There is a particular kind of incident that never appears in your incident count and costs more than most outages you page someone for. On a Tuesday in March someone rebuilds a server. The new image is clean and missing one crontab entry. The nightly export to your billing partner stops. Nothing breaks, no page fires. Twenty-three days later the partner emails asking why they have not received data since March.
Why absence is invisible
Human attention is built to notice change, not steady state. A job that stops running produces nothing — no new log line saying "I did not run tonight," only a gap, and gaps are only visible if you are specifically looking for them.
The three most common causes
Infrastructure replacement — anything configured by hand rather than by code does not come back. Silent permission and credential expiry — the job still runs, still exits zero, and does nothing useful. Scheduling drift — someone edits a crontab and moves a job from every day to every year without noticing.
What does not solve this
Log aggregation can only alert on logs that exist. Dashboards require someone to notice what is not there. Exit code alerting requires a process to exist. Discipline decays after two weeks. Any control that depends on sustained human vigilance for something boring will fail.
What does solve it
A heartbeat gives your job a URL to call when it finishes and alerts when the call does not arrive. It works with no cooperation from the failing component — the server can be gone, the crontab wiped. The monitoring system is waiting for a message that says "everything is fine," and it acts when that message does not come. This is a dead man's switch, and the metaphor is apt.
Doing an audit today
List every scheduled task — crontabs, systemd timers, Kubernetes CronJobs, cloud scheduler entries, application-level schedulers. For each, answer: when did it last successfully complete, and how would you know if it stopped? The jobs where the second answer is "someone would eventually notice" are your exposure. Wire heartbeats to the top five first.
A note on what you will find
Teams who run this audit almost always find at least one thing already broken. That is the natural consequence of the blind spot: if you could see these failures without looking, they would not be silent.