The Price of an Unlabelled Claim
A tuition page is a claims document. What happens when you read one the way you would read a preprint.
By KW Norton.
The exercise
AI-assisted schooling is arriving in my own metropolitan area at a price point of fifty thousand dollars a year for kindergarten through eighth grade, at a campus that opens in the autumn of 2026 and therefore has no local operating record. The figure is published as a single flat number: no tiers, no aid schedule, no statement of what is included and what is billed separately.
I am not interested in whether that price is too high. Families decide that, and a school is entitled to charge what someone will pay. I am interested in something narrower and more checkable. When a parent hands over that amount, what exactly have they been told, and could they verify any of it?
So I read the page the way I read a preprint. Every strong claim gets two things attached to it: a status label saying what kind of claim it is, and a falsifier saying what observation would sink it. This is the same discipline I apply to my own work, including the claims I most want to be true. It is not an adversarial reading. It is the only reading that respects the reader.
Six claims, audited
What follows is the standard vocabulary of this sector, not the invention of any one school. Every phrase here recurs across the category.
Students “crush their academics in 2 hours.”
What is missing — Which academics, measured how, against what full-day baseline. Two hours of what activity, at what completion standard, is the entire content of the claim and none of it is specified.
Honest label — Marketing summary. Not a measured result until the instrument and the baseline are named.
Classes rank “in the top 1% nationally.”
What is missing — The assessment, the norming population, the year, the unit of analysis, and who computed the percentile. The supporting link points inward to the school’s own program page rather than outward to an assessment body.
Honest label — Self-reported percentile on an unnamed instrument. Unverifiable from outside.
Children “learn twice as fast.”
What is missing — A rate needs a numerator, a denominator, and a comparison group. None appear. A doubling claim without a matched control is a testimonial with a number attached to it.
Honest label — Hypothesis. Falsifiable in principle, untested in public.
“Ranked among the best private schools.”
What is missing — The ranker. Rankings have authors, methodologies, and inclusion criteria; an unattributed ranking has none of these.
Honest label — Unattributed. Cannot be checked.
“Kids love school.”
What is missing — This is the honest kind of claim — affective, and supported by parent and student video. But the sample is selected by the party making the claim, which is the definition of a non-random sample.
Honest label — Selected testimony. Real, and not generalisable.
Accreditation status.
What is missing — The same footer states the campus “is a candidate for accreditation with Cognia” and “is accredited by Cognia.” Those two statements cannot both describe the same campus at the same time.
Honest label — Internally contradictory. One of the two is wrong.
The tell is in the placeholders
The page I read still carried its scaffolding. The frequently-asked-questions section reads, verbatim, “Q #1 / A #1” through “Q #5 / A #5.” The press section shows two entries of template text where article links belong. Two document links are labelled “update link.” The telephone field is empty. Above an events module reporting no events available sits the line “Save Your Spot — These Sessions Fill Fast!”
I want to be careful about what this does and does not indicate. It is not evidence of bad faith; pre-launch pages are assembled under deadline and every organisation has shipped a placeholder. What it indicates is sequencing. The persuasive surface was finished first and the answerable surface was not. Numbers, testimony and urgency were production-ready. The five questions a parent would actually ask were left as numbered blanks.
That ordering is the structural fact worth naming, because it is exactly the ordering that a demonstration-first project has to invert.
Why the two-hour figure is the interesting one
Set the marketing aside, because there is a real question buried in it. A compressed academic block is a plausible thing. Direct instruction wastes an enormous amount of a child’s day on transitions, on waiting for twenty-eight other children, on material already mastered. Recovering hours from that is not implausible at all.
But the claim being made is not “we recovered wasted hours.” It is “mastery in two hours.” Those differ in a way that matters, and the difference is testable. If the two hours are spent on adaptive drill to a mastery threshold, then what has been compressed is the acquisition of procedures that a well-built system can pace better than a classroom can. That is a genuine result and I would expect it to hold.
What no compression schedule can produce is time held against a hard question. Recursive time — sitting inside a problem past the point of comfort, without resolution, being asked another question instead of being handed the answer — is not an efficiency to be recovered. It is a duration that must be spent. An architecture optimised to shorten the academic block is optimising against precisely the variable I would want lengthened.
A system that measures its success by how quickly the question closes will get very good at closing questions. Whether that is education depends entirely on which questions they were.
This is not a rebuttal of the model. It is a falsifier for it, and a fair one: if students from a compressed two-hour academic day perform as well as a matched cohort on tasks requiring sustained, unresolved reasoning across days — not recall, not procedure, but the ability to hold an open question and defend a position under questioning — then my objection fails and I will say so on this page. As far as I can determine, nobody has published that comparison.
What the audit costs me
I have to run the same instrument on my own work or this essay is worthless. So: the memory-recall figure I have cited in my own material is a hypothesis with a named source and an unreplicated result, and I label it that way wherever it appears. My own account of a self-directed education begun at 73 is an existence proof with four confounds stated up front — it defeats one universal statement and establishes no average. The questioning instrument I use is documented at the Socratic Partner together with its failure modes, including the one where it flatters you.
None of that is generosity. A claims document that audits only other people’s claims is a claims document with a thumb on the scale.
The standard I would want applied to me
Three lines on a public page would settle most of this, for any school in the category:
- Name the instrument. “Top 1%” on which assessment, normed against whom, in which year, computed by whom.
- Name the comparison. “Twice as fast” against which cohort, matched on what, over what interval.
- Name the falsifier. State the observation that would cause you to withdraw the claim, and then watch for it.
A school that publishes those three lines is not weakened by them. It is the only kind of school whose numbers mean anything, and a parent reading it can tell the difference between a finding and a slogan. The absence of those lines does not prove a claim false. It does mean the claim cannot be checked — and an uncheckable claim attached to a fifty-thousand-dollar invoice is a transfer of risk from the institution to the family.
That transfer is the fee nobody itemises. It does not appear in the tuition figure, and it is paid by whoever has the least ability to audit what they bought.
Related: The N of One on what a single case can and cannot argue, and The Socratic Partner for the working instrument and its failure modes.