A lightweight audio-event classifier built around auditable extracted features and confidence-based review.