A pipeline step calls an external service that sometimes returns transient errors. A timeout may occur after the external action succeeds. Which two practices make retries safer? Select 2 answers.

1. Generate a new business operation identifier on every retry.
2. Use bounded retries with backoff and an observable terminal failure path.
3. Use a stable operation identifier and idempotent external-action handling.
4. Retry forever at the maximum rate until the dependency accepts the request.
5. Treat every timeout as proof that no side effect occurred.