The Expert Knew More

On the night of October 22, 1707, four Royal Navy warships struck rocks off the Isles of Scilly. Between 1,400 and 2,000 sailors drowned, including Admiral Sir Cloudesley Shovell aboard the ninety-gun HMS Association. The cause was not weather. It was not enemy fire. It was a problem the world's best navigators had been unable to solve for three centuries: they could not determine their east-west position at sea.¹

Seven years later, the British Parliament passed the Longitude Act. The prize was £20,000 for a method accurate to within half a degree of longitude. At the time, £20,000 could build an estate. The scientific establishment, led by the Astronomer Royal and backed by the Royal Society, pursued the solution they understood: astronomical tables. Measure the angular distance between the Moon and a known star. Compare it against pre-computed values. Derive Greenwich time. Calculate longitude from the difference. This was the approach the scientific community trusted. It was mathematical. It was published. Other astronomers found it rigorous.²

The method required a clear sky. In clouds, or when the Moon was hidden, no observation could be made. It required hours of calculation with logarithm tables. The ship kept moving during those hours. And it demanded mathematical ability most sailors did not possess. It was, in every sense, a solution validated by the approval of other astronomers.³

John Harrison was a carpenter's son from Foulby, in the West Riding of Yorkshire. He had no university education. He was not a member of any scientific society. He taught himself clockmaking as a young man, building his first longcase clock at twenty from wood, not brass. In 1730, he traveled to London with an idea for a marine timekeeper that could survive the motion and temperature changes of an ocean voyage.⁴

He walked into the shop of George Graham, London's leading watchmaker. Graham, who by all rights should have dismissed a self-taught provincial carpenter, instead talked with Harrison for ten hours. When Harrison left, Graham had lent him money with no security and no interest to build the prototype.⁵

Harrison spent the next five years in a workshop behind the Royal Oak pub yard in Barrow upon Humber, building a clock that weighed seventy-two pounds. He sailed with it to Lisbon in 1736. The ship's sailing master, using dead reckoning, placed the ship sixty miles east of where it actually was. Harrison's clock had it right. The Board of Longitude granted him £500 to continue his work.⁶

He built a second clock. A third. The third took nineteen years and produced two inventions still in use today, the bimetallic strip and the caged roller bearing, but never achieved the precision he needed. Then Harrison abandoned the large sea clock design entirely. His fourth chronometer, completed in 1759, was not a clock at all. It was an oversized pocket watch, five inches across, with a silver case wound daily by key.⁷

In 1761, Harrison's son William sailed from Portsmouth to Jamaica aboard HMS Deptford carrying the chronometer, designated H4. After eighty-one days at sea, H4 was five-point-one seconds slow. The longitude error was roughly one nautical mile. The Longitude Act required accuracy within thirty. H4 exceeded the requirement by a factor of thirty.⁸

The Board of Longitude refused to pay. They demanded more trials. They changed the conditions. They required Harrison to reveal his methods and dismantle H4 piece by piece before a panel. When the panel confirmed the mechanism was sound, the Board still withheld half the prize, pending proof that other watchmakers could duplicate it.⁹

In 1765, Nevil Maskelyne became Astronomer Royal. The appointment made him an ex officio Commissioner of Longitude. He was now simultaneously the champion of the rival lunar distance method and a judge of Harrison's claims. He wrote that H4 had "built-in errors which just happened to cancel each other out." In May 1766, he collected Harrison's earlier chronometers from his home at Red Lion Square and transported them to the Royal Observatory in an unsprung cart over cobblestone roads. At least one clock arrived damaged.¹⁰

Harrison was seventy-nine years old. He had spent more than forty years on the problem. He had solved it. The evidence was not ambiguous. He built a fifth chronometer and appealed directly to King George III, who installed it at his personal observatory at Kew. Over ten weeks, the King personally supervised daily observations. H5 was accurate to within one-third of a second per day. George III reportedly told Harrison: "By God, Harrison, I'll see you righted."¹¹

In June 1773, Parliament granted Harrison £8,750. His total payments over a lifetime came to roughly £23,065. He was eighty years old. The formal £20,000 prize was never officially awarded to anyone. Harrison died three years later.¹²

The carpenter solved the problem the astronomers could not. The system resisted the solution because it came from the wrong kind of knowing. The chronometer was tested by the ocean. It either kept time at sea or it did not. The astronomical method was tested by other astronomers. They found it elegant, rigorous, theoretically sound. The system honored the knowledge that its own members could evaluate and obstructed the knowledge that only the ocean could judge.

