
November 19, 2026

Two people reach the end of a decade of work.
One has a theorem. She stands on four centuries of verified structure — Newton, Euler, Gauss, Hilbert — none of which she re-derived, all of which she received as free, load-bearing bricks, each one certified by a process she can trust without inspecting it. Her personal contribution is one new result at the top of an enormous inherited tower.
The other has a novel. She inherited technique — structure, scene, point of view, the accumulated craft of the form — but nobody handed her an answer to the only question that matters, which is whether the thing is any good. She had to build that answer inside herself, from scratch, over about ten years. And when she finally had it, she could not give it to anyone. It will die with her.
Now rank them. That is the question this article is about, and the reason it cannot be answered the way people try to answer it is that the two women are not doing the same kind of thing at all. One is climbing a ratchet. The other is working without one.
The hypothesis I set out to test was the folk one, stated sharply: that composers, novelists, choreographers and architects exercise more raw intelligence than physicists and engineers, because creative work forces the mind to hold open-ended ambiguity, cultivated taste and tacit judgment that formal science is explicitly built to offload onto codified method and peer-reviewed procedure. Put bluntly: the creators are the smart ones, and the scientists have infrastructure.
To test it I scored eighty real tasks across twenty-five dimensions of difficulty and read a library of 104 primary sources on task complexity, expertise, tacit knowledge, creativity, uncertainty and feedback. The scores came out close to tied, and I will say up front that I do not think the scores settle anything — they are one reasoner’s structured judgment, not measurement, and a result I happen to like is the kind I should trust least.
What the research did produce was a mechanism. One property of a field determines whether its knowledge compounds, whether its novices can be trained, whether its masters can teach, how long its apprenticeships run, how trustworthy its experts’ confidence is, and what machines will take from it first. That property is not how hard the field is. It is whether anyone can tell you that you are wrong.
Sixteen principles.
Difficulty is a relationship between a problem and the representations available for it, not a property of the problem.
Task complexity and practitioner intelligence are different quantities, and the folk debate conflates them.
One property decides a field’s entire character: whether it comes with a verifier.
A verifier turns patience into competence, because it makes search work.
A verifier is what makes knowledge cumulative — it is the mechanism of the ratchet.
A verifier is what makes novices trainable, which is why some fields have prodigies and others cannot.
Feedback you cannot attribute produces confidence rather than skill.
Without a verifier, volume stops being progress.
Science is leveraged and art is not — the asymmetry is inheritance, not difficulty.
Altitude is not climb, and comparing fields without subtracting inherited scaffolding measures the wrong quantity.
Verifiability is independent of stakes, so unverifiable does not mean unserious.
What produces a verifier is a fast, attributable causal link — not the severity of consequences.
Taste is a verifier you have to grow inside yourself, which makes it the most expensive knowledge there is.
Compression has no decompression algorithm, which is why masters so often cannot teach.
The real line runs through science rather than around it: paradigm-founding work is verifier-poor.
We built a meritocracy out of the legible half of human capability, and that half is now becoming abundant.
Nothing is hard in itself. Difficulty is a relation between three things — a problem, a mind, and the representations that mind has been handed — and changing any one of the three changes the difficulty without touching the problem at all. The word handed is doing most of the work in that sentence, and we will come back to it.
Notation is the clearest case. A notation looks like a way of writing a problem down, a neutral description, and it is nothing of the kind: it is part of the machinery that solves the problem. Positional notation with a zero performs most of the work the Roman system forces a skull to do unaided. Swap the notation and you have not described the task differently. You have redistributed it between the page and the head.
Multiply CXLVII by XXIII. Sit with it for a moment — not for the answer, for the feeling of a mind reaching for a handhold and finding none. There is no column to carry into and no place value to lean on, so every partial product has to be held whole and recombined from memory. This was a profession. It had manuals, and a merchant with a ledger to close hired a specialist to do it for him.
Now multiply 147 by 23. Identical quantities, identical operation. A nine-year-old does it in forty seconds with a pencil, and the pencil is not incidental — the paper holds the partial products so the child does not have to. Nothing about the world changed between those two paragraphs.
Physics tells the same story at the frontier. Computing how two electrons scatter in quantum electrodynamics once meant pages of algebra in which the physical content was invisible; a correct calculation taught you almost nothing about what was happening. Feynman’s diagrams turned the algebra into a picture. A line is a particle, a vertex is an interaction, and every drawing you can sketch corresponds to a term you have to compute. A graduate student now reads off in an afternoon what cost a senior theorist a week, and reads off the physics as well as the number.
The history of human capability is largely a history of such swaps. Arabic numerals, algebraic symbolism, the Cartesian plane, the calculus, double-entry bookkeeping, staff notation, the periodic table, Feynman’s diagrams. In each case a class of problems that had required hired specialists became routine, not because anyone got cleverer but because someone found a shape that did the thinking.
There is direct evidence for the mechanism. Chi, Feltovich and Glaser asked PhD physicists and undergraduates to sort twenty-four mechanics problems into groups. Novices sorted by what the problems looked like — inclined planes here, springs there, pulleys in the third pile. Experts sorted by the principle that cracks them: conservation of energy, Newton’s second law. Both groups were internally consistent. They were simply sorting different objects. The finding is not that experts know more; it is that the same printed problem is a different problem for the two of them, because the expert’s representation already contains most of the solution. Simon pushed this to its edge in 1973, arguing that ill-structuredness is not a property of a task but an artifact of the solver.
So “how hard is X” has no answer in the abstract, and every argument that opens by comparing the difficulty of two fields has skipped the only question that matters. Ask instead which representations this person was handed, by whom, and at what age. The difficulty you observe is the residue left after the inherited machinery has taken its share — and the shares are wildly unequal.
None of this makes difficulty arbitrary, and it does not make it subjective. Some problems have resisted every representation anyone has found: the behaviour of turbulent fluids, whether checking is fundamentally cheaper than solving, the path of a particular economy eighteen months out. Grünwald and Vitányi’s survey of algorithmic information theory states the hard version of this. By a simple counting argument almost all objects are incompressible — they admit no short description and no simplifying rule at all. Most of the space is genuinely structureless. A better notation is not always waiting to be found, and assuming one is is its own species of error.
The folk debate about whose work is harder fuses two claims that come apart cleanly: that a kind of work is more complex, and that the people doing it are more intelligent. Those are different quantities with different units. All four combinations exist, and the two mismatches are the interesting ones.
They come apart for a structural reason. An instrument that scores tasks scores tasks. It can tell you how much indeterminacy a piece of work contains, how many constraints collide inside it, whether anything can check the answer — and it knows nothing whatsoever about the distribution of minds that happen to walk into that work. Difficulty is a property of a problem plus a representation. Intelligence is a property of a person. Reading one off the other requires an assumption nobody states aloud: that fields recruit in proportion to their difficulty.
Consider a postdoc running a standard immunoassay inside a validated protocol. Plate layout fixed, dilutions fixed, controls specified, acceptance criteria written down by somebody else years ago. The work is demanding in the way careful work is demanding — fumble the pipetting and the plate is wasted — but it is close to zero on indeterminacy. There is a right answer and a written procedure for reaching it. A brilliant person doing that assay is doing low-indeterminacy work. The hard thinking happened once, upstream, and now lives in a binder.
