A customer tells you their data is safe because the database is replicated across three availability zones. What is your response, and what would you build?

1. Separate the two properties: replication protects availability, backups protect against data loss caused by the writes themselves. 2. Show the failure replication does not cover: a bad DELETE or a faulty migration is a legitimate write and propagates to every replica within milliseconds. 3. Add the other uncovered cases — corruption propagated by the replication stream, a ransomware event, and a deletion discovered weeks later. 4. Specify point-in-time recovery: periodic base backups plus continuous write-ahead log archiving, so you can restore to a moment just before the damaging statement. 5. Store backups in a separate account or trust boundary with write-once retention, so the credentials that can destroy production cannot destroy the backups. 6. Define RPO and RTO explicitly and let them drive the retention and snapshot schedule. 7. Insist on tested restores on a schedule — an unrestored backup is a hypothesis, not a backup. 8. Keep logical dumps alongside physical backups so a single table can be recovered without restoring the whole cluster.