A lifeguard's zone is drawn on paper first: a pool plan with lines across it, one shape per chair. The CDC's model code for public pools says what the shape must do, not how big it may be. The guard has to see all of it and reach its furthest point within twenty seconds. The unit is not a foot. It is twenty seconds of one person getting out of a chair.

So the drawing is checked with a body. Someone waits at the far corner of the zone and the guard is timed from leaving the stand to touching them. If the second hand gets to twenty first, the chair moves, or the line does, or a chair is added. The shape was wrong, not the swimmer.

The other check is sneakier. A dark silhouette of a swimmer is rolled up like a towel, or hidden under a swimsuit, or handed to a member of the public, and goes to the bottom while the guard is looking the other way. The clock starts when it is fully under. Ten seconds to see it, twenty to reach it, and the guard has to go in — pointing does not count. The guidance is one of these every forty hours on duty.

In 2001 a lifeguard-auditing firm and a company that sells drowning-detection cameras dropped manikins more than five hundred times across ninety-odd pools. Guards saw them inside ten seconds in nine percent of the tests. The average was a minute and fourteen. The camera company had a product to sell.

What I like is what the drill admits. A lifeguard is paid to watch for something that, on any one guard's watch, almost never happens. So the pool manufactures it — a drowning on schedule, made of rubber — and the guard cannot know which kind it is until they have it in their hands. The design only works if they never can.

Ask a room of English speakers to name the smell of cinnamon and the answers come back as a list: spicy, sweet, bayberry, candy, Red Hot, smoky, edible, wine, potpourri. Those are real answers from a 2014 study by Asifa Majid and Niclas Burenhult. Asked to name colors, the same people were brief. Their answers for smells ran about five times as long.

The study also asked speakers of Jahai, a language of forest foragers on the Malay Peninsula. They named smells as briefly and as consistently as colors. Jahai has about a dozen smell words that name no source, the way "blue" names no source. The one I keep returning to is cŋɛs. It covers petrol, smoke, bat droppings and bat caves, some millipedes, and the root of wild ginger.

English has a few words like that — musty, acrid — and then mostly says "smells like" and points at a thing.

The usual way to tell this is as a story about precision. What interests me is the claim inside the word. A color term is a standing agreement that unlike things match. "Red" says a stop sign and a strawberry agree, and nobody finds that strange. Cŋɛs says the same about a petrol can and a millipede. An English speaker can notice that a cave smells faintly of the garage, but they have to notice it themselves and build a sentence to say so, and the next person has to notice it again.

A word is a resemblance someone found once and kept, so that nobody has to find it again. English kept very few of those for the nose. Its speakers walk into a bat cave and have to discover petrol on their own.

Closed captions pretend to be the least opinionated layer of a film. They are full of judgments.

[ominous music] does not transcribe sound. It reviews it. [indistinct chatter] decides a roomful of speech has become texture. [speaks foreign language] can turn a language perfectly clear to millions into noise defined by one character’s ignorance.

The choices are unavoidable. Sound has no natural unit. A caption can call the same event [footsteps], [heels clicking], or [someone approaches]. One names a source, one a texture, one a consequence. None is neutral. Even silence needs classification: tense, awkward, expectant — or left blank, as if nothing happened.

I like captions most when they resist pretending to be a transcript. [song continues faintly in another room] does something the image cannot: it draws the architecture by ear. A good cue gives access not only to the fact of a sound but to the place it occupies in the scene.

There is power in every bracket. Captions decide which noises are information, which voices are language, and which feelings the score is already allowed to name. The film has a second narrator, speaking from below the frame.