Then a songwriter in a bedroom, self-taught, trying to get a chorus to land. No protocol, no acceptance criteria, nothing that can be consulted to establish whether the fourth version beats the second. The problem is genuinely open: the space of possible choruses is unbounded and the only available arbiter is an instrument the writer has to grow inside her own head. That is high-indeterminacy work, and the absence of a credential does not reduce it by a point.
Something else muddies every comparison of this kind, and it has nothing to do with the work. Fields with long, verifiable training ladders select hard on a particular measurable profile, because they can. A physics department admits on the strength of exactly the operations physics will later test. Songwriting has no entrance exam, no qualifying paper, no gate at all. The result is that physicists look more uniformly able than songwriters — and probably are — without that being a fact about the comparative difficulty of physics and songwriting. It is a fact about the door, and about who was allowed to stand in front of it at seventeen.
The cleanest evidence that the two dials turn independently comes from chess, where the task is fixed and the people vary enormously. Gobet and Campitelli surveyed 104 Argentinian players from amateur to grandmaster and found that reaching master strength took a mean of 11,053 hours of practice — with a range from 3,016 to 23,608, roughly eight to one. For solitary practice alone the spread was twenty-two to one. Several players logged more than 25,000 hours and never made master. Hold the task perfectly constant and the human cost of competence still varies by an order of magnitude. Whatever that variation is, the difficulty of chess cannot be it, because chess did not change.
What follows is a discipline rather than a belief. Stop reading a field’s prestige as a verdict on the people inside it, and stop reading an individual’s brilliance as a verdict on their field. When hiring, score the task and the candidate on separate sheets — because the commonest error in senior hiring is to credit a person with the structure they inherited. The analyst who looked exceptional inside a firm with an exceptional model, and who is ordinary without it, is a type, and everyone has met one.
None of this licenses the comfortable conclusion that intelligence is unreal, unmeasurable, or merely a social artifact. Gobet and Campitelli’s own data cuts the other way too: practice was necessary, and twenty-five thousand hours without mastery says something about the player and not only about the hours. The honest claim is narrower and harder to wriggle out of. Task complexity is a real quantity, individual capability is a real quantity, and an instrument built to measure the first is simply not evidence about the second.
Of all the ways one piece of work can differ from another, a single property matters more than the rest combined. Does the task come with a verifier? A verifier is any procedure that can look at a candidate answer and tell you whether it is good — without you already knowing the right answer. That last clause is the whole definition. Drop it and everything becomes verifiable, trivially and uselessly.
The idea is borrowed, and it is worth knowing from where. In 1971 Cook proved that every problem whose candidate solutions can be checked quickly reduces to Boolean satisfiability, founding the complexity class NP and making a mathematical object out of an everyday asymmetry: some problems are agonising to solve and cheap to check. A sudoku takes an hour to fill and twenty seconds to score. The P-versus-NP question asks whether checking is fundamentally cheaper than doing. Everyone believes it is. Nobody can prove it.
Carried out of computing and into human work, the property predicts far more than it was built for. The five principles that follow are its consequences — how a field trains people, whether its knowledge accumulates, what volume of effort buys, and which of its tasks machines take first.
Start with the contrast at its sharpest. A compiler is a verifier and it answers in seconds. It does not care who wrote the code, does not get tired, and cannot be argued with. A test suite is a verifier that answers in minutes. Now look for the equivalent on the other side: nothing in the world takes a novel as input and returns a verdict. An editor is not a verifier. Editors disagree, and the same manuscript’s fate turns on which desk it lands on, which is a fact about desks rather than about the book.
The same gap opens in work with enormous stakes. A transport planner deciding whether to run a tram line down a particular street is choosing between options no available procedure can rank. Rittel and Webber named this class in 1973 and listed its properties: there is no definitive formulation of the problem, because formulating it is the problem; there is no stopping rule; solutions are good-or-bad rather than true-or-false; and there is no immediate and no ultimate test of a solution. The tram helped or it did not, twenty years later, tangled with everything else that happened to that street, with no counterfactual city to compare against.
What that absence does to a practitioner shows up in the protocol studies. Goel and Pirolli watched expert architects, mechanical engineers and instructional designers work, and extracted twelve invariants common to all three. Two are the ones to keep: no right or wrong answers, only better and worse — and personalised stopping rules and evaluation functions. Read the second one slowly. With no correctness criterion and no feedback, the designer has to supply her own termination condition. That is why two excellent architects stop at different places on the same brief and both are finished, and why “it’s done” is a sentence only one person in the room can say.
The consequence is a habit worth installing before any substantial piece of work. Name the verifier out loud, specifically, before planning anything. Not “the market will tell us” — when, how precisely, and will you be able to tell which decision it was answering? If you cannot name it, you are in the other regime, and every instinct imported from the verified half of your career is about to be wrong.
The limit here is large. Almost no real work is purely one thing or the other. Most tasks are a stack with a different verification regime at every layer: does the code compile, do the tests pass, is this architecture any good, is this the right product to be building. The bottom two layers answer in minutes and the top one may never answer. Teams routinely run a flawless loop at the bottom of the stack and mistake it for progress at the top — green pipelines, an elegant refactor, a beautifully built thing nobody wanted. Confusing which layer you are on is a distinct failure with its own signature, and it is commoner than having no verifier at all.
One more caution against over-reading the formalism. Fortnow’s survey of P versus NP stresses that NP-hardness is a worst-case statement, and that modern solvers routinely crack industrial instances with hundreds of thousands of variables. Having a verifier does not make a task easy, and being provably hard in the limit does not make the instance on your desk intractable.
Here is what a verifier buys. If you can rank candidates, you can generate a great many of them badly and let the ranking do the work. Search plus a verifier is a machine that manufactures competence out of patience, and it requires no genius at any step — which is precisely why its presence or absence matters more than the talent of the people involved.
The mechanism is that the intelligence relocates. It stops living in the generator and takes up residence in the filter. Evolution has no mind at all and produced the eye. Drug screening, engineering optimisation, chess engines, finding a decent restaurant in an unfamiliar city: in every case the generating step is dumb, repetitive and cheap, and all the discrimination happens in the test.
A materials team hunting a compound that stays conductive at a given temperature does not need anybody who can predict the winner. They need a synthesis route cheap enough to run thousands of times and a measurement honest enough to trust. The chemistry intuition in the room gets spent pruning the space, not picking the answer, and the furnace settles it. If the measurement is good, the team’s competence is largely a function of how many plates they can run before the funding ends.
The same logic at the opposite scale: an engineer at two in the morning with one failing integration test. Deliberation is the wrong tool here and it feels like the right one. Each hypothesis costs thirty seconds to check — comment out the middleware, pin the dependency, bisect the commit range, flip the feature flag — so generate-and-test beats thinking by a wide margin, every time. The engineers who are fast at this are not the ones with the best theory of the bug. They are the ones who stopped theorising soonest.
Evidence that patience genuinely substitutes for insight comes from where you would least expect it. Satisfiability is NP-complete; the theory says it is hard. Fortnow’s survey notes that modern SAT solvers nevertheless dispatch industrial instances with hundreds of thousands of variables as a matter of routine. The wall is real and practical problems went over it anyway, because a cheap checker plus relentless structured search plus compute is an engine that does not get tired. Read that as a claim about strategy rather than about logic: given a verifier, grinding wins against problems insight cannot touch.
