The Man Without Qualities Gets the Job

The complaint arrives at every dinner table, usually after the second glass. There used to be characters. In politics, on boards, in editorial offices. Now the same person seems to hold every one of those jobs: prepared, pleasant, unobjectionable, interchangeable. The observation is fair. The explanation that comes attached to it — that people got worse — is not, and it has the shape of every bad explanation: it needs someone to blame, and it predicts nothing.

It is also untestable in the way that should always make you suspicious. Human beings did not change in forty years. Filters did. And a filter, unlike a generation, is a thing you can write down and count.

The tool below is that count. It takes three numbers — how many people can end an appointment on their own, how readily any one of them objects to any one thing, and how much there is on an ordinary candidate's file — and returns what comes out the other end. Push the vetoes up and watch what actually happens to the field. The middle stays exactly where it is. Both ends vanish.

Nobody can say yes alone

Start with the rule that is never written down anywhere. In almost every modern route upward, no single person can appoint you, and almost every single person can end you. The local branch can end you. The list can end you. So can the caucus, the committee, HR, the supervisory board, compliance, the press office, one editor, one thread. Not one of them can make you. Each one, alone, can finish you.

A selection run under effective unanimity does not pick the best candidate. It picks the least objectionable one. That is not a moral claim about anybody involved, and it does not require a single person in the room to be acting badly. It is arithmetic, and the arithmetic is one line long.

Call r the chance that any given gatekeeper objects to any given thing about you. Then 1 − r is the chance they let it pass, and a candidate clears n independent gates with probability (1 − r) to the power of n. Twelve gates at a one-in-four objection rate: 3 percent. Twelve gates at one in eight: 20 percent. Twelve gates at zero: a hundred percent, and exactly one kind of person has an objection rate of zero.

You do not need a conspiracy to strip character out of a process. You need twelve reasonable people with a veto and one candidate who has qualities.

This is the same multiplication that runs through a chain of argument, where seven links you are each 90% sure of leave you 48% sure of the conclusion, and through a secret, where four hundred thousand people who are each 99.9996% reliable cannot hold one for four years. A product of numbers below one only ever goes one way, and it goes there faster than anybody's intuition for it.

Character is a count, and it sits in the exponent

The one-line version has a hole in it, and the hole is where the argument actually lives. It treats a candidate as having one thing about them. Nobody has one thing about them.

So define character the way a selection process is forced to define it: as a count of items on a file that a stranger could reasonably object to. Not virtues. Not depth. Just countable, retrievable facts. The German essay lists six, along with what each one becomes once it is written in a minute.

• A position everyone knows they hold — filed as committed.
• A mistake they talk about openly — filed as compromised.
• Contradicted their own side once, in public — filed as disloyal.
• Two years missing from the CV — filed as unexplained.
• One sentence that stuck to them — filed as a target.
• Real backing in exactly one camp — filed as not a consensus candidate.

Read the right-hand side again. Every one of those objections is reasonable. Every one of them would read well in the minutes. None of them is malice. And that is the whole mechanism: a quality nobody can object to is not a quality. Character is definitionally the surface somebody can catch on.

Now count. If each of n gates looks at each of k items and objects with probability r, then getting through is (1 − r) to the power of n times k. The exponent is gates multiplied by items. Character does not lower your odds. It multiplies the thing that lowers your odds.

Nothing in the process is aimed at character. Character is simply what puts you in the exponent.

What comes out the other end

One candidate is an anecdote. The claim worth testing is about the whole field, so the model needs a population. How many objectionable items does a person have? They are rare, roughly independent facts that accumulate over a life, which is the standard description of a Poisson count — the same shape used for arrivals at a traffic light or leaks out of a conspiracy. Say a candidate has on average λ of them.

Then the whole thing collapses into one line of algebra, and it is a small piece of luck that it does. Write ρ = (1 − r) to the power of n for the chance a single item survives the whole process. A candidate with k items gets through with probability ρ to the power of k. Multiply that by the Poisson prior, renormalise, and the survivors are Poisson again — with mean λρ. Exactly. Not approximately, not in simulation.

Set it at the German essay's own numbers — twelve gates, a one-in-four objection rate, an average of three items on file — and the machine reports the following. 5.5 percent of everyone who started gets through. Of those who do, 91 percent have nothing on file at all. Not a light file. Not a manageable file. Nothing.

