A Job represents work that should finish. Its Pod template needs a finite command and a suitable restart policy, usually Never or OnFailure. Completions describe how many successful runs are needed; parallelism bounds how many may run at once. Backoff controls retries after failures and doesn't turn a failing command into a successful result. A long-running service belongs in a different workload shape. Here the fictional batch commands finish deterministically and leave status you can inspect. Your tasks vary the completion target and retry policy, then distinguish a completed Job from a resource whose command repeatedly exits with an error.