What to do with it is unambiguous, and most organisations do the opposite. Where a verifier exists, volume and delegation are correct and agonising is merely slow. Let juniors and machines generate; let the verifier discriminate, because it is better and faster at discriminating than your judgment is. Stop reviewing line by line what the test suite already answers. Ship the experiment instead of debating it, when the experiment costs a day and the debate costs a week.
Hold that beside the inverse case, which arrives in the eighth principle. Where no verifier exists, every sentence in the paragraph above becomes dangerous advice, because volume without a ranking function is motion rather than progress, and delegation without a check cannot be audited at all.
The limit is the one that destroys companies. Search needs the verifier to be aligned with what you actually want. Optimise a proxy hard enough and you get a technically excellent product nobody wanted — the team that maximised time-on-page and built an application unpleasant to leave, hitting its target quarter after quarter. A cheap wrong verifier is more dangerous than no verifier at all, because it industrialises the error, reaches it faster, and wraps it in a dashboard that makes doubt look like innumeracy.
A verifier also makes search possible without making the space navigable. Franco, Doroc, Yadav, Bossaerts and Murawski showed that complexity measures taken from theory — chiefly how close an instance sits to the satisfiability phase transition — predict both human accuracy and time spent across three canonical hard problems, and do so after controlling for instance size. Difficulty tracks the structure of the particular instance rather than how large it looks. Some searches are cheap to check and still hopeless to run, which is why “generate more candidates” is an instruction and not yet a method.
If I can verify your result, I can build on it without redoing it. That single move is the most consequential thing our species has ever worked out. I take your theorem as a brick and put my floor on top of it, and the brick does not have to be fired again.
A ratchet is a mechanism that permits motion in one direction and forbids slippage back. Verification is the pawl. Without it a field can still move, but every generation re-establishes the ground it stands on, and the ground is never higher than one person can reach in one lifetime.
Mathematics is a four-thousand-year cathedral built from nothing but that move. A graduate student this morning used the spectral theorem, the fundamental theorem of algebra, Lebesgue’s construction of the integral and a dozen results from the 1970s whose proofs she has never read and never will. This is not laziness and nobody pretends it is. Verification happened once, the result was cemented into the structure, and the entire point of cementing it was that nobody would pay that cost again.
The price of the first turn of the ratchet is occasionally visible. Klein and colleagues’ formal verification of the seL4 microkernel — 8,700 lines of C and 600 of assembler, proved correct against an abstract specification — consumed about twenty person-years of proof work against 2.2 to write the kernel. An order of magnitude, paid once. Every system since that sits on seL4 inherits the proof for free and does not know it is inheriting anything. That is what a brick costs and what a brick is worth.
It is not only a mathematical habit, which is why the law generalises rather than describing one discipline’s good luck. A structural engineer sizing a steel beam this afternoon opens a code table and reads off a load factor. That number is a compressed century of tested failures — collapses investigated once, by people long dead, whose conclusions were checked, argued over and then cemented into a figure she multiplies by. She does not revisit the collapses. She is not expected to. The verification was the work, and the table is where the work went to live.
Now the other side, and it is not a complaint about art. A novelist inherits technique, and does so freely: free indirect style, the way a scene turns into its own consequence, how to put a gun on a wall in act one. Craft is real and it transmits through books, workshops and apprenticeship. What does not transmit is the evaluation function. Nobody can hand a writer a working answer to “is this good,” because that answer was never externalised into anything checkable in the first place. It has to be grown privately, over roughly a decade, and when it finally works it cannot be given away.
So two fields that look comparable in difficulty diverge wildly in what their practitioners inherit. A twenty-five-year-old physicist operates at an altitude four centuries of verified structure carried her to, and her own contribution can be modest while the view is enormous. A twenty-five-year-old novelist starts at sea level, as every novelist has, because the shoulders are not available to her — not because sand is harder terrain than rock.
The practical consequence runs in two directions. When judging a person or a field, separate inherited altitude from personal climb, since the folk ladder quotes the first while claiming to report the second. And when you want your own work to compound rather than evaporate, externalise the evaluation criteria: write down what would count as good before you look at the result, so that next year’s version of you inherits a brick instead of a mood.
The limit comes in two parts and both matter. Cumulation is not the same as progress toward truth — a verifier cements a shared error as efficiently as a correct result, and a paradigm shift is precisely the discovery that a long-ratcheted structure was built on a bad foundation, with the ratchet having made the error load-bearing in the meantime. And art does accumulate. Technique accumulates, vocabulary accumulates, the available moves multiply with every generation. What art does not accumulate is settled verdicts, which is a far narrower claim than the one usually made on its behalf.
With a verifier you can hand a beginner a problem, let them fail, and tell them immediately and unambiguously that they failed. That loop is the entire engine of skill acquisition. Where it is missing, training does not merely slow down — it becomes a different and much weaker activity, and no quantity of institutional seriousness compensates.
Deliberate practice, as Ericsson defined it, presupposes a verifier and cannot be built without one. The requirement is effortful work at the edge of current ability, with immediate informative feedback and repetition. Remove the feedback and the phrase stops meaning anything. You cannot stay at the edge of your ability if nothing tells you where the edge is; you drift to wherever it feels productive, which is reliably somewhere comfortable.
This is why chess has thirteen-year-old prodigies. A junior who loses a game runs it through an engine the same evening and receives a per-move verdict, with the exact move where the position turned marked for her. The engine does not care that she is thirteen, does not soften the assessment, and will do it again tomorrow night. Ten thousand of those verdicts, compressed into pattern recognition, is a master.
Now the same age in a different world. There are no teenage general counsels, and the reason is not that law is harder than chess. A twenty-six-year-old who has drafted a dozen settlement agreements has no way to learn which clauses saved the client, because the disputes that never happened leave no trace in the file. The feedback she gets is the deal closing: one terminal bit, arriving months late, confounded with everything else in the negotiation. There is nothing for an engine to score.
Notice what her firm does instead, because every low-feedback profession does the same thing. It substitutes time. Seven years to partner, a fixed sequence of rotations, promotion gates spaced by the calendar rather than by any demonstration. That is not stupidity. It is what an institution builds when it cannot test, and it has the texture of training without the loop that makes training work.
The gradient this produces is measurable and stark. Macnamara, Hambrick and Oswald’s meta-analysis of 88 studies found deliberate practice explained 26% of performance variance in games, 21% in music, 18% in sports, 4% in education — and under 1% in the professions. The figure fell as domains became less predictable and less structured. Interpret the bottom of that list properly: in professional work, more than 99% of the difference between practitioners is something other than accumulated practice. Set that beside how we actually talk about seniority, where years serve as a near-sufficient proxy for capability, and the result is devastating.
The gradient is not a ranking of those fields by difficulty. Games sit at the top because the game tells you, instantly and without negotiation. The professions sit at the bottom because the world answers late, vaguely, and about something other than what was asked.
What to do about it is to stop paying for years and start manufacturing the missing loop. Inside a four-year geopolitical forecasting tournament, Chang, Chen, Mellers and Tetlock found that a cognitive-debiasing tutorial lasting under an hour improved accuracy by 6 to 11% in Brier score over control, with the benefit persisting across years. An hour of structure beat decades of unstructured experience, in a domain famous for teaching nobody anything. Forecasts with resolution dates, a decision journal that records the reasoning before the outcome, a pre-mortem: these are manufactured verifiers, and they are cheap.
