Seven Days, No Pages

Seven days. Nothing. Not a page, not a warning that escalated, not a single one of those 3 a.m. alerts that turns out to be a disk filling on a box nobody owns.

I should say plainly that this is not luck. The disk alert is gone because somebody wrote the cleanup job in May. The flapping health check is gone because we finally fixed the timeout rather than raising the threshold for the fourth time. There is a list, and every item on it is somebody's afternoon.

The strange part is how it felt, which was not restful. I slept badly for the first three nights in exactly the way I sleep badly during a loud rotation, because the thing that wrecks the sleep is not the pages, it is carrying the phone.

By Wednesday I had stopped checking it. By Friday I had left it in another room for two hours, which I have not done since 2024, and I want to record that as the actual outcome of the work rather than the zero.

Everybody who has done a rotation knows this and almost nobody writes it down, because it sounds like complaining about a thing that did not happen.

A page at 3 a.m. costs you the rest of that night, and you can put a number on it. What is harder to see is the tax on the six nights where nothing happened, where you are asleep but not fully, at a level where a notification from anything at all gets you to sitting upright before you are conscious.

I have watched people come off a quiet week and be told they had an easy one. They did have an easier one. They were also carrying it the entire time, and the difference between those two things is the thing every on-call policy I have ever read fails to price.

The fix is not fewer alerts, although fewer alerts is good and we have done that. The fix is somebody else genuinely holding it for a full night, which requires two people, which requires headcount, which is where every version of this conversation ends.

Face down is supposed to be the whole technique. You put it face down and you stop seeing the screen light up, and the theory is that the checking stops.

The checking did not stop. It got worse, because face down removes the cheap glance and replaces it with a deliberate act, and a deliberate act is much more memorable, so by eight o'clock I could tell you exactly how many times I had done it.

Eleven. In forty minutes. On a night when the rotation had already ended.

It used to be a page and a half. Every rotation added something — a caveat, a known issue, a note about a service that behaves oddly on Sundays — and nobody ever removed anything, because removing something from a handover feels like taking a risk on somebody else's night.

By last summer it was unreadable, which meant it was unread, which meant it was worse than nothing: it looked like a handover had happened.

The four lines are what is left after we did the actual work each line was standing in for. The Sunday caveat is gone because the service was fixed. The known issue is gone because it was closed. Every line we deleted was deleted by making it untrue.

This is the thing I would tell anybody starting on an on-call rotation. A long runbook is not thoroughness. It is a list of things you have all agreed to keep working around, written down so that the working-around can be inherited.