f"Bearer {token}"},
-            params=params or {},
-        )
-        if resp.status_code == 401:
-            print("ERROR: Spotify token expired. Re-run auth script.", file=sys.stderr)
-            sys.exit(1)
-        if resp.status_code == 429:
-            retry_after = int(resp.headers.get("Retry-After", 2))
-            if attempt < max_retries:
-                print(f"Rate limited. Waiting {retry_after}s...", file=sys.stderr)
-                time.sleep(retry_after)
-                continue
-        if resp.status_code >= 500 and attempt < max_retries:
-            time.sleep(2 ** attempt)
-            continue
-        resp.raise_for_status()
-        return resp.json()
-    resp.raise_for_status()
-    return {}  # unreachable
-
-
-def load_cached_profile() -> MusicProfile | None:
-    if not CACHE_FILE.exists():
-        return None
-    try:
-        data = json.loads(CACHE_FILE.read_text())
-        if time.time() - data.get("_cached_at", 0) > CACHE_TTL_SECONDS:
-            return None
-        del data["_cached_at"]
-        return MusicProfile(**data)
-    except Exception:
-        return None
-
-
-def save_cached_profile(profile: MusicProfile) -> None:
-    data = asdict(profile)
-    data["_cached_at"] = time.time()
-    CONFIG_DIR.mkdir(parents=True, exist_ok=True)
-    tmp_file = CACHE_FILE.with_suffix(".tmp")
-    tmp_file.write_text(json.dumps(data, indent=2))
-    tmp_file.chmod(stat.S_IRUSR | stat.S_IWUSR)
-    tmp_file.replace(CACHE_FILE)
-
-
-def build_music_profile(token: str) -> MusicProfile:
-    # Check cache first
-    cached = load_cached_profile()
-    if cached:
-        return cached
-
-    # Fetch top tracks and artists (always available)
-    top_tracks_data = spotify_get(token, "/me/top/tracks", {"time_range": "medium_term", "limit": 50})
-    top_artists_data = spotify_get(token, "/me/top/artists", {"time_range": "medium_term", "limit": 50})
-
-    tracks = top_tracks_data.get("items", [])
-    artists = top_artists_data.get("items", [])
-
-    if not tracks:
-        print("ERROR: No listening data found. Listen to more music on Spotify first!", file=sys.stderr)
-        sys.exit(1)
-
-    # Try audio features (may be deprecated/restricted)
-    features: list[dict] = []
-    track_ids = [t["id"] for t in tracks]
-    try:
-        features_data = spotify_get(token, "/audio-features", {"ids": ",".join(track_ids[:100])})
-        features = [f for f in features_data.get("audio_features", []) if f is not None]
-    except Exception as e:
-        print(f"Audio features unavailable (may be deprecated): {e}", file=sys.stderr)
-        print("Falling back to genre-based profiling.", file=sys.stderr)
-
-    profile = aggregate_profile(tracks, artists, features)
-    save_cached_profile(profile)
-    return profile
-
-
-def aggregate_profile(tracks: list, artists: list, features: list) -> MusicProfile:
-    has_audio_features = len(features) > 0
-    n = len(features) or 1
-
-    if has_audio_features:
-        avg_valence = sum(f["valence"] for f in features) / n
-        avg_energy = sum(f["energy"] for f in features) / n
-        avg_danceability = sum(f["danceability"] for f in features) / n
-        avg_acousticness = sum(f["acousticness"] for f in features) / n
-        avg_tempo = sum(f["tempo"] for f in features) / n
-
-        avg_instrumentalness = sum(f["instrumentalness"] for f in features) / n