The gradient is about the availability of the learning loop, not about whether skill exists in those fields. Professional skill is real — a good general counsel is worth a great deal and a bad one is a catastrophe — it is simply not acquired the way chess skill is acquired, and it therefore does not arrive at thirteen. And the underlying literature is arguing with itself: Macnamara and Maitra’s double-blind replication of the original violinist study failed to reproduce its core claim that practice differentiates every level of skill. Hours matter less than the canonical story says, in both directions. Resist any reading of this principle that makes practice decisive, and any that makes it irrelevant.
Three things get lumped together under the word feedback and they need separating. Latency: how long until the world answers. Granularity: whether the answer is a rich signal or one terminal bit. And credit assignment: whether you can tell which of your actions caused the outcome. The third is the one nobody tracks, and it is the one that decides whether experience teaches anything at all.
Credit assignment fails in specific, identifiable ways. Pignatelli and colleagues’ survey of temporal credit assignment sets out the structure: long delays between action and consequence, stochastic outcomes, sparse reward, and transposition — many different action sequences arriving at the same state — all dilute the learning signal per decision, independently of how large the reward is. Their summary of real-world decision problems is that feedback is typically noisy, delayed, and carries little or no information about causes. That is not a description of a bad environment. It is a description of most environments.
Day-trading is the clean case, because it fails on exactly one of the three dimensions. The feedback is instant and it is granular — a number, every second, denominated in money. Attribution is near zero. A trader up on the quarter cannot separate her thesis from the sector, the sector from the index, or the index from the month she happened to start in. She can run that loop for a decade, feel herself learning throughout, and finish with a strong, specific, articulate account of a market that does not work the way she says. Fast feedback you cannot attribute is worse than none, because it feels like learning and therefore closes the question.
A manager’s weekly one-to-one is the same failure in a quieter register. Twenty years of them, several hundred people, a career of teams that mostly did well. Which conversation mattered? Which intervention moved anything, as against the ones that merely preceded things going fine? The counterfactual never runs, outcomes arrive years later carried by other causes, and the people who thrived would mostly have thrived. Confidence here is not dishonest. It is uninformative.
Hogarth, Lejarraga and Soyer supply the general account. Inference always spans two settings — one where information is acquired and one where it is applied — and accuracy requires a kind match between them. Wicked environments are those with systematic mismatches: censored or selected samples, missing or biased feedback, shifting targets. The part to hold onto is that wickedness is a property of the environment rather than of the person, which means no amount of individual rigour fixes it from the inside.
Kahneman and Klein’s adversarial collaboration settled the conditions. Valid intuitive expertise requires an environment of sufficiently stable, learnable cue-outcome regularities, and prolonged practice with rapid, unambiguous feedback. Both, not either. Where one fails they state flatly that subjective confidence and years of practice are not reliable indicators of judgment accuracy, and among the failures they name stock picking, long-range political forecasting, clinical prediction and personnel selection. Those are not marginal activities. They are a large fraction of what senior professional life consists of.
Heuer’s account of a horse-race handicapping study shows what the inside of that failure feels like. As the number of cues available to the experts rose from five to forty, accuracy stayed flat while expressed confidence roughly doubled. More information bought certainty, not performance. Anyone who has watched a well-briefed committee grow decisive in proportion to the size of its data pack has seen it happen live.
The consequence is a pricing correction. In low-attribution domains, “twenty years of experience” is not evidence and should not be paid as though it were. Ask instead for the reasoning recorded before the outcome — the memo written in advance, the forecast with a resolution date, the pre-mortem on file. Someone who cannot produce those has twenty years of exposure, which is a different asset entirely.
The limit needs stating narrowly or it curdles into contempt. This does not mean experienced people in low-attribution fields know nothing. It means their confidence carries no information about their accuracy, which is both a smaller claim and a more useful one, because it tells you what to discount rather than whom to dismiss. And the rule people tend to extract — faster feedback is better — is false as stated. Gabay and colleagues found adults with ADHD were impaired at probabilistic learning under immediate feedback but indistinguishable from controls when the identical feedback was delayed by a few seconds, because delay shifts the work from procedural to declarative systems. Latency is not a quality dial. It selects which machinery does the learning.
You cannot brute-force a novel. A thousand plot outlines gets you no closer to a good book, and the reason is not that a thousand is too few. It is that you cannot rank the thousand. Quantity converts into quality only through a ranking function, and in this kind of work there isn’t one.
Volume feels like progress because across most of our experience it is. Effort, candidates and iterations are the things we can watch ourselves doing, so we use them as proxies for movement. The proxy holds only where something downstream is discarding the bad candidates. Remove the filter and the pile grows without the work advancing a single step — motion rather than progress, and indistinguishable from progress from the inside.
A novelist three hundred pages into a book that is not working knows this precisely. The obvious move is to write the fourth hundred, because writing is what she does and pages are what she can count. The fourth hundred may be the mistake. No procedure can tell her whether the problem is on page 40 or in the premise, and more pages generate more structure to be wrong about. The engineer’s grind reliably converts effort into progress; the writer’s grind sometimes converts a year into nothing, and more effort does not repair it, because effort was never the binding constraint.
The same distinction inside a company is sharper, because both halves happen in the same building, often in the same week. Running twelve A/B tests improves a conversion rate, reliably, almost mechanically — an arbiter exists and it discards ten of the twelve for you without consulting anyone’s seniority. Producing twelve strategy decks does not improve a strategy decision. The twelfth deck is more fluent than the first, better argued, more confident, and no likelier to be right. Its fluency is actively harmful, because fluency is experienced as conviction, and conviction is what gets funded.
This is why cheap generation has not helped the work that most needed help. A model will produce forty taglines in nine seconds, or sixty product strategies, or a hundred names. In every one of those tasks the candidates were never the bottleneck; a competent team could already generate more options than it could evaluate before lunch. Multiplying the supply of candidates in a domain whose constraint is evaluation tightens the constraint rather than loosening it. The queue gets longer and the arbiter is still one person with an opinion.
The structure behind all of this is documented where you would expect it: in studies of designers. Goel and Pirolli’s protocol work found, across architects, engineers and instructional designers alike, no right or wrong answers, no immediate feedback, and personalised stopping rules and evaluation functions. Rittel and Webber had already put it as a bare property of wicked problems — there is no stopping rule. A search with no termination condition does not converge. It simply continues, and then it continues some more, and the practitioner’s own instrument is the only thing that can end it.
So the operating rule in verifier-poor work inverts the one from the fourth principle. Cut the candidate count deliberately. Spend the recovered time on the instrument and on who is in the room, because where nothing can adjudicate, the most assured voice in a meeting carries no information at all. Stop measuring output, and stop rewarding it, since output is the one thing in this regime that is not evidence.
The limit is easy to abuse in exactly the opposite direction. Volume is not useless in verifier-poor fields; its function changes completely. Writing a thousand pages is how a writer builds the internal instrument that will one day tell her the chapter is dead — that is formation, not search. Practice volume, yes. Search volume, no. The two look identical from outside and have nothing in common: one is building the arbiter, the other is pretending one already exists. And the distinction gets used as a licence by people who want permission to produce nothing: a writer who has written four hundred pages this year and a writer who has written four have not had the same year, whatever either of them says about taste.
