Whenever you're ready, walk me through how you'd approach this.
How do you ensure data consistency in a distributed system?
Discuss the consistency model you would choose, how data moves across services and replicas, and how you handle concurrent writes, retries, failures, and delayed events. Address both correctness and availability, including how you would detect and recover from inconsistencies. Relate the design to ML concerns such as online and offline feature parity, reproducible training data, and model-serving behavior.