-        time_sigs = set(f["time_signature"] for f in features)
-        time_sig_variety = min(len(time_sigs) / 5, 1.0)
-        avg_complexity = avg_instrumentalness * 0.3 + time_sig_variety * 0.3 + (1 - avg_danceability) * 0.4
-    else:
-        # Fallback: estimate from genres
-        avg_valence, avg_energy, avg_danceability, avg_acousticness, avg_tempo, avg_complexity = (
-            estimate_features_from_genres(artists)
-        )
-
-    avg_popularity = sum(t.get("popularity", 50) for t in tracks) / (len(tracks) or 1)
-    obscurity_score = 1 - avg_popularity / 100
-
-    genre_counts: dict[str, int] = {}
-    for artist in artists:
-        for genre in artist.get("genres", []):
-            genre_counts[genre] = genre_counts.get(genre, 0) + 1
-    total_genres = sum(genre_counts.values()) or 1
-    genre_distribution = {g: c / total_genres for g, c in sorted(genre_counts.items(), key=lambda x: -x[1])}
-
-    mood_label = derive_mood(avg_valence, avg_energy)
-
-    top_artist_names = [a["name"] for a in artists[:10]]
-    top_track_names = [f"{t['name']} — {t['artists'][0]['name']}" for t in tracks[:10] if t.get('artists')]
-
-    return MusicProfile(
-        avg_valence=avg_valence,
-        avg_energy=avg_energy,
-        avg_danceability=avg_danceability,
-        avg_acousticness=avg_acousticness,
-        avg_tempo=avg_tempo,
-        avg_complexity=avg_complexity,
-        obscurity_score=obscurity_score,
-        genre_distribution=genre_distribution,
-        mood_label=mood_label,
-        top_artists=top_artist_names,
-        top_tracks=top_track_names,
-        has_audio_features=has_audio_features,
-    )
-
-
-# ──────────────────────────────────────────────
-# Genre-based feature estimation (fallback)
-# ──────────────────────────────────────────────
-
-# Genre archetypes: (valence, energy, danceability, acousticness, tempo_norm, complexity)
-GENRE_ARCHETYPES: dict[str, tuple[float, float, float, float, float, float]] = {
-    # Electronic / Dance
-    "edm": (0.6, 0.8, 0.8, 0.1, 0.65, 0.2),
-    "house": (0.6, 0.7, 0.8, 0.1, 0.62, 0.3),
-    "techno": (0.3, 0.8, 0.7, 0.05, 0.65, 0.4),
-    "ambient": (0.3, 0.2, 0.2, 0.6, 0.40, 0.6),
-    "electronic": (0.5, 0.6, 0.6, 0.15, 0.60, 0.5),
-    "drum and bass": (0.4, 0.9, 0.6, 0.05, 0.85, 0.4),
-    "trip hop": (0.3, 0.4, 0.5, 0.3, 0.45, 0.6),
-    # Rock
-    "rock": (0.5, 0.7, 0.5, 0.3, 0.60, 0.4),
-    "indie rock": (0.4, 0.6, 0.5, 0.4, 0.55, 0.5),
-    "art rock": (0.3, 0.5, 0.4, 0.4, 0.50, 0.8),
-    "post-punk": (0.3, 0.6, 0.5, 0.3, 0.55, 0.6),
-    "shoegaze": (0.3, 0.5, 0.3, 0.3, 0.50, 0.6),
-    "post-rock": (0.3, 0.5, 0.2, 0.4, 0.50, 0.7),
-    "punk": (0.5, 0.9, 0.4, 0.3, 0.70, 0.2),
-    "metal": (0.3, 0.9, 0.3, 0.1, 0.70, 0.5),
-    "alternative": (0.4, 0.6, 0.5, 0.3, 0.55, 0.5),
-    "grunge": (0.3, 0.7, 0.4, 0.3, 0.55, 0.4),
-    "classic rock": (0.5, 0.7, 0.5, 0.3, 0.60, 0.4),
-    "progressive rock": (0.4, 0.6, 0.3, 0.3, 0.55, 0.9),
-    # Jazz / Classical / Complex
-    "jazz": (0.4, 0.4, 0.4, 0.7, 0.55, 0.9),
-    "classical": (0.4, 0.3, 0.2, 0.9, 0.50, 0.9),
-    "contemporary classical": (0.3, 0.3, 0.2, 0.8, 0.45, 0.9),
-    "experimental": (0.3, 0.5, 0.3, 0.4, 0.50, 0.9),
-    "avant-garde": (0.3, 0.5, 0.2, 0.4, 0.50, 0.9),
-    # Folk / Acoustic / Singer-Songwriter
-    "folk": (0.5, 0.3, 0.4, 0.8, 0.50, 0.4),
-    "singer-songwriter": (0.4, 0.3, 0.4, 0.7, 0.50, 0.4),
-    "acoustic": (0.5, 0.3, 0.4, 0.9, 0.50, 0.3),