Two practitioners can work equally hard, equally well, for equally long, and one of them ends up standing four centuries higher than the other. The difference is not talent and it is not difficulty. It is how much verified structure the field was able to hand over for free.
The mechanism is verification done once and then cemented. When a result can be checked, it has to be checked a single time; afterwards everybody treats it as a brick rather than a question, and the next person builds a floor on top of it without reopening it. Fields that can do this accumulate. Fields that cannot are obliged to re-grow their foundations inside every practitioner, privately, from the beginning — because the thing that would have to transmit is not a fact but a working judgment about what counts as good. Facts travel. Judgments do not.
Watch a second-year doctoral student start a lattice simulation of quark confinement. In one afternoon she uses calculus, linear algebra, the gauge structure of the Standard Model, a Monte Carlo sampler and a floating-point library, and she has verified none of them and intends to verify none of them. She did not re-derive Maxwell before lunch. Her personal contribution that year may amount to one tightened estimate of one parameter. That modest contribution sits on top of a tower so high that the view from it is genuinely new. Her climb was a few metres; her altitude is enormous. Both things are true and they are not in tension.
Now a novelist three hundred pages into a second book that is not working. She has inherited a great deal of craft — free indirect style, the chapter break, how to move a character across a room and make the reader feel the room, everything the form learned between Defoe and last Tuesday. What nobody handed her is the one instrument that matters at page three hundred: a reliable answer to whether this chapter is alive or merely competent. There is no reference she can consult. Her editor has an opinion, her agent has a different one, and neither opinion can be checked against anything outside itself.
She built that instrument over roughly a decade, by making several hundred thousand words that were not good enough and slowly noticing which ones were. When she dies it goes with her. The next novelist starts at zero on exactly that, and will spend her own decade on it, and the two decades do not add up to anything. There is no ratchet in the room.
The pattern surfaces as soon as you stop counting how many people are involved. Among the eighteen tasks in our catalogue that are genuinely solo — one person, no institution, nobody to coordinate — the five most demanding alternate between the two worlds: theoretical physics, a graphic novel, a body of abstract painting, a debut novel, the invention of a new statistical method. That alternation is the whole finding. Strip out the institutional weight that lifts science’s apparent position and what remains is not a ladder with laboratories at the top. It is two kinds of unassisted work that turn out to be about equally unassisted.
So drop the question of which field is harder; it has no answer and never had one. Ask instead what fraction of someone’s position was inherited and what fraction was climbed, and then judge the climb. A hiring panel comparing a physics PhD against ten years of unfunded fiction is comparing a received altitude with a personal one, and it will reliably mistake the first for the second — not because the panel is lazy but because altitude is the only one of the two that comes with documentation.
The honest limit is that art is not scaffolding-free, and the strong version of this claim is false. Genre transmits. Form transmits. Technique, canon, conservatoire training, the entire apparatus of workshops and editors and masterclasses — all of it transmits, and a novelist today is not starting where Defoe started. Narrative tools exist now that took two centuries to invent, and she has them for nothing, exactly as the physicist has calculus for nothing. The asymmetry is narrower than it first looks and it sits in one specific place: the evaluation function. Craft is inheritable; taste is not. The scientist stands on shoulders and the novelist stands on sand, and the point has never been that sand is harder terrain. It is that the shoulders are not available to her.
When someone tells you theoretical physics is obviously harder than songwriting, listen to what the sentence actually compares. It puts four hundred years of accumulated, verified, inherited structure on one side and one songwriter’s decade on the other. That is nonsense in the specific way that comparing a mountaineer at eight thousand metres to a swimmer at sea level is nonsense.
The error is an accounting error, and it survives because the two quantities use the same unit. Height above sea level and metres climbed this morning are both measured in metres, so a number quoted without saying which one it is sounds like either. The folk ladder quotes altitude and claims to be reporting climb. Nobody notices the switch, because the inherited part of a practitioner’s position is invisible from inside — the physicist does not experience Maxwell as assistance, any more than you experience the floor as assistance while walking.
There is a clean operational test, and it takes about a minute. Ask what each practitioner would have to rebuild from nothing if their field’s accumulated record vanished overnight. Not what they know. What they would have to re-derive before Monday.
Run it on a competent working chemist — the kind who develops a scalable synthesis route for a mid-size pharmaceutical intermediate. Burn the library and he is finished. The periodic table’s organising logic goes, and with it every reaction mechanism, every spectroscopic reference table, two centuries of accumulated synthesis routes, the solubility data, the catalyst literature. He cannot reproduce the route he ran last week, because that route was a search through a space that other people had already mapped and labelled. He is reduced to a careful person with glassware and no map. His capability was real and almost none of it was inside him.
Run it on a competent working poet. Burn the library and she loses something that genuinely hurts: the canon, which is where her ear was trained, and the thousand poems against which her sense of a dead line was calibrated. But the instrument itself is in her head, and it still works. She can write tomorrow morning and the work will be as good as it was. Say the uncomfortable half out loud: this is not a demonstration that the poet is more capable than the chemist. It is a demonstration that her apparatus is internal, and therefore uninheritable, and therefore also unremovable.
Our own scoring tripped over the same accounting error from the other direction, which is the best evidence we have that it is not a hypothetical. Engineering and institutional design led the table, and a good share of that lead came from coordination load, adversarial pressure and regulatory constraint. Those are real difficulties. They are also difficulties that a field’s accumulated apparatus generates and then manages — the regulator exists because the field matured — rather than difficulties one practitioner climbs alone. The table was quietly reporting altitude plus institutional mass, labelled as difficulty. Once that was discounted, the gap between the sciences and the arts stopped being a gap.
Two things follow for anyone who evaluates people. Discount the inherited component before you admire the output, and discount hardest where the documentation is thickest, because thick documentation is a signature of inheritance rather than of climb. And apply it to yourself in the other direction: if your field hands you a great deal, you are higher than you climbed, and the part of your career that is actually yours is smaller and more interesting than your CV implies.
The limit matters more here than anywhere else, because this principle is the one most often weaponised. Subtracting inheritance does not make inheritance cheating, and it does not make the chemist a lesser figure than the poet. The entire point of civilisation is that you should not have to re-derive Maxwell. A species that made every generation rebuild the periodic table would still be dying of infected teeth. Using the ratchet is not coasting, and building a result that others can stand on is a contribution of a kind the poet cannot make at all. The error is in the accounting, never in the practice — and the correct conclusion is not that physics is overrated but that the folk ladder is measuring a quantity nobody intended to measure.
Here is the strongest objection to everything in this article, and it deserves to be put properly rather than knocked down. Creative fields lack verifiers because their outputs do not matter enough to be checked. Where being wrong has consequences — a bridge, a drug, a proof — reality supplies a verifier whether anyone wants one or not. Where being wrong costs nothing, none develops. So low verifiability is a symptom of low stakes rather than high difficulty, and the novelist is only working without a net because nothing falls.
