Durability means that once a database reports a transaction as committed, the system should be able to recover that transaction after a failure.

That sounds simple, but it hides several hard questions. What if the database process crashes? What if the operating system restarts before dirty pages have been flushed? What if the disk itself fails, or a replica is behind, or someone deletes the wrong rows by mistake?

Different mechanisms protect against different failures. Write-ahead logging protects against crashes on one machine. Checkpointing makes crash recovery faster. Replication protects against machine or disk loss. Backups and point-in-time recovery protect against corruption, bad deploys, and human mistakes.

Durability is a stack of safeguards rather than one feature.