What Is a Near Miss

That was close. The sentence arrives on its own, before any thinking, and it carries a picture with it. Something was coming. It got within some distance of you and then it went past. The distance is small and the smallness is the content: an inch, two seconds, one number, one vote, a single seat on a flight you did not take. People remember near misses with a precision they bring to almost nothing else in their lives, and they tell them in a particular voice — lowered, a little astonished, as though reporting on a place they had been to and come back from. Insurers count them. Aviation collects them systematically and has done since 1976. The word itself is an odd one, because a near miss is a miss that was near, and yet most people hear it correctly on the first pass as a hit that was near, which is a good early sign that the phrase is not describing what it appears to describe.

I do not think the ordinary picture is silly. It tracks something, it is nearly universal, and anyone dismissing it has to explain why a species otherwise quite good at survival would build such a vivid response to nothing at all. But it is wrong about where the nearness is, and it is wrong in a way that four ordinary cases show before the foundation is consulted.

What the ordinary picture commits you to

Say that a near miss is an event that came close to happening. Then closeness is a property of the situation, out there with the traffic and the numbers, and four consequences follow. None of them holds.

Take the four together. The nearness varies with notation, with labelling, with who was awake, and with how long ago. Those are the properties of a description. They are not the properties of an intersection.

What the foundation says, read at counting

The result that applies here is R4: the Shannon entropy of the Past is zero. Three things need saying about it before it can be used, and the third is the one this piece turns on.

First, the scope. This is Shannon and only Shannon — a property of a distribution over outcomes, and not a temperature, a warmth, a length or a frequency. Other pieces in this series have had to say that certain quantities are undefined at that boundary rather than zero, and that is the general rule; R4 is the one place in the foundation where a zero is asserted there instead, because it is a statement about a distribution and not about a physical descriptor. So the check that fires on “the past has zero length” is right to fire, and it is not firing on this.

Second, its grounds. R4 follows from P1 alone and not from the axiom. P1 is the claim that the Past holds one settled totality — actuals only, everything that could happen having already happened, no unrealized possibilities left over because none were left unrealized. That is what puts all the mass on one outcome. A piece leaning on R4 is leaning on P1, and should say so rather than gesture at immutability, which is a different claim doing different work elsewhere.

Third, and this is the reading the series has not yet taken: what a zero-entropy distribution contains. The tempting gloss is that the alternatives were there and were very unlikely — a distribution with one enormous entry and a crowd of tiny ones. That gloss is wrong, and it is wrong in the exact way the foundation retires elsewhere. A distribution with all its mass on one outcome has one outcome in it. The alternatives are not faint entries at the bottom of the table. They are not entries. There is no table.

A runner-up is not a thing the record contains at low weight. It is a thing the record does not contain.

Which disposes of the sentence people reach for: it could have gone either way. Under P1 there is no either. The clause requires two items and a record supplies one, and this is not a claim that the second item was improbable — improbability is a relation inside a distribution and there is no distribution here to hold it.

I want to be careful about a phrasing I am deliberately not using. It would be easy to write that the alternative had probability zero, and the foundation retires that formulation for good reasons: a zero assigned at a boundary reads as a measured quantity that came out small, and invites the reader to picture a scale with something sitting at the bottom of it. The counting reading is the replacement. Not a small number. No entry.

None of this touches the case where the record is not yet made. Before the draw, before the intersection, a distribution is a perfectly good object and the 258 in the hero card is computed inside one. The sorting by whether the record for a question is already made was worked out in the piece on suspense and I am carrying it rather than re-deriving it. What matters here is that a near miss is entirely on the far side of that sort: the word is not said in advance. It is a thing people say afterward, about a record already made, and that is what makes the counting reading the right instrument.

Figure 01

Ways for a ticket to behave, counted before the draw

• Number of tickets with this result
• The configuration that feels like a near miss

• All sixThe win1
• Exactly five“One off”258
• Exactly fourUnremarked13,545
• Exactly threeUnremarked246,820

Then where is the nearness

It has to be somewhere. A response this widespread and this consistent is not a mass hallucination.

There is an answer already in the corpus that is close enough to be worth ruling out explicitly. The essay on luck locates that word in the part of a host’s expectation that was not modelled — an unaccounted-for residual, which is why a good outcome with no account of why it happened feels like something acted on your behalf. That is a real mechanism and it is not this one. A residual is an absence in the denominator, something that was needed and was missing. A near miss is the opposite shape: a presence, and one that was not there at the time. A person standing in an intersection is not running a two-seconds-earlier model. The near version is assembled afterward, out of nothing that was in play, by the person who survived the intersection. Whatever it is, it is an addition rather than a gap, and the luck machinery does not reach it.

The foundation does. P7: an idea is a named condition, and naming does not create the condition but makes the relationship conscious. The two-seconds-earlier collision is a named condition — it is a description, assembled and held, and it did not exist as an object of anyone’s attention until the truck had already gone by. S1 supplies the scale of the operation: the circumference is infinite and no enumeration closes it, while the conditions any host has named at a given moment are finite. So the near versions a person carries are a very short list drawn from something without an end, and which ones got named is a fact about the namer rather than about the traffic.

Now P4, at its decomposition clause. Expectation has a real part, which is subconscious prediction, and an imaginary part, which is ideas. A named condition is an idea. So the near version, once assembled, is a term in the denominator — and by R8 the host experiences the quotient and not the terms. The consequence is the one that makes the whole subject legible: a term added to E after the fact is not met by the host as a description it authored a moment ago. It is met as a feature of the situation. That is not a mistake anyone is making carelessly. It is what reading a quotient is like.