Concede the part that lands, and concede it without flinching. Some verifier-poor fields are verifier-poor precisely because they are unserious. You can sit in a seminar where the same four positions have circulated for thirty years and not one of them has ever been obliged to retire, and the reason they never resolve is not that the questions are deep. It is that nothing whatever rides on resolving them, so no mechanism ever arrives to force the issue. Mistaking that condition for profundity is a real failure mode, it is common, and the people who commit it are usually the loudest about how their field cannot be measured.
Then look at what the objection has to survive. Drafting a national constitution. Negotiating a peace treaty between parties who were shelling each other last year. Sentencing a defendant in a morally hard case. Deciding which parent a seven-year-old lives with. Setting interest rates for a currency union of three hundred million people. In every one of those, being wrong is catastrophic and effectively irreversible. In every one, nothing can tell you at the time whether you were right.
Sit with the judge for a moment. A nineteen-year-old drove the car in an armed robbery; the shopkeeper died; the driver never went inside and did not know a gun was in the bag. Twelve years or four. The judge has sentencing guidelines, which bracket the answer and do not choose inside the bracket, and she has to choose by Thursday. Forty years from now there will be a man with a life, and that life will not be an experiment with a control arm. The version of him who got the other sentence does not exist and cannot be consulted. She will never find out. Nobody will.
Now the same structure in a different world entirely. A rate-setting committee is looking at an inflation print that is high and an employment recovery that is incomplete. Raise, and they may cause a recession that did not have to happen. Wait, and they may let expectations embed and spend five years grinding them back out. Whatever they do, the counterfactual economy is unavailable forever, which is why the argument about 2021 will still be running in 2041 with both sides citing the same data. Maximal consequence, zero verification.
This is the standing description of such problems rather than a new observation. Rittel and Webber established it in 1973 for the class of problems they called wicked: their solutions are not true-or-false but good-or-bad, and there is no immediate and no ultimate test. Read what that sentence is actually conceding. Not that quality is imaginary — these authors were studying city planning, where bad work ruins lives — but that no procedure exists to adjudicate it, at the time or ever. Unverifiable is a fact about the available checking apparatus. It is not a fact about how much the answer matters.
So the two properties are orthogonal, and treating them as one is how respectable people end up dismissing the hardest work there is. Verifier-poverty comes in two kinds: one caused by triviality, where nothing rides on the answer and no mechanism develops because none is needed, and one caused by genuine causal opacity, where everything rides on it and the world still will not say. Telling those two apart is most of the skill of judging which fields deserve respect, and almost nobody does it, because the surface signature is identical — the endless unresolved argument.
The limit is that this cuts both ways and does not hand creative fields a clean bill of health. Showing that stakes do not create verifiers does not show that any particular unverifiable field is serious; it only removes the shortcut by which you could have dismissed it without looking. The work of looking remains. A poem and a monetary policy decision are both unverifiable, and only one of them is unverifiable because nothing rides on it — but which is which has to be established case by case, and the practitioner’s own confidence is no help at all, since confidence grows just as well in the trivial soil.
If stakes do not produce verifiers, something else does. Three properties, and all three have to be present: the link between action and outcome must be stable, it must resolve fast enough to learn from, and it must be attributable to the specific choice you made. Drop any one and the feedback stops teaching.
Attribution is the one people forget, and it is the one that does the damage. An outcome that arrives entangled with four hundred other causes is not a signal about your decision; it is a signal about the world. Day-trading is the pure case: feedback inside the hour, attribution near zero, which is why it reliably manufactures practitioners who are confident and unskilled. Fast feedback you cannot attribute is worse than no feedback at all, because it feels exactly like learning.
Bridges have all three. A structural engineer finishes a girder design and the load test answers unambiguously, in a day, against that design. Deflection under known load is not a matter of interpretation; the number either falls inside tolerance or it does not, and no component of the result is traceable to the political mood of the city council. That is why a young engineer can become genuinely good in eight years. The world is willing to mark her homework, promptly, and to mark the right question.
Constitutions have none of the three and never will. The outcome arrives in fifty years, by which time it is tangled with a commodity boom, two wars, a demographic transition and a technology nobody drafting the document could name. There is no counterfactual country. The drafter who wrote the emergency-powers clause in 1991 cannot learn whether that clause was the thing that held in 2034 or the thing that nearly broke, because the same document also contained ninety other clauses and the country also contained sixty million people with their own ideas.
Then put a third case between them, because the middle is where the mechanism becomes visible. A colorectal surgeon adopts a modified anastomosis technique. Leak rates show up in his own series within weeks, and because the rest of the protocol held steady, the difference is attributable to the technique. He runs sixty cases and he knows something. This is why surgical practice measurably improves decade on decade while constitutional design does not — not because surgeons are more rigorous than constitutional lawyers, but because the patient answers, quickly, and about the right thing.
Which leads to the move that matters in practice: verifiers can sometimes be manufactured by engineering those three properties in deliberately. A prototype attacks latency, pulling a two-year answer into a fortnight. A red team supplies the adversarial check that a friendly environment withholds. A decision journal attacks attribution directly — write down the reasoning and the expected outcome before you act, and a year later you can audit the reasoning rather than only the result, which is the difference between learning and storytelling.
Tetlock’s calibration work is the anchor here, and the number is less interesting than its shape. A training tutorial of under an hour improved geopolitical forecasting accuracy by six to eleven percent, and the gain persisted across subsequent years. Sit with that. Geopolitical forecasting is a domain with famously atrocious feedback, where decades of unstructured experience produce pundits rather than forecasters. One hour of deliberate structure beat the decades. That is not a claim about how clever the tutorial was; it is a claim that in verifier-poor domains, the binding constraint is the absence of a learning loop rather than any shortage of effort or intelligence.
So if you are in a field where nothing answers, stop waiting for experience to make you good, because it will not. Build the loop by hand, accept that it is cruder than a load test, and use it anyway.
The honest limit is that a manufactured verifier is a proxy, and every proxy inherits the standard pathology: the moment it becomes the target, people optimise the proxy rather than the thing. A team that builds the prototype that will pass the prototype test has manufactured a verifier that verifies nothing. A checkout flow that wins a two-week conversion test by nagging customers can also train them, over eighteen months, to distrust you. The three properties tell you whether a signal can teach. They say nothing about whether it is measuring what you care about, and that second question is yours alone.
Taste is usually filed under subjective preference — a matter of opinion, not a thing serious people theorise about. That filing is wrong, and the wrongness is most of why creative work is so hard to think about clearly. Taste is a verifier: a working evaluation procedure that reliably separates better from worse where no external procedure exists.
It does the same job a proof-checker does in mathematics, or a wind tunnel does in aerodynamics. Something takes a candidate as input and returns a verdict you can act on. The only difference is location. The proof-checker sits out in the world where anyone can use it; taste sits inside one person’s head, where nobody else can reach it, including the person, who typically cannot explain its output.
That single relocation explains a set of otherwise baffling facts, and the clearest is the length of creative apprenticeships. If there were genuinely no right answer, training would be quick and pointless — you cannot train someone toward nothing. Yet the domains with no external verifier carry among the longest formation periods in human activity, with a decade being unremarkable. That is what growing an instrument costs. The only known calibration method is to make thousands of things, most of them not good enough, and slowly notice which ones were alive.
