Whenever you're ready, walk me through how you'd approach this.
How would you approach designing an anomaly detection system for high-frequency trading logs?
Explain how you would ingest and normalize heterogeneous logs, detect operational, security, and trading-behavior anomalies, and distinguish genuine incidents from expected market or system changes. Address streaming versus batch processing, model selection, alert prioritization, evaluation with limited labels, latency and availability requirements, investigation workflows, feature drift, training-serving skew, and monitoring. State the clarifying questions you would ask before choosing capacity, retention, and model thresholds.