The middle never moved

Here is the part that turns the dinner-table complaint from a grumble into a claim you can be wrong about. Give each item on the file an equal chance of cutting well or cutting badly — the position that turns out to be right, or the position that turns out to be a liability; the mistake that made someone wiser, or the one that repeats. Nobody at the gate knows which. Then a person's eventual outcome is a short random walk with as many steps as they have items.

Which means the average outcome of the field is unchanged by construction, at every setting of every dial. The spread is not. Variance scales with the number of items, so it falls from λ to λρ, and the standard deviation falls by the square root of ρ. At the default settings that is 18 percent of the original spread. Not a worse field. A thinner one.

The average did not fall. The distribution got narrow. Those are different complaints, and only one of them is true.

The comparison people actually make is broken in a way that hides this. When we say back then, we mean three names. Five at the outside. We set today's median against 1972's outliers, and the three hundred forgettable people sitting next to those outliers have dropped out of the comparison without dropping out of history. Correct the comparison and the loss does not disappear — it gets sharper, because now it is a claim about a shape rather than a level.

Count the ends. At the default settings, 22 percent of the unfiltered field lands more than two steps from the median in one direction or the other. After the process: half of one percent. Roughly one in four becomes roughly one in two hundred.

The insurance works. That is the problem.

Read that last number in both directions, because the model has no way to read it in only one. Both ends fall by the same factor, always, at every setting. There is no dial anywhere in this thing that shortens the left tail without shortening the right one by exactly as much, and there is no such dial in the real process either, for the same reason: an objection does not arrive labelled with which way the person was going to turn out.

So the process bought insurance, and the insurance is real. Fewer boards captured by somebody's grandiosity. Fewer premiers with a drinking problem and a firearms licence. Fewer people who drive a functioning company into the ground out of sheer self-belief. Anyone who thinks the narrowing is pure loss should sit with that list for a moment, because it is the reason nobody will ever vote to widen the distribution.

The premium was paid at the other end, and the premium never appeared on an agenda. It could not have. A prevented catastrophe leaves evidence — the competitor who did hire that person, and the wreckage you can point at. A prevented great appointment leaves nothing at all. There is no counterfactual file. Nobody has ever written a report on the person they did not hire and what that person would have built.

You can count the disasters you avoided. Nobody has ever counted the giants.

This is the same asymmetry that runs the cemetery: one side of the ledger gives talks and the other side is silent, and a system that only ever hears from one side will keep concluding it is doing well. The feedback loop that would correct the process is missing the entire cost term. That is not an oversight anybody made. It is a property of what can be observed.

The second filter, and why it moves the same number

There is a newer filter stacked on top, and it works on the other variable. A biography used to have an effective statute of limitations, not because anyone was generous but because looking things up was expensive. Something foolish said in 1983 required somebody to drive to an archive, find the microfilm and pull a copy. Almost nobody ever did. The past was not forgiven. It was merely uneconomical.

That cost is now zero. Nobody increased the penalty; the search got cheap. The outcome looks identical and has a completely different cause — and it changes what is rational. If every utterance is permanently retrievable and any utterance can eventually be used against you, the optimal strategy is not to speak well. It is to produce no evidence. The optimal biography is the one with nothing in it.

In the model that is not a change to r or to n. It is a change to λ: candidates now arrive at the first gate with fewer items on file, on purpose, having learned before they were twenty-five that a file is a liability. Both terms of λρ are being pushed down at once, by two mechanisms that have nothing to do with each other. People starting out today are not more cowardly than their parents were. They are better informed about the price.

Elias, and why nobody is to blame

Norbert Elias found the best word for this behaviour in a place nobody looks for it: at court. His point, compressed: the more densely people depend on one another, the further ahead they must control their own conduct — not out of decency, but because in a dense network every impulse has consequences you can no longer see the end of. He called it the compulsion toward self-restraint, and the figure he drew with it is the courtier, a person whose most important instrument is the damping of his own affect.

Versailles did not defeat the nobility. It disarmed them and paid them in ceremony. Whoever still flared up there lost; whoever held himself in rose. Within two generations a warrior caste had become a society of observers studying each other at dinner. The court Elias described had a few thousand members. Ours has a few billion, permanently, verifiably, with minutes. The courtier is not a historical type any more. He is the ordinary case, and adaptation is not a moral failing of the present. It is the price of density.

