Andrew Paul Klein

At a rock shelter called Abric Romaní, near Capellades in Catalonia, archaeologists have been excavating since 1983. Across sixteen levels spanning roughly 62,000 to 40,000 years ago, they recovered 597 wooden remains and imprints. Thirteen were identified as things Neanderthals made or used: a thrusting spear two and a half metres long, a short-hafted club with scrape marks along its head, four pieces of charred bark sitting near hearths, a crossed-stick frame, a branchless pole of five metres lying beside five hearths in what earlier work read as a sleeping area.

The wood itself is gone. Calcium carbonate built up around it, and when the organic material decayed it left hollow cavities in the travertine — molds that replicate the original shape of the wood. Some are fine enough to preserve the microscopic anatomy: tracheids, cross-field pits.

The excavators are careful about what a mould can show. The paper concedes the difficulty of analysing manufacturing processes from a cast. It notes that the pointed end of the long spear is, in its own words, “unfortunately, the most poorly preserved.” It acknowledges the difficulty of distinguishing deliberately worked bark containers from bark fragments that detached naturally. Thirteen objects out of 597. The record is thin, and the paper says so.

What survives is not the object but the space the object occupied. A precise record of a form, and no record at all of the substance that gave the form its weight.

The prehistorians know exactly what they are holding. A cavity. Honestly described. Limits stated.

We do the same thing with numbers, and we are less careful about it.

In 1972, the pollster Daniel Yankelovich named a pattern of reasoning after Robert McNamara, who had brought it to the United States Department of Défense from the Ford Motor Company. The pattern has four steps.

Measure whatever can be easily measured. Disregard that which cannot be easily measured, or give it an arbitrary quantitative value. Presume that what cannot be easily measured is not really important. Say that what cannot be easily measured does not exist.

Yankelovich called it the McNamara fallacy. It is not a description of stupidity. McNamara was, by any measure, a formidably intelligent man. It is a description of what happens when a quantitative apparatus is treated as a description of the world rather than as a lens trained on part of it.

The canonical example is the body count. The United States measured its progress in Vietnam by the number of enemy dead. The figure was quantifiable, reportable, and could be aggregated upward into briefings. It was also, as a measure of whether the war was being won or was worth fighting, close to worthless. But once it was installed as the primary indicator, everything that could not be expressed in it — the political allegiances of villagers, the durability of the insurgency, the question of whether any of it should be happening — became invisible to the apparatus.

The metric did not lie about what it measured. It reported the count with precision. The error lay in the reading: in the assumption that what had been counted was what mattered.

Note the third and fourth steps, because they are where the damage is done. The failure is not in choosing a poor measure. It is in what happens afterwards — when the thing that cannot be measured is first presumed unimportant, and then treated as though it were not there at all. Nobody decides this. It is the ordinary operation of an instrument that records what it can register and is silent about the rest.

And the silence gets read as absence. As though the thing not counted had been examined and found wanting, rather than never examined at all.

In October 1966, McNamara’s office announced that the armed forces would accept 100,000 additional men a year who would previously have been rejected — men who had failed the mental or physical standards, drawn disproportionately from poor and minority communities. The programme was called Project 100,000. The men who entered through it were officially described as “New Standards Men.”

The logic was optimisation. The systems analysts in McNamara’s office understood the war as a production problem with inputs and constraints. Manpower was a constraint, and the constraint could be relaxed by lowering the threshold. The men were, in the administrative vocabulary of the time, “substandard” — a word that names a person by their failure to fit a specification.

What happened to them is documented, though the precise figures are contested and should be treated with care. Approximately 354,000 men were inducted through the programme over five years. They were assigned disproportionately to combat units. Their casualty rates were higher than those of other soldiers — by some accounts substantially higher, though the studies differ and the range should be stated rather than a single figure seized on. Many had intellectual disabilities or limited literacy. Some could not read their own orders. The programme was ended in 1971. Estimates of those who died run into the thousands, and the uncertainty about the number is itself part of the indictment, because nobody was counting carefully enough to know.

