Whenever you're ready, walk me through how you'd approach this.
How do you ensure data consistency and integrity across distributed database systems?
Explain how you would reason about consistency requirements, transactions, replication, conflict resolution, schema evolution, validation, recovery, and observability. Address how the design should protect downstream ML feature and training data from duplication, loss, corruption, stale reads, and training-serving skew.