Design Real-Time Driver Pay Aggregation

Design a production system that ingests a high-volume stream of delivery-order events and continuously computes each driver's daily pay. The system should support: - JSON event ingestion from upstream publishers - Peak-period multipliers that may require splitting a work interval at boundary times - Malformed, duplicate, or late-arriving events - Data volume large enough that a full day's raw events may not fit in memory - Recovery from consumer or service outages - Safe interaction with upstream APIs in production Discuss the end-to-end architecture, including publisher/consumer flow, idempotency under at-least-once delivery, partitioning strategy, streaming computation, persistence, replay and reconciliation, handling large data with external sorting or batch recomputation, and operational protections such as retries, circuit breakers, rate limiting, and backpressure.