Watch it operate. A documentary editor has two versions of the same six-minute sequence on her timeline at two in the morning. Both are defensible; both would pass; a producer would sign off on either. One of them holds, and the other merely works, and she can tell. What she cannot do is say why in terms that would survive being written down, and she certainly cannot hand the discrimination to the assistant editor, who has been with her four years and is still wrong about this sort of thing roughly a third of the time.
The same structure appears in places nobody files under art. A mechanical designer four months into a surgical instrument has a housing that satisfies every requirement on the list — tolerance, sterilisation cycles, cost, grip envelope — and knows it is not right. He cannot name a violated constraint, because none is violated. He adds two weeks and changes the parting line, and the senior partner picks the new version out of a row of six without being told which is which. No procedure in that building can adjudicate between the two housings, and yet the two of them agree, and they agree because they have been calibrated on the same decade of objects.
The neuroscience says why no shared external scale is available to appeal to. Vessel, Starr and Rubin found that sensory regions scaled linearly with how much observers liked a work, while frontal default-mode regions jumped in a step only for the most moving pieces — and those step responses were highly idiosyncratic across observers. Read the implication rather than the method. The low-level part of aesthetic response is stimulus-driven and roughly common to everybody, which is why there are rules about composition that can actually be taught. The part that decides whether something is a revelation is not stimulus-driven, and it does not line up between people. There is no public scale there, because the mechanism generating the verdict runs on an individual’s own history.
Goel and Pirolli found the operational consequence decades earlier, in design protocols from architects and engineers: practitioners carry personalised stopping rules and personalised evaluation functions. Of course they do. Nothing external supplies either, so each person has to build a private one, and the built ones differ.
Treat taste, then, as a capital asset with a ten-year build time, and invest in it the way you would in any instrument — through volume of attempts and brutal exposure to work better than yours. Stop describing it as soft. It is the only verifier available in half of serious human work, and the half that includes which problem to attack.
The limit is the one that keeps this from being a licence. A private verifier can be badly miscalibrated with no mechanism to discover it, which is precisely the condition of the confident bad artist — a real instrument, operating, and wrong, for thirty years. Nothing internal can catch that, because the thing doing the checking is the thing that needs checking. Which is why taste is grown by prolonged collision with a canon and with people whose judgment outruns yours, and never by introspection, and why the practitioner who has withdrawn from all comparison is not protecting a vision. They are protecting an uncalibrated instrument.
Not will not — cannot. Polanyi’s line about knowing more than we can tell gets quoted as gentle mysticism, a graceful shrug at the ineffable, and it is nothing of the kind. It is a mechanical claim about what expertise physically is, and it has a consequence that most mentoring arrangements are built to ignore.
Expertise is compression. Ten thousand cases get squeezed into a representation that fires in milliseconds, and the squeezing discards the cases. What survives is a disposition to respond, not a library of examples with reasons attached. Compression of that kind has no decompression algorithm: you cannot unpack an instinct back into the episodes that produced it, because the episodes are gone, and what replaced them was never in words to begin with. The master is not withholding anything. He genuinely cannot hand it over.
A jazz pianist of forty years’ standing voices a chord behind a singer who has drifted slightly flat, and the voicing rescues the phrase. His student asks why that voicing. He says something about leaving room, which is true and useless, because the real answer is some thousands of nights of hearing what crowded a singer. Ask him to do it slowly and explain as he goes, and the playing degrades — he starts choosing voicings a competent sideman would choose.
The identical thing happens in a reading room. A radiologist flags a nodule a second and a half into a chest film and is right. Asked to justify it, he produces a plausible account about margins and spiculation, and the account is a reconstruction rather than a transcript: he saw it, then explained it. Hand that explanation to a resident and the resident finds nothing, because the explanation was never what did the work.
There is direct experimental evidence that articulation damages the skill rather than merely failing to transmit it. Flegal and Anderson found that experienced golfers asked to describe verbally what they had just done while putting performed worse afterwards, while novices were unaffected. The asymmetry is the finding. Novices have nothing compressed, so narrating costs them nothing; experts do, and narration forces the compressed thing back through a channel too narrow to carry it, and something breaks on the way.
The perceptual side has an older and even sharper result. Chase and Simon found that after five seconds with a position, a chess master replaced about sixteen of twenty-five pieces correctly, against a class-A player’s eight and a beginner’s four — and the entire advantage vanished on randomly arranged boards. That last clause is what matters. The master was not seeing harder or remembering better in any general sense. He was seeing different objects, supplied to perception by long-term memory, and when the board stopped containing those objects he was an ordinary person with an ordinary span. Expertise had restructured what was there to be perceived. You cannot lecture someone into perceiving different objects.
Dreyfus completes the picture from the other end: force experts to justify their decisions by explicit rules and their performance degrades back toward merely competent. So an apprenticeship is not, and has never been, a transfer of knowledge. It is a supervised environment for growing your own instrument — which is why it takes years, why it cannot be compressed into a course, and why the mentor’s real function is to put you in front of enough cases and tell you when you were wrong, rather than to explain.
Two things follow immediately. Stop asking experts to narrate while they work, because you are degrading the performance you called them in for; debrief afterwards instead, and treat what they say as a hypothesis about what they did. And stop reading an inability to explain as evidence of not knowing — the interview question that rewards articulate accounts of judgment is selecting for articulacy, which is a different trait and more common.
The limit is that none of this licenses mystique, and the field is full of people who have noticed how convenient this argument is. Plenty of claimed tacit knowledge is unexamined habit wearing a robe: a practitioner doing something arbitrary for twenty years and calling it feel. The test is behavioural and it is not negotiable. Are the person’s judgments reliably better than a competent alternative, checked against outcomes where outcomes exist? If they are, the inarticulacy is a real compression artefact. If they are not, the poetry about intuition is just poetry, and the right response to it is the question of what they have been right about lately.
The fault line in human work does not run between the sciences and the arts. It runs between working inside an inherited evaluation function and having to supply your own — and that cut puts people on sides our institutions would find surprising, with some physicists on the novelist’s side and some artists on the accountant’s.
The reason is structural. An evaluation function is a field’s apparatus for saying no: the replication, the statistics, the referees, the merciless indifference of the apparatus to what you hoped. Normal science is thick with it, and that thickness is what makes normal science trainable, cumulative and fast. But the apparatus is built out of the current framework’s assumptions about what counts as a result, which means at the exact moment the framework is what is in question, the apparatus goes quiet.
When Darwin was assembling natural selection across those twenty years, no procedure existed that could tell him he was right. There was no experiment. There was a vast pile of heterogeneous observations — barnacles, pigeons, finches, the fossil record, the breeders’ pamphlets — and a single idea that either organised them or did not, and the only instrument available for judging whether it did was Darwin. He delayed for two decades, which is not the behaviour of a man with a verifier.
Einstein’s case is cleaner still, because the thing he was reconstructing was simultaneity itself. There was no test he could run before the theory existed to be tested, since the theory was what specified which measurements were even meaningful. By every account from the people who have done work of this kind, it feels precisely the way creative work feels: long stretches of not knowing, reliance on a privately grown sense of rightness, aesthetic criteria doing load-bearing structural work rather than decorative work. Kuhn drew the same line from the historian’s side — normal science as puzzle-solving inside shared criteria, revolutionary episodes as periods when the criteria themselves are in dispute and there is nothing neutral to appeal to.