-    "americana": (0.5, 0.4, 0.4, 0.7, 0.55, 0.4),
-    "country": (0.6, 0.5, 0.5, 0.6, 0.55, 0.3),
-    "bluegrass": (0.5, 0.5, 0.5, 0.9, 0.60, 0.5),
-    # Pop / R&B
-    "pop": (0.7, 0.6, 0.7, 0.2, 0.60, 0.2),
-    "indie pop": (0.6, 0.5, 0.6, 0.3, 0.55, 0.3),
-    "r&b": (0.5, 0.5, 0.7, 0.3, 0.55, 0.4),
-    "soul": (0.5, 0.5, 0.6, 0.5, 0.55, 0.5),
-    "funk": (0.7, 0.7, 0.8, 0.3, 0.55, 0.5),
-    "neo soul": (0.5, 0.4, 0.6, 0.4, 0.50, 0.6),
-    # Hip-Hop
-    "hip hop": (0.5, 0.6, 0.8, 0.1, 0.55, 0.3),
-    "rap": (0.5, 0.7, 0.7, 0.1, 0.55, 0.3),
-    "underground hip hop": (0.4, 0.5, 0.6, 0.2, 0.50, 0.5),
-    # World / Latin
-    "bossa nova": (0.6, 0.3, 0.6, 0.7, 0.55, 0.5),
-    "latin": (0.7, 0.7, 0.8, 0.3, 0.55, 0.3),
-    "afrobeat": (0.6, 0.7, 0.8, 0.3, 0.55, 0.6),
-    "reggae": (0.6, 0.5, 0.7, 0.4, 0.45, 0.3),
-    # Blues
-    "blues": (0.4, 0.5, 0.5, 0.6, 0.50, 0.5),
-    "delta blues": (0.3, 0.4, 0.4, 0.8, 0.50, 0.5),
-}
-
-# Default for unrecognized genres
-_DEFAULT_ARCHETYPE = (0.5, 0.5, 0.5, 0.5, 0.55, 0.5)
-
-
-def estimate_features_from_genres(
-    artists: list[dict],
-) -> tuple[float, float, float, float, float, float]:
-    """Estimate audio features from genre distribution when audio features API is unavailable."""
-    genre_weights: dict[str, float] = {}
-    for artist in artists:
-        for genre in artist.get("genres", []):
-            genre_weights[genre] = genre_weights.get(genre, 0) + 1
-
-    if not genre_weights:
-        v, e, d, a, t, c = _DEFAULT_ARCHETYPE
-        return (v, e, d, a, t * 200, c)
-
-    total = sum(genre_weights.values())
-    valence = energy = dance = acoustic = tempo = complexity = 0.0
-
-    for genre, count in genre_weights.items():
-        w = count / total
-        # Find best matching archetype (longest key match wins)
-        archetype = _DEFAULT_ARCHETYPE
-        best_len = 0
-        for key, vals in GENRE_ARCHETYPES.items():
-            if key in genre or genre in key:
-                if len(key) > best_len:
-                    archetype = vals
-                    best_len = len(key)
-        valence += archetype[0] * w
-        energy += archetype[1] * w
-        dance += archetype[2] * w
-        acoustic += archetype[3] * w
-        tempo += archetype[4] * w
-        complexity += archetype[5] * w
-
-    return valence, energy, dance, acoustic, tempo * 200, complexity
-
-
-def derive_mood(valence: float, energy: float) -> str:
-    if valence > 0.6 and energy > 0.6:
-        return "euphoric"
-    if valence > 0.6 and energy < 0.4:
-        return "serene"
-    if valence < 0.4 and energy > 0.6:
-        return "intense"
-    if valence < 0.4 and energy < 0.4:
-        return "melancholic"
-    if valence > 0.5:
-        return "upbeat"
-    if energy > 0.5:
-        return "driven"
-    return "contemplative"
-
-
-# ──────────────────────────────────────────────
-# Cross-domain mapping
-# ──────────────────────────────────────────────
-
-def clamp01(n: float) -> float:
-    return max(0.0, min(1.0, n))
-
-
-def weighted_avg(pairs: list[tuple[float, float]]) -> float:
-    total = sum(clamp01(v) * w for v, w in pairs)
-    weight_sum = sum(w for _, w in pairs)
-    return clamp01(total / weight_sum) if weight_sum else 0.0
-
-
-def genre_affinity(genres: dict[str, float], keywords: list[str]) -> float:
-    """How much of the genre distribution matches the given keywords (0-1).