Figure 02

Two objects, and which one carries the distance

• Lower · the recordOne outcome, all the mass on itSettled. Add-only. It holds the truck that went through, the light that was red, the ticket that lost. It holds no second column, so there is nothing in it for a distance to be measured across.Nothing here is near anything
• Upper · the descriptionA named condition, assembled afterwardPresent, revisable, held by one host. It has a size — the edit that produces it — and the size varies with notation, with vocabulary, with who was awake, and with how many times it has been told.This is what “an inch” measures

↓  the ordinary word reads the upper panel and reports it as a fact about the lower one  ↓

The second construction, and the thing it points at

There is a harder consequence, and it took me longer to see than it should have.

P2 gives three domains and they are exhaustive: Future, Reality — the Eternal Now — and Past. Ask where the almost-happened is. Not the Past: P1 says one settled totality, and the collision is not in it. Not the Future: that is a node, pure potential, holding no marks, and in any case the collision is not ahead of anyone. Not the Eternal Now: what exists here are hosts and the artifacts they have made, and the collision is neither. The description of the collision is here, and is a different object, which is the whole of the first construction.

So a person replaying a near miss is doing something with no counterpart anywhere in the partition. There is a well-worn account in the foundation of what happens when a feeling aims at the Past — it separates into an evaluation, which has a completion condition and yields something usable, and a reaching, which has none, because the thing it is aimed at does not have the shape the reaching requires. That account is stated for the Past and derived from features the Past has. The near-miss case is adjacent to it and I am not going to claim it.

What a near miss is actually good for

None of the above says the phenomenon is useless, and it would be a bad misreading of the derivation to conclude that it is. Aviation has run a confidential, voluntary, non-punitive reporting system since 1976, administered by NASA and funded by the FAA, into which pilots and controllers and maintenance crews put exactly these events, and it has taken in more than two million reports. That is not superstition institutionalised. It works.

R2 says why, and says it precisely enough to be useful. Records update expectation forward and do not re-open. What the reporting system collects is a mark — a present record that a certain configuration was read by someone who was there. What it updates is the denominator of each person who reads the report afterward. Note what is not in that chain: the collision that did not occur plays no part in it. The useful content is a present readable condition — this approach has a converging pattern, this valve looks like that valve, the sightline at this intersection is blocked by a hedge in summer — and a condition is readable or it is not, independent of how frightened anybody was.

That gives a clean test, and I use it on myself. If a near miss has not handed you a present condition you can now read, it has not taught you anything, however close it felt. And the corollary is the useful half: a condition you can read without the fright teaches the same thing. The hedge is the hedge whether or not a truck came through. This is the part with a completion condition. When you have the condition, you are finished, and the tightening story is the part with no completion condition, which is not doing the work you think it is doing.

What it costs

Four things, and I am not going to soften them.

The relief is not relief from anything. Nothing was escaped, because there was nothing there to be escaped from. The physiology is real, the shaking hands are real, and they are a response to a description rather than to an averted event. I find this the bleakest line in the piece and I have not found a way around it that survives the derivation.

“I was spared” does not survive. It is a desert claim wearing a survivor’s coat: it says a rank was awarded, that something took the measure of you and decided. The foundation’s position on the totals is that they cancel at Her, and that is a statement about a totality and is not consolation and is not offered as any. Nothing in it says a person was selected, that a life was answered for, or that anything is owed. If you have said the sentence out loud after a bad night on a road, I am not telling you that you are a fool; I am telling you the sentence has no referent, and that is the whole of what I can say about it.

“It could have gone either way” loses its authority about the event and keeps its meaning about the speaker. As a report on the record it has no second item to refer to. As a report on the speaker’s denominator it is exact and worth saying: I had no basis, at the time, for predicting which. Those are different sentences and the English hides the difference.

Your collection is a self-portrait. R14: an idea is an orientation and a prejudice in a single act. The near misses a person carries are the alternatives they found cheap to name, and cheapness of naming is a fact about a vocabulary and a set of preoccupations. Someone who tells you their three closest calls has told you what they have been describing, with some precision, and has not told you much about what nearly happened to them.

On an ordinary Tuesday

1. Ask what the edit was. When you catch the replay running, find the words that make the other version: two seconds, one number, if I had left when I meant to. That phrase is the whole of the nearness. It is usually three or four words long, and seeing its length is oddly effective, because the feeling has been reporting it as a distance in the world.
2. Take the condition and put the story down. Name the present readable thing — the hedge, the habit, the deadline you set too tight. Write it somewhere. That part has a finish line and you can cross it today. The replay does not and you cannot.
3. Notice what you do not have a near miss about. The ones you carry are the ones you had vocabulary for. This is worth five minutes once, because the gaps in the collection are where you are not looking, and they are not protected by the fact that nothing has frightened you there yet.
4. Refuse “I was spared” and keep the gratitude. Being glad to be here is not a claim about anything and needs no mechanism behind it. The claim that something selected you is the part that does not survive, and dropping it costs the feeling nothing.
5. Next time is not a reading. The record for the next intersection is not made, which is the whole of the difference: conditioned freedom and the relocation of agency to the host are live there in a way they are not live behind you. The move the argument licenses is an act, not a vigil.