The AI slop track trap: what makes AI music sound like AI

# The AI-slop checklist. Score any generated track.
CHECKS = [
    # tag, description
    ("over_compressed",    "Peaks all glued at -1dB, no dynamic range. Sounds tiring after 30s."),
    ("muddy_textures",     "Every instrument occupies 100-800Hz. No separation. Listen on headphones and you'll notice."),
    ("repetitive_structure", "Same 8-bar loop through 3 minutes with token variation. No bridge, no drop, no arc."),
    ("cliche_lyrics",      "'neon', 'digital', 'electric dreams', 'soul on fire', 'lights on the dance floor'."),
    ("dead_autotune",      "Every vocal note snapped to grid. No slides, no vibrato, no breath."),
    ("mixed_too_loud",     "LUFS above -8 (streaming platforms normalize to -14; you're just losing headroom)."),
    ("perfect_end",        "Track ends exactly on the downbeat with a crescendo that wasn't earned by tension."),
    ("no_silence",         "No pauses anywhere. Real music breathes; AI music often doesn't."),
]

def score_track(observations: list[str]) -> dict:
    """Pass a list of tag names you observed. Score = # tripped."""
    tripped = [c for c in CHECKS if c[0] in observations]
    return {
        "score": len(tripped),
        "verdict": "slop" if len(tripped) >= 4 else "borderline" if len(tripped) >= 2 else "clean",
        "tripped": [t[0] for t in tripped],
    }

# Example: a Suno first-take with default settings
r = score_track(["over_compressed", "cliche_lyrics", "dead_autotune", "perfect_end"])
print(r)
# → {"score": 4, "verdict": "slop", ...} — regenerate.