Two things are worth naming about this.

The first is the euphemism. “New Standards Men” is a phrase that hides what it describes behind a technical term. The vernacular name was crueller — “McNamara’s Morons” — and there is a dark honesty in it. The official name concealed; the cruel name saw. Sometimes the accurate naming of a thing is done by the people being harmed, and the institution’s own vocabulary is the more deceptive of the two.

The second is that nobody in the chain of decision was doing anything other than their job. McNamara was applying methods that had made him successful in industry. The analysts were optimising a system. The recruiters were meeting quotas. The officers were filling units. Each step was defensible in the language of the institution, and the cumulative result was the systematic exposure of the most vulnerable young men in the country to the highest risk of death.

The metric that organised this — acceptable induction standards — was accurate. It was also a mould, and the men were poured into it.

In Australia in 2026, an older person applying for a home care package completes an assessment. The questionnaire has more than five hundred questions. It was recommended by a royal commission, tested on twenty-two thousand people, developed with clinical input, and has a published trial report.

The answers are then scored by an algorithm. The algorithm is a separate thing. It was never recommended by the royal commission. It underwent no live trial before launch. The department that built it has been unable to name the clinicians who signed off on it, and has refused to release the documents that would show how it was developed.

A professor who designed the classification system used to fund residential aged care reverse-engineered it. Her findings: the vast majority of the five hundred questions do not feed into the algorithm. The detailed clinical notes recorded in more than a hundred text boxes cannot be read by it. Frailty, cognitive problems, mental health, carer fatigue — these enter the calculation, in her words, “far too late” to affect funding.

A one-point difference at a particular juncture equates to almost forty thousand dollars a year. In one documented case, two people with otherwise identical needs but functional scores of 24 and 25 were allocated to funding levels eight and five. A fifty-one per cent difference in the care they would receive, from a single point.

The assessor records. The algorithm decides. The record — the observation, the person’s own account of their life, the carer’s exhaustion — is collected and then not used.

Nobody decided that frailty did not matter. The apparatus made the decision by being unable to register it.

This is what Yankelovich’s third and fourth steps look like when they are performed by a procurement system rather than by a person. The material that cannot be processed is first given no weight, then treated as unimportant, then treated as though it were not there. The woman who was managing last year and is not managing this year becomes a score that has no category for the change. The score is accurate. The change is real. The two never meet.

The temptation here is to look for the villain, and the temptation should be resisted, because the villain is not there and looking for one is how the argument gets dismissed.

The aged care algorithm was not imposed by a cabal. It emerged from a procurement process. A prototype, believed to have been designed by a consultancy, was taken in-house, modified and validated by public servants, and launched. The public servants were asked to do work of a technical sophistication they had not been trained for. The department could not name the clinicians who signed off because the sign-off process was not structured to produce names. The minister defends the system by appealing to a royal commission that recommended something else.

Each person in that chain can give an account of their conduct that is internally coherent. The consultant delivered a product. The public servants implemented a policy. The department followed procurement rules. The minister relied on advice. The parliament passed the legislation. No single point in the chain is a place where someone decided to harm old people.

And yet old people are being harmed, systematically, by an apparatus staffed by ordinary people doing their jobs.

This is the thing that is harder to face than malice, and it is the reason the essay is titled as it is. A monster can be found and stopped. A structure cannot, because there is no point at which to apply the lever, and because everyone inside it is, in their own understanding, innocent.

Hannah Arendt’s phrase for this — the banality of evil — is often misunderstood. It does not mean the evil is trivial. It means the evil is ordinary: performed by people who are not exceptional in their cruelty, who would be genuinely surprised to be told what they were part of, and who are protected by the ordinariness of what they did.

What holds the structure together is not secrecy. It is something closer to unfalsifiability.