-    Uses word-boundary-aware matching to avoid false positives like 'art' matching 'martial'."""
-    total = 0.0
-    for genre, weight in genres.items():
-        genre_tokens = genre.replace("-", " ").split()
-        genre_parts = set(genre_tokens)
-        for k in keywords:
-            k_tokens = k.replace("-", " ").split()
-            k_parts = set(k_tokens)
-            # Match if keyword parts are a subset of genre parts, or genre is substring of keyword
-            if k_parts <= genre_parts or genre in k:
-                total += weight
-                break
-            # Prefix matching: a keyword token is a prefix of a genre token or vice versa
-            if any(
-                gt.startswith(kt) or kt.startswith(gt)
-                for kt in k_tokens
-                for gt in genre_tokens
-            ):
-                total += weight
-                break
-    return total
-
-
-def music_to_wine_profile(m: MusicProfile) -> WineProfile:
-    # Genre signals (supplements audio features, serves as primary signal when unavailable)
-    g = m.genre_distribution
-    genre_dark = genre_affinity(g, ["metal", "punk", "goth", "dark", "doom", "black", "death", "grunge", "post-punk"])
-    genre_complex = genre_affinity(g, ["jazz", "classical", "prog", "experimental", "avant", "art rock", "contemporary"])
-    genre_acoustic = genre_affinity(g, ["folk", "acoustic", "singer-songwriter", "bluegrass", "americana", "chamber"])
-    genre_electronic = genre_affinity(g, ["edm", "house", "techno", "electronic", "synth", "drum and bass", "dubstep"])
-    genre_bright = genre_affinity(g, ["pop", "indie pop", "dance", "disco", "funk", "latin", "reggaeton", "k-pop"])
-    genre_earthy = genre_affinity(g, ["blues", "delta", "soul", "gospel", "country", "roots", "world", "afrobeat"])
-
-    # Normalize tempo to 0-1 (cap at 180 to handle DnB sensibly, floor at 60)
-    tempo_norm = clamp01((m.avg_tempo - 60) / 120) if m.avg_tempo > 0 else 0.5
-
-    return WineProfile(
-        body=weighted_avg([
-            (m.avg_energy, 0.4), (tempo_norm, 0.2), (1 - m.avg_acousticness, 0.15),
-            (genre_dark, 0.15), (genre_electronic, 0.1),
-        ]),
-        sweetness=weighted_avg([
-            (m.avg_valence, 0.4), (m.avg_danceability, 0.2), (1 - m.avg_complexity, 0.1),
-            (genre_bright, 0.2), (1 - genre_dark, 0.1),
-        ]),
-        tannin=weighted_avg([
-            (1 - m.avg_valence, 0.3), (m.avg_complexity, 0.2), (m.avg_energy, 0.2),
-            (genre_dark, 0.2), (genre_complex, 0.1),
-        ]),
-        acidity=weighted_avg([
-            (m.avg_complexity, 0.3), (1 - m.avg_valence, 0.2), (m.obscurity_score, 0.2),
-            (genre_complex, 0.2), (genre_acoustic, 0.1),
-        ]),
-        complexity=weighted_avg([
-            (m.avg_complexity, 0.3), (m.obscurity_score, 0.2), (1 - m.avg_danceability, 0.1),
-            (genre_complex, 0.3), (genre_earthy, 0.1),
-        ]),
-        fruitiness=weighted_avg([
-            (m.avg_valence, 0.3), (m.avg_energy, 0.2), (1 - m.avg_complexity, 0.1),
-            (genre_bright, 0.25), (1 - genre_dark, 0.15),
-        ]),
-        earthiness=weighted_avg([
-            (m.avg_acousticness, 0.3), (m.avg_complexity, 0.2), (m.obscurity_score, 0.2),
-            (genre_acoustic, 0.15), (genre_earthy, 0.15),
-        ]),
-        spiciness=weighted_avg([
-            (m.avg_energy, 0.3), (1 - m.avg_valence, 0.2), (m.avg_complexity, 0.2),
-            (genre_dark, 0.15), (genre_electronic, 0.15),
-        ]),
-    )
-
-
-def score_wine_match(target: WineProfile, wine: Wine) -> float:
-    """Score using squared differences to amplify spread between good and bad matches."""
