- Ping at the end, not the start, so a job that crashes halfway counts as missed.
- Report failures explicitly with the error, so you're not waiting for the grace period.
- Set a grace period a little longer than normal run-time variation.
- Use the job's time zone, so daylight saving doesn't create false alarms.
- Wake only for jobs with real cost: billing, backups, data syncs customers depend on.
Pagerless check-ins do all five: docs.