The gradient

There is a passage in Thomas Sowell's Knowledge and Decisions that states the problem in two sentences:¹³

To say that a farm boy knows how to milk a cow is to say that we can send him out to the barn with an empty pail and expect him to return with milk. To say that a criminologist understands crime is not to say that we can send him out with a grant or a law and expect him to return with a lower crime rate. Thomas Sowell, Knowledge and Decisions (1980)

The farm boy's knowledge is tested every morning. Wrong technique, no milk. Repeated failure, no farm. The feedback is immediate and personal. The criminologist's knowledge is tested by whether other criminologists find it interesting. The peer review process evaluates methodology, novelty, and internal consistency. It does not evaluate whether the crime rate went down. The knowledge is authenticated by a different mechanism entirely.

The question runs along a gradient, not a binary. At one end: maximum consequence. The farmer. The engineer whose bridge must hold. The surgeon whose patient must survive. The pilot who must land the aircraft. Their knowledge is validated by what happens next. At the other end: zero consequence. The tenured academic whose ideas are judged by whether colleagues find them compelling. The think tank fellow whose policy recommendation is evaluated by other fellows. The knowledge is validated by peers, not by outcomes. The feedback loop runs through the profession, not through reality.¹⁴

Sowell identified the same gradient in a different passage. "The standards by which engineers and financiers are judged are external standards, beyond the realm of ideas and beyond the control of their peers," he wrote. "An engineer whose bridges or buildings collapse is ruined, as is a financier who goes broke." But "the ultimate test of a deconstructionist's ideas is whether other deconstructionists find those ideas interesting, original, persuasive, elegant, or ingenious. There is no external test."¹⁵

Between the farmer and the academic, the gradient runs through every domain. The financial analyst is tested by returns, but with enough lag and enough noise to obscure errors for years. The management consultant is tested by client satisfaction surveys that measure whether the client liked the presentation, not whether the recommendation worked. The policy advisor is tested by nothing measurable within the advisor's career timeline.

The gradient runs from maximum consequence to zero consequence. The system's prestige hierarchy runs in the opposite direction. The less your knowledge is tested by reality, the higher your status. The person closest to the outcome has the least authority. The person farthest from the outcome has the most. This is the structure Harrison encountered. The clock was tested by the Atlantic. The tables were tested by other astronomers. The system gave authority to the tables.

The gradient is not a theory. It produces measurable results. In 2015, a team of researchers attempted to replicate one hundred published psychology studies from three leading journals. Ninety-seven percent of the original studies had reported statistically significant results. These were findings that had survived peer review, been published, been cited. When the experiments were repeated with the same methods and materials, only thirty-six percent replicated. Nearly two-thirds of the findings vanished the first time they were tested by something other than peer approval.¹⁶

The room

Turbocharge

In May 2013, Purdue Pharma retained McKinsey & Company to conduct a rapid assessment of OxyContin performance. The project was called "Evolve to Excellence." McKinsey had been advising Purdue since 2004. This was nine years in, and six years after Purdue had already pleaded guilty to a federal felony for misbranding OxyContin as less addictive than it was. The guilty plea cost Purdue $600 million. McKinsey re-engaged anyway.¹⁷

McKinsey's consultants recommended that Purdue "turbocharge" OxyContin sales by intensifying marketing to what they called "High Value Prescribers," including doctors who were already prescribing opioids for uses that were unsafe and medically unnecessary. They recommended bypassing retail pharmacy restrictions by delivering OxyContin directly through mail-order pharmacies. They recommended minimizing sales representative discretion in identifying which prescribers to target.¹⁸