Musil got there first, and he got there by inversion

Robert Musil began publishing Der Mann ohne Eigenschaften in 1930 and died in 1942 without finishing it. Its thirty-ninth chapter carries the joke the whole novel turns on, and it is a better summary of this arithmetic than anything written since: a man without qualities consists of qualities without a man.

Musil meant something specific by it. His Ulrich is not empty; he is unwilling to be fixed into a set of properties, in a world where the properties have detached themselves and float free of anyone. "A world of qualities without a man has arisen," the chapter says, "of experiences without the person who experiences them." He was describing the modern subject dissolving into a system of formulas of possible meanings — the phrase is his — decades before anybody had a CV parser.

Which is exactly what comes out of the filter. Not empty people. A dossier full of qualities — competencies, certifications, stations, endorsements, every one of them real and none of them attached to anybody in particular, because the one thing the process is structurally able to reward is the absence of a person you could catch on.

The loop that keeps it from correcting

The two halves close on each other, and that circle is the actual reason nothing fixes itself. The people at the top got there by clearing twelve vetoes, which means they are by construction the people nothing sticks to. Nobody can attach themselves to a person nothing sticks to, so no bond forms. Without a bond there is no constituency that defends anyone after a first bad week. Without that constituency, any quality at all is an uninsured risk — and nobody carries an uninsured risk.

So the only people who survive at the top are the ones who never needed cover. Character is not a property somebody has. It is a risk somebody is carrying. While nobody is willing to carry it, nobody can have it. That is the entire diagnosis, and it needs no decline, no decadence, and not one claim that people used to be better. It needs two ingredients: many vetoes and little cover.

And now you

Up to here it has been about cabinets and boards, which is the comfortable version, because it is far away. The uncomfortable version: the same machine runs at forty employees, and you are the one operating it. It runs whenever a hiring round concludes that there was nothing remarkable. When a promotion is settled by consensus. When a candidate has four conversations and any one of them can end it. When the feedback form says somebody rubs people the wrong way and nobody asks which people, or about what.

None of which is an argument for hiring the bully. The bully is the other end of the same distribution, and the same insurance cut him off for good reasons. Set the tool to one decision-maker and look honestly at what comes back: the giants return, and so do the disasters, in precisely the same proportion. That is the trade. It is a real trade, and the point of doing the arithmetic is to be able to make it deliberately rather than to keep making it by default in one direction.

Four ways this is wrong

The gates are assumed independent, and they are not. Real gatekeepers read the same file and take each other's calls, and correlated objections make the product too harsh: at the limit where every gate objects to the same things, twelve gates behave like one. The honest answer is a band, from (1 − r) to the power of n·k at one end to (1 − r) to the power of k at the other. The tool prints both. Note which end of that band requires the gatekeepers to be thinking for themselves.

An item on your file is not a coin flip. The model gives each one an equal chance of cutting well or badly, which is plainly false of any specific item — some really are just liabilities. But that is the assumption doing the least work, because the claim is about what the filter can see, and the filter sees an objection, not a direction. A process that could reliably tell the two apart would be a good process, and would not be this one.

All three dials are assumptions. Nobody has measured the real number of vetoes in a real appointment, or the objection rate, or how much sits on an average file. The German essay says so about its own twelve, and it is right to. What survives the uncertainty is the shape, not the level: at every setting the middle stays put, and the two ends move together.

Selection is not only on objections. People are also chosen for competence, and competence is not in this model at all. That is deliberate, and it is the honest limit of the thing: the model does not claim the people who got through are bad at their jobs. Most of them are extremely good at their jobs. It claims something narrower and harder to argue with — that once absence-of-objection is in the product, the variance of everything else goes with it.

And there is a fifth, which is not a flaw in the model but the reason the argument is hard to win. The cost side of this ledger is unobservable in principle. Nobody can produce the list of people who would have been remarkable, because they were filtered at the branch meeting in 2011 and became competent regional managers instead. Anyone defending the process can therefore always demand evidence that cannot exist. The arithmetic is what you have instead of that evidence — not a measurement, but a reason to expect the missing column to be exactly as large as the one you can see.

We did not select the wrong people. We built a procedure with one possible output and then expressed surprise that nobody notices him. Push the vetoes around yourself in The Unanimity Filter. The Chain of Argument is the same multiplication running over the steps of a case instead of the gates of a career, and The Cemetery is what happens to your reasoning when only one side of the ledger is allowed to speak.