The watchman’s clock

A hundred years ago, a factory night watchman carried a clock.

Not to tell time. The clock was a leather-cased drum with a paper dial inside, and at the far corners of the building — the boiler room, the loading dock, the paint store at the back — a numbered key hung on a chain. The watchman walked to each key and turned it in the clock, and the key pressed its number into the paper. No ink, no wires, nothing to run down. In the morning the owner opened the drum and read the night: 12:42, boiler room. 12:54, dock. A skipped round showed as a blank arc of paper. The clocks were built, nearly unchanged, for a hundred and thirty years; the last mechanical one shipped in 2011.

The clock never knew what the watchman saw. It recorded one fact only — that a person had stood, at a certain minute, in a place where his eyes could reach the thing that mattered. Everything else was the person.

We relearned the difference this summer. A system we run stopped doing its one real job, and stayed stopped for twelve days, behind four layers of monitoring. Nobody was asleep. Every light was green, and every green light was honest. The service answered when asked. The scheduled errands ran on their minute, and each errand collected a fresh pass at the door, so the door log filled with successful entries around the clock. Activity everywhere. The board counted it faithfully.

Monitoring is not looking. Every signal we had measured activity. The outcome — whether the work came out the other end — had no signal at all. The by-the-minute jobs were minting fresh proof-of-life on a loop, and their proof drowned the silence of the one path that had died. The monitoring did not fail. It reported, precisely and continuously, on things adjacent to the point.

One person walked the rounds and found it in an afternoon. Not a cleverer dashboard — a walk. Stand in front of each system in turn and ask it the outcome question directly: show me the work, not the heartbeat.

There is an eighteen-minute film of this idea, just below. In 1958 the British Transport Films unit — the unit a BFI curator called the classiest industrial film unit in the world — followed the night gangs who go down into the London Underground in the few hours the current is off. Women sweeping lint from the tunnel walls by hand lamp. Gangs re-laying rail before dawn. At half past three, a man walking the track finds a broken rail, and it is mended before the first train. The railway did not learn the rail was broken from a report. A man stood on it. We named a company after a small fish that burns like a candle; this film belongs in the same lineage of attention.

Skip past the film

Then the part of our story that matters more than the finding. The walk got made recurring — and deliberately boring. Not an intelligent agent deciding each week what seems worth checking. A fixed list, the same probes, the same order, on a schedule, each stop ending in one plain line of evidence. A clock, not a mind. We had the clever option on the shelf and declined it on purpose. A clever walker gets interested in something and skips a room; a boring walker cannot skip. The blank on the paper would show.

The watchman’s clock was the same design decision, made in brass. The owner never asked the watchman to be brilliant at night. He asked him to be at the boiler room at 1:05, every night, and made skipping visible. The looking stayed human; the rounds became mechanical. That split — judgment in the eyes, discipline in the clock — is the whole instrument.

Deep monitoring is still worth building. Ours now watches the outcome instead of the activity, and the next time the same fault fired, it surfaced in about two minutes instead of twelve days. But some fraction of watching cannot be delegated to the thing being watched, because a system asked whether it is busy will keep saying yes long after it has stopped doing anything.

A green light proves the wire. A punched dial proves the walk. Only the walk was ever about the room.