-    weights = {
-        "body": 1.5, "sweetness": 1.2, "tannin": 1.0, "acidity": 1.0,
-        "complexity": 1.5, "fruitiness": 0.8, "earthiness": 0.8, "spiciness": 0.7,
-    }
-    total_weight = 0.0
-    weighted_sq_distance = 0.0
-    for dim, w in weights.items():
-        diff = getattr(target, dim) - getattr(wine.profile, dim)
-        weighted_sq_distance += (diff ** 2) * w
-        total_weight += w
-    # Squared distance max is 1.0 per dim, so normalize and convert to similarity
-    return 1 - (weighted_sq_distance / total_weight) ** 0.5
-
-
-def normalize_scores(recs: list[tuple]) -> list[tuple]:
-    """Rescale scores so best match = 95-99% and worst = proportionally lower."""
-    if len(recs) <= 1:
-        return recs
-    scores = [r[1] for r in recs]
-    lo, hi = min(scores), max(scores)
-    spread = hi - lo
-    if spread < 0.01:
-        return [(w, 0.90, c) for w, _, c in recs]
-    return [
-        (w, 0.60 + 0.38 * (s - lo) / spread, c)
-        for w, s, c in recs
-    ]
-
-
-def generate_connections(music: MusicProfile, wine: Wine) -> list[MusicWineConnection]:
-    """Generate cross-domain connections. Uses graduated thresholds to ensure
-    most users get at least some connections (the entertainment value)."""
-    connections = []
-    g = music.genre_distribution
-
-    # Dark/brooding music → tannic wines (graduated threshold)
-    dark_score = (1 - music.avg_valence) * 0.5 + genre_affinity(g, ["post-punk", "metal", "goth", "dark", "doom", "grunge"]) * 0.5
-    if dark_score > 0.25 and wine.profile.tannin > 0.5:
-        connections.append(MusicWineConnection(
-            music_signal=f"Your music leans dark and introspective (valence: {music.avg_valence * 100:.0f}%)",
-            wine_signal="This wine has serious tannin structure",
-            explanation="You like your art with tension — this wine delivers that same brooding complexity",
-            strength=min(0.5 + dark_score, 0.95),
-        ))
-
-    # Acoustic/organic → old-world wines (graduated)
-    acoustic_score = music.avg_acousticness * 0.5 + genre_affinity(g, ["folk", "acoustic", "singer-songwriter", "bluegrass", "classical"]) * 0.5
-    if acoustic_score > 0.3 and "old-world" in wine.tags:
-        connections.append(MusicWineConnection(
-            music_signal="You gravitate toward acoustic, organic sounds",
-            wine_signal=f"{wine.region} has centuries of winemaking tradition",
-            explanation="Analog music lover, traditional winemaking. No synthetic shortcuts",
-            strength=min(0.4 + acoustic_score, 0.9),
-        ))
-
-    # Complexity seekers (graduated)
-    complexity_score = music.avg_complexity * 0.4 + genre_affinity(g, ["jazz", "classical", "prog", "experimental", "art rock"]) * 0.6
-    if complexity_score > 0.3 and wine.profile.complexity > 0.6:
-        connections.append(MusicWineConnection(
-            music_signal="You seek out musically complex, layered compositions",
-            wine_signal=f"This {wine.varietal} rewards patient attention",
-            explanation="You don't want background music, and you don't want background wine",
-            strength=min(0.5 + complexity_score * 0.5, 0.95),
-        ))
-
-    # Niche taste → obscure varietals (graduated)
-    niche_score = music.obscurity_score
-    if niche_score > 0.4 and "obscure-varietal" in wine.tags:
-        artist_ref = music.top_artists[0] if music.top_artists else "niche artists"
-        connections.append(MusicWineConnection(
-            music_signal="Your taste runs deep into niche territory",
-            wine_signal=f"{wine.varietal} from {wine.region} — not exactly grocery store fare",
-            explanation=f"Someone who listens to {artist_ref} doesn't want the wine equivalent of Top 40",
-            strength=min(0.4 + niche_score * 0.5, 0.95),
-        ))
-
-    # High energy → bold wines (graduated)
-    energy_score = music.avg_energy * 0.6 + genre_affinity(g, ["metal", "punk", "edm", "drum and bass", "hard rock"]) * 0.4
-    if energy_score > 0.4 and wine.profile.body > 0.6:
-        connections.append(MusicWineConnection(
-            music_signal="Your playlists hit hard — high energy, high intensity",
-            wine_signal=f"This is a full-bodied {wine.color} that doesn't hold back",
-            explanation="You like your music loud and your wine bold",
-            strength=min(0.4 + energy_score * 0.4, 0.85),
-        ))
-
-    # Bright/happy → fruit-forward (graduated)
-    bright_score = music.avg_valence * 0.5 + genre_affinity(g, ["pop", "indie pop", "funk", "disco", "latin"]) * 0.5
-    if bright_score > 0.35 and wine.profile.fruitiness > 0.5:
-        connections.append(MusicWineConnection(
-            music_signal="Your music is bright and upbeat",
-            wine_signal=f"Fruit-forward {wine.varietal} with approachable charm",
-            explanation="Happy music, happy wine — nothing wrong with joy",
-            strength=min(0.3 + bright_score * 0.5, 0.8),
-        ))
-
-    # Danceable → approachable/sweet
-    if music.avg_danceability > 0.55 and wine.profile.sweetness > 0.2:
-        connections.append(MusicWineConnection(
-            music_signal="You love a groove — high danceability across your library",
-            wine_signal=f"This {wine.varietal} has a touch of approachable sweetness",
-            explanation="Dance music and a hint of sweetness — both are about pleasure, not pretension",
-            strength=min(0.3 + music.avg_danceability * 0.4, 0.75),
-        ))
-
-    # Quiet/ambient → delicate wines
-    quiet_score = (1 - music.avg_energy) * 0.5 + genre_affinity(g, ["ambient", "drone", "minimal", "new age"]) * 0.5
-    if quiet_score > 0.35 and wine.profile.body < 0.5:
-        connections.append(MusicWineConnection(
-            music_signal="Your listening is quiet, ambient, introspective",
-            wine_signal=f"A delicate {wine.color} that whispers rather than shouts",
-            explanation="You don't need volume to feel something. This wine understands that",
-            strength=min(0.4 + quiet_score * 0.4, 0.85),
-        ))
-
-    # Genre-region affinity
-    if genre_affinity(g, ["bossa nova", "latin", "flamenco"]) > 0.1 and wine.country in ("Spain", "Argentina", "Portugal"):
-        connections.append(MusicWineConnection(
-            music_signal="Latin and Mediterranean rhythms in your library",
-            wine_signal=f"This {wine.varietal} comes from {wine.country}'s winemaking tradition",
-            explanation="Your music has warmth and rhythm — so does this wine's homeland",
-            strength=0.65,
-        ))
-
-    if genre_affinity(g, ["blues", "soul", "gospel", "americana"]) > 0.1 and wine.country == "USA":
-        connections.append(MusicWineConnection(
-            music_signal="American roots music runs through your listening",
-            wine_signal=f"{wine.producer} in {wine.region} — American terroir",
-            explanation="Rooted in American soil, both the music and the wine",
-            strength=0.6,
-        ))
-
-    return connections
-
-
-def recommend_wines(music: MusicProfile, wines: list[Wine], top_n: int = 3) -> list[Recommendation]:
-    target = music_to_wine_profile(music)
-
-    rating_boosts = load_rating_boosts()
-
-    scored = []
-    for wine in wines:
-        score = score_wine_match(target, wine)
-        # Multiplicative boost from ratings (caps total adjustment at +/-15%)
-        boost = sum(rating_boosts.get(tag, 0.0) for tag in wine.tags)
-        boost = max(-0.15, min(0.15, boost * 0.03))
-        score = clamp01(score * (1 + boost))
-        connections = generate_connections(music, wine)
-        scored.append((wine, score, connections))
-
-    scored.sort(key=lambda x: -x[1])
-    # Normalize so scores use the full display range
-    scored = normalize_scores(scored)
-
-    results = []
-    for wine, score, connections in scored[:top_n]:
-        best_conn = sorted(connections, key=lambda c: -c.strength) if connections else []
-        if best_conn:
-            reasoning = f"{best_conn[0].explanation}. Try this {wine.name} — a {wine.varietal} from {wine.producer} in {wine.region}."