Then run the inversion, because it is the half people miss. A session guitarist on a Tuesday afternoon is doing art and is in a verifier-rich regime: there is a chart, a click track, a producer with an opinion, and a take either sits in the pocket or does not, with the answer arriving in ninety seconds. He can get measurably better every year of his career, because something tells him he was wrong, immediately. He is on the same side of the line as the biochemist running a standard assay and the accountant closing the books — all three executing inside an inherited function that does the discriminating for them.
The within-career version is sharper than the between-career one, and it is the version that applies to you. The same structural biologist crosses this line several times a week. Choosing which protein to spend the next five years on is verifier-poor in the extreme — nothing can tell her, and the honest answer may not arrive before the grant does. Running the crystallography on the protein she chose is verifier-rich: the diffraction pattern resolves or it does not. Same person, same morning, two regimes requiring opposite behaviour, and most researchers have never been taught that they are different.
So use this cut rather than the institutional one, which sorts by subject matter and is nearly useless. It explains why the folk ranking feels wrong to people who cannot articulate why: they have noticed that the paradigm-founder and the novelist are doing the same kind of thing, and that the departmental boundary between them is an administrative artefact. And it tells you what to look at when you assess a role. Not the discipline. Whether anybody in the building can tell the person they are wrong, and how long it takes.
The limit is that this is a gradient rather than a wall, and treating it as a wall produces its own errors. Darwin had a great deal of inherited apparatus even while founding a framework — comparative anatomy, stratigraphy, Malthus — and the novelist has craft. Most careers straddle, most weeks contain both regimes, and almost nobody sits at either pole for long. The useful question is therefore not which side someone is on but what proportion of their week runs in each, and whether they have noticed, because the behaviour that is correct in one regime is actively destructive in the other.
The features our institutions are proudest of measuring — depth of codified knowledge, abstraction, symbolic manipulation, everything that appears on a transcript — are disproportionately the verifiable ones. That is not a coincidence and it is not a conspiracy. It is a selection effect operating over a century, and it has left us with a measuring apparatus that can only see one half of what people are good at.
The causal chain is short. A skill becomes certifiable when there is a reliable test for it. A reliable test is also precisely what a machine needs to train against. So legibility made a skill certifiable, and certifiability made it mechanisable, and credentialing and automation have therefore been aimed at the same target the whole time, for the same underlying reason, without either noticing the other. We spent a century building an apparatus to identify and reward exactly the capabilities that were about to become abundant.
Set two people side by side. An acquiring editor at a publishing house reads forty pages of a manuscript and knows the book is alive, in a market where most such judgments are wrong and hers are right often enough to have built a list. No examination exists for what she just did. There is no certificate, no qualification, no portable evidence beyond a backlist that anyone can attribute to luck, and in a hiring process she is outranked on paper by a graduate with three certifications in editorial project management.
Her counterpart sits in an engineering consultancy, deciding which of eleven possible bids to chase with a team that can only prepare three. Getting that right is the whole business — the firm’s margin lives or dies on it — and the partner who is good at it cannot explain how, cannot train it in under a decade, and would struggle to describe the skill on a CV without sounding like he was describing a personality. Meanwhile every engineer in the building holds a chartered qualification certifying the part of the work that a test can see.
The empirical signature has already inverted the folk ladder, which is the part that should make everybody uneasy. Eloundou and colleagues scored every occupational task for language-model exposure and found that requirements for science and critical thinking were negatively associated with exposure, while programming and writing skills were strongly positively associated. Read what that means rather than what it sounds like. It is not that writing is trivial or that science is safe. It is that the better-specified a skill is — the more cleanly you could write its exam — the more exposed it turns out to be, and we have spent decades steering our most capable young people toward exactly the skills whose exams are cleanest, on the grounds that those were the rigorous ones.
Meanwhile the traits we have always been faintly embarrassed to put on a CV because they sound like vibes — taste, judgment, knowing which problem is worth attacking, the capacity to work for two years with nothing confirming anything — are the ones with no verifier. Which is why they resist automation. And why we never learned to measure them. Those are the same fact stated twice, and the whole argument of this piece collapses into that sentence.
If you are choosing what to spend a decade on, the sorting is therefore no longer by subject matter and has not been for a while. Ask whether the skill you are buying has a cheap external test. Where it does, you are acquiring something legible, certifiable, portable and increasingly abundant, and you should expect it to be worth less in ten years than the brochure implies. Where it does not, you are acquiring something nobody will give you credit for, nothing can take from you, and no machine has a signal to grind against.
The limit is that this is not an argument for abandoning credentials or a romance about the illegible. Credentials are not worthless; they certify real capability, they let strangers trust a surgeon, and a world that stopped testing anything would be considerably worse than this one. Nor is the illegible half automatically more valuable — most of what cannot be measured cannot be measured because it is not there, and an uncertifiable claim to judgment is usually just an uncertifiable claim.
The narrow point is the only one worth keeping: we have been pricing two halves of human capability with an instrument that can see one of them, we did not know we were doing it, and the relative scarcity of the two halves is reversing underneath us while the instrument goes on reporting as though nothing has changed.
Creative people are not smarter than scientists. Scientists are not smarter than creative people. The question was malformed, and the malformation has a precise shape: it compares inherited scaffolding against personal climb and reports the sum as though it were the second.
But the folk hypothesis has a real kernel, and the sixteen principles let us name it exactly. Our picture of intelligence is anchored on the traits that are easy to certify — depth of codified knowledge, abstraction, symbolic manipulation — and those are precisely the traits that a verifier-rich field makes certifiable in the first place. The traits that dominate verifier-poor work are, by construction, the ones no test can see. So when someone says the creative people are the smart ones, they are pointing at something true through a bad instrument. What they have noticed is that the novelist does without help what the scientist does with four hundred years of assistance, and that nobody subtracts the difference.
What they get wrong is the inference — that the scientist must therefore be coasting. Elite science is not coasting, and paradigm-founding science is among the hardest things anyone does. It is simply hard in the way we normally reserve for artists.
The practical residue of all sixteen is one question, and it is worth more than any ranking: who can tell you that you are wrong, and how fast?
Where something can — a compiler, a test suite, a market, a customer who returns or doesn’t — generate aggressively, delegate freely, fail cheaply, and let the verifier do the discriminating, because it is better and faster at judging than you are. Agonising there is simply slow. Where nothing can, invert all of it: volume is useless, delegation is dangerous because you cannot check what comes back, and the most confident voice in the room is providing no information at all. What matters instead is the slow construction of the instrument, and the patience to be wrong for long stretches with nothing confirming anything.
And if you are choosing what to spend a decade on, the sorting is no longer by subject matter. It is by whether the skill you are buying has a cheap external test. Where it does, you are acquiring something legible, certifiable, and increasingly abundant. Where it does not, you are acquiring something nobody will give you credit for and nothing can take from you.
The ratchet is the greatest thing our species ever built. It is also the reason we have never learned to see the people who work without one.
Eighty tasks scored across twenty-five dimensions, plus a purpose-built library of 104 primary sources spanning task-complexity theory, cognitive load, expertise and deliberate practice, tacit knowledge, creativity, uncertainty, feedback environments, computational complexity, coordination, occupational task analysis and AI task exposure. The task scores are structured expert judgments rather than measurements and are treated here as a reason to look for a mechanism, never as proof of one.