Is it still running?

You set something up to run by itself every day. It worked for weeks. Then it stopped, and nothing told you. Here’s a simple way to find out fast. It’s an idea, not a program.

Why it fails quietly

Most jobs that run on a timer only speak up when they hit an error. But many failures make no error at all. The computer was asleep. The job ran but found nothing to do. The timer itself stopped. Each time, you hear nothing, and silence looks just like success.

Leave a mark, check it from somewhere else

Each time the job finishes, it leaves a mark: the time, and one fact about what it did, like “read 12 emails”.
Something else checks that mark on its own timer. It asks two things: is the mark fresh, and does the fact look like real work?
If the mark is too old, or the work looks empty, it tells you. Silence is the alarm, not the all-clear.

For example, your daily summary adds a line to a note each morning. Once a week, a calendar reminder tells you to look. No new line since yesterday? Something stopped.

Pick the time limit with care. For a daily job, “no mark for 26 hours” lets it run a little late without a false alarm. People often call this mark a heartbeat.

Four rules that make it work

Rule	Why
Judge the work, not the run	A job can start, find nothing and stop with no error. “It ran” is not the same as “it did its job”.
Check from somewhere else	If the check runs on the same timer as the job, they stop together, and you hear nothing.
Missed runs don’t come back	Most timers skip a run that was missed while the computer was off. Make the job catch up, or have the check say so.
Test the alarm once	Stop the job on purpose and wait. If no alert arrives, the check is the thing that’s broken.

Try it with your AI

Paste this into a chat and fill in the part in brackets. It asks for a plan first and changes nothing.

Plan an “is it still running?” check

I have a job that runs by itself: [what it does, and how often].
Help me add a simple check that it's still running and still doing its work.
1. What mark should it leave each time it finishes?
2. Where should a separate check run, so it doesn't stop when the job stops?
3. How old can the mark get before you warn me?
4. How do I test the warning once, on purpose?
Keep it simple, and don't change anything until I say OK.

What you should see: four short answers and a plan you can say yes or no to. Nothing is set up until you agree.

If it doesn’t look right

If it jumps straight to code, reply: “Plan first, in plain words.”
If the check would run on the same timer as the job, ask it to move the check somewhere else.

For a longer, first-person example, read Watching the watcher in the network watcher write-up. It’s more technical.

Back to the Toolshed