Why We Send Ourselves a Fake Emergency Every Monday

Every Monday at 9:00, FlowAlert pages us. High priority, acknowledgement required, a real push through the real pipeline — API, routing rules, phones. It says, in effect: this is a drill, prove you'd have heard the real thing. Someone on the team has to physically tap Acknowledge, and the result lands on the record like any other alert.

We built this into the product, we run it on ourselves, and we think anyone who carries any kind of pager should steal the habit. Here's the reasoning.

---

Pagers don't break loudly. They rot quietly.

An alerting pipeline is a chain of parts that all decay silently: push tokens expire when a phone is replaced or an app reinstalled; a well-meaning tap on "hide notifications" during a movie sticks forever; a new phone ships with the app's sound off; a routing rule that made sense in March quietly matches nothing in August; a firewall rule, a changed API key, an OS update's new battery policy. None of these announce themselves. Every one of them produces the same symptom: silence — indistinguishable from "everything's fine" until the night it isn't.

Monitoring tools are supposed to defeat silent failure, but the monitoring pipeline itself is just another system that can fail silently. Who watches the watcher? Nobody — unless watching it is scheduled.

Rehearsal is the only honest test

You can't unit-test whether a human's phone makes a sound on a nightstand. The only test that covers the whole chain — event in, rules evaluated, push dispatched, OS delivers, sound plays, human notices, acknowledgement recorded — is sending a real alert and requiring a real thumb. So that's the mechanism: one synthetic emergency a week, through production routing, that demands acknowledgement.

The demand matters. A drill you can glance at and ignore tests delivery. A drill you must acknowledge tests response — and response is the product. If Monday's drill sits unacknowledged, we haven't had an awkward morning; we've had the cheapest possible discovery that the pager is broken, seven days at most after it broke, during a rehearsal instead of an outage.

Fire drills, but for software

Buildings don't run fire drills because anyone doubts fire exists. They run them because the failure mode isn't disbelief — it's unpracticed response: doors nobody knew were locked, alarms nobody recognized. Alerting has the same physics. The weekly drill is the fire-drill instinct applied to the pager: scheduled, boring, slightly annoying, and the only reason you get to feel confident on the other 6 days, 23 hours and 55 minutes.

It also does something subtler: it keeps the meaning of the urgent sound alive. A tone you hear once a week — always deserving a response — stays wired to "act." A pager that's silent for six months becomes deniable ("is that thing even still set up?").

Steal the habit

If you run anything whose failure should wake someone, schedule a drill: weekly, through your real rules, acknowledgement required, on the record. In FlowAlert it's one toggle — Settings → Workspace → Weekly test alert — Monday 9:00 in your workspace's timezone. It needs a routing rule that matches test.* (a catch-all counts); if you enable the drill without one, your dashboard will tell you before Monday does. Our own drill history rides the same public machinery as everything else — flowalert.io/status — because a trust ritual you perform in private is only half a ritual.

You're not trusting the pager. You're testing it. Weekly, on the record.