Blog · Monitoring
Why your monitoring tools miss silent failures
Most monitoring waits for something to break and report it. The costliest outages are the ones where nothing reports anything at all.
Your dashboard is green, and yet the nightly backup hasn't written a byte in nine days. The reason is structural: error-based monitoring can only alert on signals it receives. A job that never starts, a cron dropped in a deploy, a worker killed mid-run: none of them send an error, so nothing fires. To an error-based system, silence and success look identical.
These are silent failures, and they're the default failure mode of anything scheduled:
- A backup that exits 0 after writing an empty archive.
- A cert-renewal job removed in a migration, ticking toward expiry.
- A data sync that swallows an error and "succeeds" with zero rows.
A dead man's switch flips the model: instead of waiting for a job to report failure, you wait for it to report success, and alert when that stops. The job makes one HTTP request when it finishes; miss it, and Mortemain alerts you. The silence becomes the alarm.
# the last thing your job does curl -fsS -m 10 https://ping.mortemain.com/your-check-uuid
No agent, no SDK, EU-hosted. Pick the one job whose silent failure would hurt most, and give it a check-in this afternoon.