The pharmacists had been flagging suspicious prescriptions. They were the people closest to the patient. Their knowledge was authenticated the way the farmer's is: by what happened next. The patient who filled the prescription and came back needing more. The patient who stopped coming back. The pharmacist's judgment was consequence-validated. McKinsey's advice was to route around it.¹⁹

The sales representatives had discretion about which doctors to visit. Some of them, in the field, reading the room, seeing what these prescriptions were doing, used that discretion to avoid the most problematic prescribers. Their knowledge was authenticated by proximity. McKinsey's recommendation was to override their judgment with a spreadsheet. Target the high-volume prescribers regardless.²⁰

The consultants' knowledge was authenticated differently. Partner approval. Engagement fees. Methodological frameworks. Whether other McKinsey partners found the analysis rigorous. The authentication ran through the profession, not through the patients. In August 2013, McKinsey presented to Purdue's board. The value at stake, they said, was "hundreds of millions, not tens of millions." OxyContin revenues tripled from roughly one billion dollars to three billion in the years following the 2007 guilty plea.²¹

In 2017, McKinsey proposed that Purdue offer distributors a rebate of $14,810 for each overdose or opioid use disorder attributable to OxyContin. They projected that in 2019, approximately 2,484 customers at a single pharmacy chain would overdose or develop a dependency. At $14,810 per event, the payout would be $36.8 million. The rebate was proposed. It was not implemented. But the proposal itself reveals what the authentication gradient produces. The patient's overdose was an input in the model, not an outcome that forced the model to change.²²

Between 1999 and 2023, opioid overdoses killed more than 800,000 Americans. McKinsey advised the company at the center of the crisis for fifteen of those years. In 2021, the firm settled with forty-seven states for $573 million. In December 2024, the Department of Justice secured a $650 million settlement and charged McKinsey with conspiracy to misbrand a drug. It was the first time a management consulting firm had been held criminally responsible for advice that led to a client's crime. A senior partner was sentenced to six months in federal prison for destroying documents related to the work.²³

The system gave decision-making authority to the people with the least consequential feedback and removed it from the people with the most. The consultants had the frameworks. The pharmacists had the patients. The consultants won the room. The patients bore the cost.

Not a retailer

In 2000, the board of Home Depot hired Robert Nardelli as chief executive. Nardelli was a protege of Jack Welch at General Electric. He had spent twenty-seven years at GE. He had no experience in retail.²⁴

His knowledge was financial. Process optimization. Cost discipline. Centralized purchasing. Metrics-driven management. This knowledge was authenticated by other GE executives, by Wall Street analysts, by the business press that had elevated GE's management culture to an object of study. Nardelli could articulate a strategy for Home Depot in language the board understood because the board came from the same world.

The people on the floor had a different kind of knowledge. Home Depot had been built by experienced tradespeople who could tell a contractor which grade of lumber to use, which adhesive would hold in humidity, which fitting matched which pipe. This knowledge was authenticated by whether the customer came back. Nardelli replaced many of them with lower-cost part-time workers. Revenue doubled. Net income rose 130 percent. The stock fell nine percent while the chief competitor, Lowe's, rose 188 percent. Customer satisfaction, measured by the American Customer Satisfaction Index, fell to dead last among major United States retailers.²⁵

The person who could articulate won the boardroom. The person who knew which adhesive to use was replaced. Revenue went up. The thing that made the revenue possible went down. The board saw the first number. The customer felt the second.

The pattern extended beyond Home Depot. In 2024, Spencer Stuart's annual Board Index reported that financial expertise accounted for twenty-nine percent of new S&P 500 board appointments, up from eighteen percent in 2008. Profit-and-loss leadership, the kind tested by operational consequences, accounted for eight percent.²⁶