-        else:
-            reasoning = f"Based on your overall taste profile, this {wine.varietal} from {wine.region} should resonate with you."
-        results.append(Recommendation(wine=wine, score=score, reasoning=reasoning, connections=connections))
-
-    return results
-
-
-def load_rating_boosts() -> dict[str, float]:
-    """Load tag-level boosts from past ratings. Positive = liked, negative = disliked."""
-    if not RATINGS_FILE.exists():
-        return {}
-    try:
-        ratings = json.loads(RATINGS_FILE.read_text())
-    except (json.JSONDecodeError, OSError):
-        return {}
-    if not isinstance(ratings, list):
-        return {}
-    tag_scores: dict[str, list[float]] = {}
-    for entry in ratings:
-        normalized = (entry.get("rating", 3) - 3) / 2  # 1=-1, 2=-0.5, 3=0, 4=0.5, 5=1
-        for tag in entry.get("tags", []):
-            tag_scores.setdefault(tag, []).append(normalized)
-    return {tag: sum(scores) / len(scores) for tag, scores in tag_scores.items()}
-
-
-# ──────────────────────────────────────────────
-# Demo profile
-# ──────────────────────────────────────────────
-
-DEMO_PROFILE = MusicProfile(
-    avg_valence=0.32,
-    avg_energy=0.55,
-    avg_danceability=0.45,
-    avg_acousticness=0.62,
-    avg_tempo=115.0,
-    avg_complexity=0.72,
-    obscurity_score=0.68,
-    genre_distribution={
-        "art rock": 0.15, "post-punk": 0.12, "jazz": 0.10,
-        "indie rock": 0.09, "ambient": 0.08, "shoegaze": 0.07, "classical": 0.06,
-    },
-    mood_label="contemplative",
-    top_artists=["Radiohead", "Nick Cave", "Talk Talk", "Bjork", "John Coltrane",
-                 "Portishead", "The National", "Brian Eno", "Deerhunter", "Low"],
-    top_tracks=["Reckoner — Radiohead", "Into My Arms — Nick Cave",
-                "Bloodbuzz Ohio — The National", "Hyperballad — Bjork",
-                "A Love Supreme Pt. 1 — John Coltrane"],
-    has_audio_features=True,
-)
-
-
-# ──────────────────────────────────────────────
-# Output formatting
-# ──────────────────────────────────────────────
-
-def format_bar(value: float, width: int = 20) -> str:
-    filled = round(max(0.0, min(1.0, value)) * width)
-    return "=" * filled + "-" * (width - filled)
-
-
-def print_profile(music: MusicProfile, wine_profile: WineProfile) -> None:
-    print("=" * 50)
-    print("YOUR MUSIC PROFILE")
-    print("=" * 50)
-    if not music.has_audio_features:
-        print("  (estimated from genres — audio features unavailable)")
-    print(f"  Mood:          {music.mood_label}")
-    print(f"  Valence:       [{format_bar(music.avg_valence)}] {music.avg_valence * 100:.0f}%")
-    print(f"  Energy:        [{format_bar(music.avg_energy)}] {music.avg_energy * 100:.0f}%")
-    print(f"  Danceability:  [{format_bar(music.avg_danceability)}] {music.avg_danceability * 100:.0f}%")
-    print(f"  Acousticness:  [{format_bar(music.avg_acousticness)}] {music.avg_acousticness * 100:.0f}%")
-    print(f"  Complexity:    [{format_bar(music.avg_complexity)}] {music.avg_complexity * 100:.0f}%")
-    print(f"  Obscurity:     [{format_bar(music.obscurity_score)}] {music.obscurity_score * 100:.0f}%")
-    print(f"  Tempo:         {music.avg_tempo:.0f} BPM")
-    if music.top_artists:
-        print(f"  Top artists:   {', '.join(music.top_artists[:5])}")
-    if music.genre_distribution:
-        top_genres = list(music.genre_distribution.keys())[:5]
-        print(f"  Top genres:    {', '.join(top_genres)}")
-    print()
-
-    print("=" * 50)
-    print("DERIVED WINE PROFILE")
-    print("=" * 50)
-    for dim in ["body", "sweetness", "tannin", "acidity", "complexity", "fruitiness", "earthiness", "spiciness"]:
-        val = getattr(wine_profile, dim)
-        print(f"  {dim:<12}   [{format_bar(val)}] {val * 100:.0f}%")
