The steelman of the whole house — with its strongest opponent kept on the board

The Clerk and the Jury

TL;DRThe strongest version of Only Humans Score isn't 'AI bad, humans good' — it's separation of powers.

The strongest version of this house isn’t “AI bad, humans good.” It’s separation of powers.

The clerk
The machine presents; tirelessly, and well.

The machine is the clerk: it can present the record, generate the drafts, paint the pictures — perfectly and tirelessly. Nothing in this house denies that; the house runs on it. But scoring is a different kind of act. A score isn’t a measurement; it’s a claim that something mattered — and mattering requires a being for whom things can go well or badly.

The line the whole page turns on

A machine’s verdict is a weather report. A human’s verdict is a judgment — because the judge has stakes: a body, a reputation, a limited number of Tuesdays.

That’s jury logic, and it’s why verdicts bind. We don’t seat juries because twelve strangers are the most accurate instrument available; we seat them because a judgment delivered by people with skin — people who can be wrong in public and live with it — is the only kind a free society agrees to be bound by. The capacity to be hurt is the qualification to judge. Feeling isn’t the obstacle to objectivity; it’s the credential.

On that reading, this museum isn’t a nostalgia project about human specialness. It’s a constitutional design for the AI era: machines draft, humans ratify, and no output becomes valuable until a mortal says so. The clerk keeps the record; the jury scores it.

The strongest counter
Named out loud, so the steelman is honest.

Here is the best case against this house, at full weight: humans are noisy, biased, inconsistent scorers. Inter-rater reliability is famously poor; on defined rubrics, models already out-grade crowds. If scoring is what humans are for, why are we so measurably bad at it?

The house’s best reply is that the objection smuggles in its own defeat: a rubric is frozen prior human judgment. “The model scores better” means “the model matches old human scores faster.” There is no standard behind the audience to appeal to. The noise is the signal — aggregate human judgment isn’t an imperfect approximation of what “good” means; it’s what “good” means. Every rubric a machine can ace was pressed, first, by a jury.

That’s the altitude this house chooses to argue at — against the smartest opponent, not the easiest one. The devil’s advocate gets a fair hearing here; a steelman that hides its counter is a brochure.

Why it belongs in this house
The page is the game.

Notice what you’re holding: a machine drafted this page — the strongest case for the house and the strongest case against it — and a human read it end to end and pressed it. The clerk presented; the jury scored. If the argument moved you, that was the drafting. Whether it’s true is still yours to rule on. That’s the whole game.

Machines draft. Humans ratify. Nothing is valuable until a mortal says so.

Provenance (the no-lying floor). This steelman was drafted by the machine — one of the curator’s instruments, arguing the house’s best case and its best counter on request — and pressed to the wall by the curator’s hand. Its scope is honest too: it was argued from the front door, the Ethos, and the map — not a full crawl of every room. If a room in this museum contradicts it, the record wins and the page gets corrected on the wall. Only humans score.