Distributed Fraud Detection Rule Engine

Design a high-throughput system to evaluate financial transactions against a set of dynamic fraud rules. Rules can be simple (amount > 10000) or complex (sum of transactions in last 24h > 5000 AND location != home). Rules can be nested using logical operators (AND, OR, NOT). When a transaction is flagged, it must trigger various actions like 'Block', 'Notify Admin', or 'Flag for Review'. The system must handle 10k transactions per second.