In 2022, the cryptocurrency exchange FTX collapsed. More than sixty investors, including Sequoia Capital, had authenticated Sam Bankman-Fried's knowledge through the usual channels. MIT degree. Quantitative trading background. An effective altruism philosophy that made his ambition sound principled. Sequoia published a thirteen-thousand-word profile of him. They invested $214 million. The company had no chief risk officer. No functioning board oversight. When the bankruptcy administrator, John J. Ray III, examined the wreckage, he described "virtually no internal controls and no separateness whatsoever" between FTX and the hedge fund it was supposed to be separated from. The knowledge that would have caught this, basic auditing, basic reconciliation, basic financial controls, is consequence-validated. It is the kind that returns from the barn with milk or does not. It was never in the room. Thirty-two billion dollars in value was authenticated by peer approval and destroyed by the first external test.²⁷

Worthless

In 2013, Laszlo Bock, then head of People Operations at Google, told the New York Times: "G.P.A.'s are worthless as a criteria for hiring, and test scores are worthless. We found that they don't predict anything." Google had analyzed its own hiring data and found no correlation between academic performance and job performance. Some teams had fourteen percent of members with no college education at all.²⁸

The degree validates exposure to a curriculum. It does not validate the ability to produce an outcome. A meta-analysis covering eighty-five years of data found that years of education correlate with job performance at 0.10. Cognitive ability correlates at 0.51. The credential is five times weaker than the underlying capacity it claims to represent.²⁹ The degree is peer-validated. Professors assess whether the student learned the material. The job is consequence-validated. The market assesses whether the person can do the work. The system uses the first signal to predict the second. The correlation is weaker than anyone paying for the degree assumes.

By 2023, IBM had dropped degree requirements from more than half its job postings, down from ninety-five percent. Delta Air Lines reported that eighty-two percent of its external hires required no degree. Sixteen states eliminated four-year degree requirements for most government positions. No measurable decline in performance followed any of these changes.³⁰

The test

Where did the system get the authentication right?

In 1959, commercial aviation recorded approximately forty fatal accidents per million departures. By 2024, seven fatal accidents occurred in 40.6 million flights. The accident rate had fallen by more than two orders of magnitude in sixty-five years.³¹

The structural reason is not better planes, though planes improved. It is the feedback mechanism. Every crash is investigated. Every finding produces a safety recommendation. Every recommendation is implemented. Pilots return to simulators every six months to rehearse scenarios that have killed people. The authentication runs through consequences. Did the pilot land the aircraft? Not: did other pilots find the technique interesting.³²

The airline industry's expertise is real because the system that validates it requires the expert to be right. The surgeon's expertise is real for the same reason. The bridge builder's expertise is real because gravity does not conduct peer review. The question is not whether expertise is valuable. The question is whether the system that authenticates it requires the expert to produce an outcome or merely to produce an argument.

Aviation does. Consulting does not. Academic social science, when finally subjected to an external test in 2015, saw sixty-four percent of its findings disappear. In cancer biology, the numbers were worse. Amgen attempted to replicate fifty-three preclinical studies. Six reproduced. Eighty-nine percent of the published findings, findings that had been peer-reviewed, cited, and built upon, did not survive contact with a second laboratory.³³

The authentication determines the reliability. Where the system tests knowledge against consequences, the knowledge improves. Where the system tests knowledge against peers, the knowledge might improve. The difference between "does" and "might" is where the cost lives.

Sowell's farm boy returns from the barn with milk or he does not. The criminologist returns from the field with a publication. The publication is evaluated by other criminologists. The crime rate is not consulted. Both are called experts. The word is doing different work in each case.

Harrison built a clock that was tested by the Atlantic Ocean. Maskelyne built tables that were tested by the Royal Society. The clock kept time within five seconds over eighty-one days. The tables were, in the judgment of other astronomers, elegant. The Board of Longitude authenticated knowledge by asking what other astronomers thought. The ocean authenticated knowledge by asking whether ships arrived.

The reader hires experts. Consults experts. Defers to experts. The question the reader might consider asking is not whether the person across the table is credentialed or experienced or confident. The question is simpler and older than any of those words. What tests this person's knowledge?

If the answer is consequences, the knowledge corrects itself. If the answer is peers, it might. Harrison knew the difference. It cost him more than forty years to prove it. The Board never changed its mind. The ocean did not require the Board's approval.