A multi-task job fails after its first task successfully writes output. What must retry design consider?

1. Delete all checkpoints before every retry
2. Disable every downstream validation rule
3. Idempotent writes and the scope of rerun tasks
4. Assume all successful writes are automatically rolled back