The Best People
Description
There are two honest ways to choose the people who run things: a test, or a lottery. Test the candidates and promote the high scorers, and over time the people who can afford to prepare for the test become the people in charge — a single fixed definition of merit creates one preparation pipeline, and a class captures it. China's keju civil-service exams were open in principle and produced, over centuries, a self-reproducing scholar-gentry; this is Sandel's general argument about meritocracy hardening into hereditary advantage. Switching to a 'better' metric does not escape it, because the metric itself rots: Goodhart's law says a measure that becomes a target stops measuring, and Frankel and Kartik make it precise — as the stakes on a signal rise, the signal reveals gaming ability rather than the underlying trait. The opposite move, pure lottery, has no fixed target to capture, which is why Athens filled its Council of 500 and most offices by lot — but it elected its generals and treasurers, the posts where demonstrated skill mattered, so even the home of sortition drew a line. The sober hybrid is old: Ming and Qing China made you pass the exam to qualify and then assigned posts partly by lot, and Sandel proposes the same shape today (a threshold, then a lottery above it). But every such hybrid hits a wall. In 1977 Allan Gibbard proved that even allowing randomness, the only selection rule that treats everyone equally and cannot be strategically gamed is the pure lottery with no competence filter at all; any quality bar necessarily reintroduces a way to game it. Capture-resistance and competence therefore trade off in principle, not merely in practice. The series takes that wall as fixed and asks the only question it leaves open: not how to make merit un-gameable, but how to make gaming it expensive — and the effort spent trying worth something.
Sources & further reading (7)
- The Dark Side Of Meritocracy (Sandel) — Noemahttps://www.noemamag.com/the-dark-side-of-meritocracy/
- The Problem with Metrics — Goodhart's law (arXiv)https://arxiv.org/pdf/2002.08512
- Muddled Information — Frankel & Kartikhttps://navinkartik.com/Papers/FK-MudInf.pdf
- Athenian democracy — Wikipediahttps://en.wikipedia.org/wiki/Athenian_democracy
- The Logic of Randomness: Can Lottery Democracy Replace Electoral Democracy? — Springerhttps://link.springer.com/article/10.1007/s41111-025-00293-3
- Draw lots for Harvard entry, says Michael Sandel — Times Higher Educationhttps://www.timeshighereducation.com/news/draw-lots-harvard-entry-says-michael-sandel
- Gibbard's theorem — Wikipediahttps://en.wikipedia.org/wiki/Gibbard%27s_theorem
Script
Cold open
There are two honest ways to choose the people who run things. Give everyone a test, and put the high scorers in charge — and slowly, the people who can afford to prepare for the test become the people in charge. Or skip the test, draw names at random, and accept a fair sample of people who mostly don't know the job. We have been arguing about which is worse since Athens. The reason we can't simply take the best of both is not a failure of nerve, or of imagination. It's closer to a theorem.
Frame
Find the best people and put them in charge. It sounds like the one thing nobody could object to. But every system ever built to find the best people gets taken over — reliably, and with a mechanism — by one particular kind of person: whoever is optimized for the thing the system measures. This episode is about why that happens, and about a hard limit on escaping it — a result from nineteen seventy-seven that says cleverness won't save you. The rest of the series asks how much you can still do once you stop pretending the limit isn't there.
What goes wrong with the test?
Start with the test. Imperial China ran the most serious version ever attempted — the keju, competitive written exams, open in principle to almost any man, the single gateway into the entire civil service. And over centuries it produced a class: the scholar-gentry, the families who owned the tutors, the books, and the years of free time the exams demanded — and who passed that advantage straight down. One definition of merit creates one way to prepare for it. Whoever can afford that preparation takes it over. Open exam, hereditary result.
Isn't that just a bad metric — pick a better one?
Maybe that's just a bad metric — pick a better one. But the metric itself rots. Goodhart's law: the moment a measure becomes the thing you're selected on, it stops measuring what it used to. The economists Frankel and Kartik made it exact — as the stakes riding on a signal go up, the signal stops revealing whether you have the trait, and starts revealing how good you are at gaming the signal. The better the prize, the worse the test. Not because you picked the wrong test — because it's a test with a prize attached.
So drop the test and draw names at random?
So drop the test. Draw names at random — a lottery has no fixed target to capture, and no preparation to inherit. Athens did exactly this: its Council of five hundred, its juries, most of its offices, all filled by lot. But not all of them. The Athenians elected their generals, and elected their treasurers — the handful of jobs where getting it wrong got people killed, or robbed. Even the home of the lottery drew a line: some work needs demonstrated skill, and they would not leave that to chance.
Then filter first and randomize second?
Then do both — filter first, randomize second. This isn't hypothetical. Ming and Qing China made you pass the exam to qualify, and then assigned many posts partly by lot. Michael Sandel proposes the same shape for elite universities now: set a real bar, and above that bar stop ranking people and draw by lottery. Qualify, then randomize. It is an old, sober idea. And it is exactly the idea that walks into the wall.
Why can't we close the gap at all?
Here is the wall. In nineteen seventy-seven the philosopher Allan Gibbard proved something true of every possible selection rule. Even if you allow randomness, the only rule that treats everyone equally and cannot be strategically gamed is the pure lottery — the one with no competence filter at all. The instant you add any quality bar, any filter for merit, you put back a way to game it. Not as a practical nuisance you might engineer away. As a mathematical certainty.
Turn
So the argument everyone keeps having — what is the right test for merit — has no clean winner, and not because the right test is still out there waiting. Gibbard's result says any filter that screens for competence can be gamed; the only thing that can't be gamed is a filter that screens for nothing. Capture-resistance and competence don't only trade off in practice. They trade off in principle. The usual move from here is to pick a side and quietly swallow the cost — the meritocrats accept the hardening caste, the lottery's defenders accept the lost skill. This series makes a different move. It takes the wall as fixed, and asks the one question the theorem leaves open: not how to make merit impossible to game — you can't — but how to make gaming it expensive, and how to make the effort people pour into trying actually worth something.
Closer
Athens lived on both sides of this wall at once — lots for almost everything, elections for the generals — and never once pretended the tension away. That is the honest place to start. You cannot build a definition of merit that can't be captured. The open question is whether you can build one that is costly to capture — one where the only reliable way to beat it is to actually become good at something. Which points at an idea that sounds, at first, like a mistake. Don't fix the definition of merit. Let it move.