There was no live trial of the algorithm. The clinicians who signed off cannot be named. The documents cannot be released. The assessor cannot override the score. The algorithm was changed substantially in July 2026 — thirteen of the thirty-three functional-score items removed, three introduced — invalidating whatever validation had existed. The appeal is heard by the department that built the system. The legislated escalation pathway operates after the decision, places the burden on the older person or their family, and was described by one crossbench member as “a bandaid patch-up.”

Each of these is defensible in isolation. Together they form a structure in which the question is this working? has no institutional answer.

That is not an accident of implementation. It is the property that allows a failing system to describe itself as reforming. And it is the same property, in a different form, that allowed the men of Project 100,000 to be inducted: the standards were met, the quotas were filled, the reporting was accurate, and no mechanism existed by which the question are these men dying because of what we did? could be asked inside the apparatus in a way that would produce an answer.

Bandura called the mechanism displacement of responsibility — the transfer of accountability onto a procedure, a superior, a system, until no individual holds it. In the older literature this is described as something people do in speech. What the aged care system shows is that it can be built into a procurement process. The algorithm does the displacing. Nobody has to give the speech.

It is not enough to say that metrics miss what matters. That objection, left on its own, becomes an argument against measurement itself, which hands the field to whoever controls the numbers.

The constructive question is sharper. What would a measure that captured lived experience actually require?

Three things.

First, it must be qualitative at intake. The person’s own account — of what they can and cannot do, of what has changed, of what they fear — must enter the record as data, weighted and considered, not as free text that no formula can read. This is not sentimentality. It is an engineering requirement. A system that cannot take in the primary evidence of the person it is assessing is operating on a subset of the available information, and will therefore misallocate, however consistently.

Second, it must be participatory. The person assessed must have standing, not merely consent. Consent is passive: it permits a decision someone else has made. Standing is active: it gives the person a right to contest the characterisation of their own life, to have that contestation heard by someone with authority to change the outcome, and to be told why. A review process that the person cannot practically reach is not a safeguard. It is a form of words.

Third, it must be correctable by the person it measures. This requires an override that can be exercised by a human with authority, and it requires a plurality of measures. The second follows from Goodhart’s law — when a measure becomes a target, it ceases to be a good measure — and from the more basic point that a single metric is a single mold. Any system that reduces a person to one number will, in time, be gamed by institutions and will misdescribe the people who do not fit the shape. The answer is not a better single number. It is several measures, held in tension, with human judgement given the final word and required to justify it.

None of these three requirements is expensive. None is technically difficult. What each of them costs is speed and control — the two things procurement systems exist to maximise. A qualitative intake is slower than a score. Participatory standing is slower than a determination. Correction by the person measured is slower than an appeal process most people abandon.

That is the honest formulation, and it is more damning than a claim of cruelty, because it names a property of the structure rather than a failing of the people inside it. The mold is not a mistake. The mold is what the instrument produces. An institution that has built its operations around the mold will keep producing it, correctly and consistently, for as long as the instrument is the thing that counts.

There are people who profit, knowingly, and the structure gives them cover and rewards them for it. This essay is not an argument that nobody is culpable. It is an argument that culpability is distributed so widely, and hidden so effectively behind procedure, that the structure functions as a machine for producing harm without producing perpetrators.

The people inside it are not innocent. They are also not the explanation.

The excavators at Abric Romaní know what they are holding. A cavity, honestly described, its limits stated. They do not present the shape as the thing. They do not report thirteen wooden objects and call it the whole of Neanderthal life. They say: this is what survived, and here is what a mold cannot show.

That is the standard. Not perfection. Honesty about what the instrument can and cannot register, and a refusal to let the cavity stand in for the thing that decayed.

The question this leaves is not addressed to governments, because governments answer in the language of their instruments and the answer will always be a number. It is addressed to the rest of us, and it is narrower than it sounds.

The next time you are handed a figure about a person — a score, a rating, a funding level, a risk assessment, a performance target — ask two questions. What did the instrument measure? And what could it not see?

Then ask the third question, which is the one that matters.

Who decided that what it could not see was not important?

Andrew Paul Klein

The Patrician’s Watch