One downstream task fails after earlier tasks completed successfully. Which recovery option should be evaluated before rerunning every mutation?

1. Blindly rerun every external side effect with new ids
2. Mark the failed task successful without execution
3. A supported repair run targeting the failed work and required dependencies
4. Delete the entire job definition