Retry

Things fail for reasons that pass: a service is briefly down, a network blips, a rate limit trips, a dependency lags. A retry re-runs the failed step after a pause instead of giving up. Waiting a little longer each time (back-off) keeps retries from hammering a struggling service.

Retries are only safe when a step is idempotent, meaning running it twice does no extra harm. Paired with that, a retry turns most stumbles into non-events, and a human is only alerted for the failures that truly cannot recover.