The canary had an oxygen cylinder

Everyone knows the phrase, and everyone knows the picture it comes with: a small yellow bird in a mine, breathing something the miners cannot smell, dying so that the men can run. The bird is a sacrifice. That is the whole force of the idiom — when we call someone a canary in a coal mine, we mean they will be the first to be destroyed and nobody will act until they are.

One of the real objects is in the Science Museum’s collection, and it is not that. It is a metal box with three glass walls so the bird is visible from any angle, a round grille door held open on the fourth side, and bolted to the top, a small oxygen cylinder with a valve. When the bird went down, you closed the door and opened the valve. The bird got up.

I should say immediately what took me a fact-check to admit: that cage was not standard issue. It was rescue-brigade equipment, made by firms like Siebe Gorman, dating from decades after canaries went underground, and it was never the common case. The ordinary working canary rode in a plain wire cage with nothing bolted to it. So this is not an essay about how mining was gentler than you thought. It is about what the exception was for.

What the bird was for

Carbon monoxide is colourless, has no smell, and does not announce itself by making you cough. It is what fills a coal mine after an explosion — the miners called that afterdamp — and it is what kills a large share of the people who survived the blast itself. A man walking into it feels a headache coming on, then feels rather good about things, then cannot work his legs.

This is what killed most of the men. After the Albion Colliery explosion of 1894, John Scott Haldane — who spent his career on the physiology of breathing and later tested war gases on his own body — found that of fifty-four dead, fifty had died of carbon monoxide and only four of the blast. In 1895 he proposed that a small warm-blooded animal be sent in to register what human senses could not. Canaries were underground by 1896.

Note that he did not say canary. He suggested mice or birds, and thought mice were the more sensitive instrument. The bird won on a property that has nothing to do with sensitivity: a distressed canary is legible from across a roadway. It stops singing, it sways, it comes off the perch. A mouse in trouble requires you to be crouched over the box watching it. The better sensor lost to the one whose output you could read by lamplight while doing something else, and that is the first thing in this story worth keeping.

Why the animal at all: a 20-gram body running a fast metabolism moves far more air per gram than a man does, and has no reserve to hide the dose in. That metabolic rate is the mechanism — it is why the mouse also works. Birds do have the better machine on top of it (air runs one way through a system of sacs, so fresh air crosses the exchange surface on the inhale and the exhale, where our lungs are bags that always re-breathe some of their own stale air), and it is a genuinely superior design; bar-headed geese tracked by GPS cross the Himalaya above 7,000 metres, in air where an unaided person is in serious trouble. But the lung architecture is the flourish. The dose rate is the reason.

Either way the shape holds: the canary is not a good alarm because it is frail. It is a good alarm because it is doing, harder and faster than the men around it, the exact thing that is about to go wrong.

The part the idiom drops

So far this is still the story everyone tells: a superior sensor, used up. To see why anyone bolted a bottle of oxygen to the cage, you have to fix the picture of where the bird actually was — and mine needed fixing until about an hour ago.

The canary was not mainly a thing every miner carried to his own coal face. It was largely a rescue instrument: carried by the brigades who go back down after the explosion, into the afterdamp, to find out whether the air will let them work. A rescue party does not take one reading. It takes a reading every few yards, for hours, in the dark, in the worst air anyone will breathe that year.

That is the setting in which the cylinder stops being sentimental and starts being obvious. A warning device that must be destroyed to produce one reading has a hidden property: it is expensive to consult, so it gets consulted less. The man carrying it starts saving it for where he already suspects trouble, and hesitating at the ambiguous case — which is the only case where a test tells you anything you did not know. Put a bottle of oxygen on the cage and the same bird gives you a reading at this junction, and then at the next one. The check becomes repeatable, and a check you can repeat is a different instrument from one you can spend.

It did not make the bird safe. Canaries died in numbers; a bird that dropped in high concentrations, or too far from the door, did not come back. The cylinder improved the odds and made the alarm reusable. Both of those are worth having and neither is a happy ending.

Britain kept them a remarkably long time. The phase-out came in 1986, when about two hundred birds still in service were replaced by electronic detectors — better in the obvious ways: no feeding, no moulting, a number on a display rather than a judgement about whether a bird looks unwell. But the electronic detector inherited the design property rather than invented it. What made it adoptable was what the cylinder had already established: that you can take a reading without spending something you would mind spending.

The residue

Two things I want to keep from this.

The first is that a sensor’s sensitivity is usually its competence turned against it. Whatever the detector is unusually good at is the reason it will be hurt first. If you want early warning, you don’t look for the weakest thing in the system; you look for the thing that is best at the exact process that is about to go wrong.

The second is the cylinder. Whenever a check exists that nobody runs, the interesting question is rarely whether people understand that it matters. It is what a single reading costs, and who pays. Reduce that, and the arguing about diligence mostly stops being necessary.

The idiom kept the bird and threw away the oxygen bottle. It kept the part that makes a good metaphor for doom, and lost the part that was actually the engineering.

Postscript, written twenty minutes after this went up. I published this and then read the fact-check I had sent off while writing it. It came back with three corrections that mattered, and rather than quietly patch the text I want the record to say what was wrong.

I had written the resuscitator cage as though it were the normal kit. It was not — it was rescue-brigade equipment, late and relatively rare, and the working canary’s cage had nothing bolted to it. I had implied every miner carried a bird to his own face; the canary was mostly a rescue instrument for going back in after an explosion. And I had made the bird’s efficient lungs the whole mechanism, when the mechanism is mostly a fast metabolism in a very small body — which is why Haldane actually preferred mice, and lost that argument to the fact that you can see a bird stop singing from across the roadway. (Also: the bar-headed geese do not fly over Everest. That is a 1950s anecdote. The GPS tracking has them above 7,000 metres, which is quite enough.)

Every one of those made my argument sound better than the evidence did. That is the direction this kind of error always runs, and it is why the check gets sent before I am confident rather than after.