-    print()
-
-
-def print_recommendations(recs: list[Recommendation]) -> None:
-    print("=" * 50)
-    print("RECOMMENDATIONS")
-    print("=" * 50)
-    print()
-
-    for i, rec in enumerate(recs, 1):
-        w = rec.wine
-        stars = "*" * round(rec.score * 5) + "." * (5 - round(rec.score * 5))
-        print(f"{i}. {w.name}")
-        print(f"   {w.varietal} | {w.region}, {w.country} | {w.price_range}")
-        if w.vintage:
-            print(f"   Vintage: {w.vintage}")
-        print(f"   Match: [{stars}] ({rec.score * 100:.0f}%)")
-        print(f"   \"{w.description}\"")
-        print(f"   Why: {rec.reasoning}")
-        for conn in rec.connections:
-            print(f"   ~ {conn.explanation}")
-        print()
-
-
-def print_json_output(music: MusicProfile, wine_profile: WineProfile, recs: list[Recommendation]) -> None:
-    output = {
-        "music_profile": asdict(music),
-        "wine_profile": asdict(wine_profile),
-        "recommendations": [
-            {
-                "wine": {
-                    "id": r.wine.id,
-                    "name": r.wine.name,
-                    "producer": r.wine.producer,
-                    "region": r.wine.region,
-                    "country": r.wine.country,
-                    "varietal": r.wine.varietal,
-                    "color": r.wine.color,
-                    "price_range": r.wine.price_range,
-                    "description": r.wine.description,
-                    "tags": r.wine.tags,
-                    "vintage": r.wine.vintage,
-                    "profile": asdict(r.wine.profile),
-                },
-                "score": round(r.score, 3),
-                "reasoning": r.reasoning,
-                "connections": [asdict(c) for c in r.connections],
-            }
-            for r in recs
-        ],
-    }
-    print(json.dumps(output, indent=2))
-
-
-# ──────────────────────────────────────────────
-# Main
-# ──────────────────────────────────────────────
-
-def main():
-    parser = argparse.ArgumentParser(description="SommeliAgent — Wine recommendations from your Spotify")
-    parser.add_argument("--demo", action="store_true", help="Use demo profile (no Spotify needed)")
-    parser.add_argument("--color", choices=["red", "white", "rose", "orange", "sparkling", "dessert"], help="Filter by wine color")
-    parser.add_argument("--price", choices=["budget", "mid", "premium", "luxury"], help="Filter by price range")
-    parser.add_argument("--count", type=int, default=3, help="Number of recommendations (default: 3)")
-    parser.add_argument("--profile", action="store_true", help="Show full music + wine profile")
-    parser.add_argument("--json", action="store_true", help="Output as JSON (for programmatic use)")
-    parser.add_argument("--no-cache", action="store_true", help="Bypass profile cache")
-    args = parser.parse_args()
-
-    if args.demo:
-        music = DEMO_PROFILE
-    else:
-        if args.no_cache and CACHE_FILE.exists():
-            CACHE_FILE.unlink()
-        token = get_access_token()
-        music = build_music_profile(token)
-
-    wines = WINE_DB
-    if args.color:
-        wines = [w for w in wines if w.color == args.color]
-    if args.price:
-        wines = [w for w in wines if w.price_range == args.price]
-
-    if not wines:
-        print("No wines match your filters.", file=sys.stderr)
-        sys.exit(1)
-
-    wine_profile = music_to_wine_profile(music)
-    count = max(1, min(args.count, len(wines)))
-    recs = recommend_wines(music, wines, count)
-
-    if args.json:
-        print_json_output(music, wine_profile, recs)
-    else:
-        if args.profile:
-            print_profile(music, wine_profile)
-        print_recommendations(recs)
-
-
-if __name__ == "__main__":
-    main()