Solving the Sheepskin Effect

His strongest evidence is the sheepskin effect. The returns to education cluster in graduation years. Three years of college without a diploma is worth a fraction of four years with one, though the difference in learning is a single year. If school were mainly building human capital, the payoff would accumulate steadily. It doesn’t. It arrives when you cross the line.

Caplan puts the split at roughly 80 percent signaling, 20 percent skill. We could argue about the number, but the direction is not seriously in dispute.

The consequence is a zero-sum game. If the degree mainly sorts people, everyone has an incentive to get one, and each new graduate makes the last one’s credential worth slightly less. Students spend more years and more money to hold the same relative position. American degree-granting institutions spend around $700 billion a year. Students carry about $1.8 trillion in debt. Caplan’s conclusion is that we should stop subsidizing the race.

Signals do real work. No employer can see productivity at the moment of hire. They can only see proxies. The better the proxy, the better the match between a person and a job, and matching is not zero-sum. Better matching means lower churn, higher productivity, less time between jobs, higher incomes. Delete the signal and you do not free the economy from a wasteful contest. You blind it.

The literature makes the point precisely. Altonji and Pierret showed that employers lean on education early in a career and lean off it as they learn what a worker can actually do. The weight on the credential falls; the weight on hard-to-observe ability rises. Read that finding forward. The degree is a stopgap for information employers do not have yet, and they abandon it the moment they get something better. The gap between what an employer knows on day one and what they know in year three is exactly what a better signal is worth.

So the problem is not that we signal, but the instrument for doing so. The most important sorting mechanism in the labor market runs on a device that costs four years and six figures and outputs a few bits: degree or no degree, elite or not, and a GPA nobody trusts. It’s the most expensive low-bandwidth data transfer in the economy.

If such a device would exist, employers would require the brain map instead of the degree. Not out of hostility to universities, but because the report is better, faster and cheaper. Applicants would follow, because $200 beats $100,000 for the same job. The teaching would then have to survive on its own merits, priced against everything else a person could do with four years and the money. That’s the honest test of what higher education sells. Most of it doesn’t survive the brain mapper.

Unfortunately, the device doesn’t exist, but something close enough to it can be built now.

The important thing is the byproduct. A system that teaches you and then tests you necessarily knows an enormous amount about you: what you know, at what depth, how fast you got there, what you do when you’re stuck, whether you finish. Run that across a full curriculum and you have a continuously updated map of a person’s capability. Add serious collaborative project work and the map extends to everything a transcript has never captured: how someone performs against a deadline, with other people, on work that’s real.

That map is a knowledge graph, and it’s a better signal than a diploma has ever been, granular and current, while the degree is a high-level snapshot of someone still in development. And it’s verified, unlike a resume.

A graph is also machine-readable, so the endpoint is not a document at all. An agent holding your record can read a job description, judge the fit, apply on your behalf and tell you exactly which parts of your history to raise in the room. On the dimensions an employer actually cares about, it knows you better than you know yourself.

That’s the brain mapper, approximated in software, at a price much closer to $200 than to $100,000.

Of course they did. Online courses competed with the part of the university that was never the product. Nobody pays six figures for the lectures. You can get better lectures for free on the internet.

To displace a university you have to displace its signal. That’s the position nobody attacked, because until recently nobody could produce a signal an employer would accept in place of a diploma. Now it can be built, and whoever builds it takes the part of higher education that carries the value.

The moat has the same shape as the incumbent’s. A credential is worth whatever employers believe about the population standing behind it, and that belief compounds. Better members produce better outcomes. Better outcomes make the record more credible. A more credible record attracts better members. It’s a two-sided network in which the currency is trust, and the incumbents’ trust is depreciating in public.

Which changes who pays. Universities are paid up front by the student for seat time, regardless of what happens next. We think the signal should be paid for by the party that wants it, and that the institution producing it should hold a real stake in the member’s long-term success.

Employers. Recruiting is a market measured in hundreds of billions of dollars a year, in which intermediaries charge 15 to 30 percent of first-year compensation to forward resumes a hiring manager could have read themselves. Our placement fee is 10 percent, against a pool already filtered on cognitive ability and proof of agency and continuously verified afterwards. We can charge less and deliver more because the expensive part, finding out whether the candidate is any good, is already done and doesn’t have to be repeated for every employer who asks.

Members. Online membership is free. Campus membership is $449 a month. No tuition, no four-year lock-in, no debt. (Pricing is indicative for the first cohort.)

Equity. Membership is priced partly in equity in whatever a member builds: one percent on joining, and one percent for each year of membership.

The last line is the one to look hardest at. Our revenue is a claim on our members’ outcomes. A university is paid the same whether its graduate founds a company or never finds work. We are not. The incentive to select well, teach well and signal accurately is structural here, not aspirational.

The sheepskin effect is not a quirk in the data. It is the business model. Value is concentrated at a line, so the institution charges for the line. The fourth year costs what the first year costs, though what you learn in it is worth no more, because the fourth year is the one that hands over the certificate.

Everything cruel about the arrangement follows from that. A student who leaves after three years has paid three quarters of the price for a fraction of the signal, and the debt does not adjust. So people stay enrolled for reasons that have nothing to do with what the next year will teach them. And the person punished hardest for leaving is the one with the best reason to go: the job offer, the company, the honest realization that the program is wrong for them. The threshold selects against exactly the people worth having.

FrontierU doesn’t grant a degree, and that is deliberate. Members who want an accredited one enroll in parallel with a partner university at a fraction of traditional tuition, and that stays their choice rather than our product. What we issue is not a certificate awarded at the end. It’s a record that updates the day something happens: an exam passed, a project shipped, a skill verified. It is worth something from the first month, and none of it sits behind a line you have to cross.

Remove the line and the dropout disappears as a category. Nobody drops out of FrontierU. Members arrive when it serves them and leave when it stops serving them, and eight months of membership is worth eight months of record whether or not a ninth follows. Membership is open-ended and can be paused or ended at any time. A member who leaves to run a company can come back, and the record picks up where it stopped.

The obvious objection is that the diploma proves persistence, and that a system without a finish line cannot. The truth is the reverse. Persistence stops being inferred from a certificate and becomes visible in detail: what someone actually finished, how long they stayed with the hard thing, what they shipped when it got difficult. It is the trait the diploma was standing in for, measured instead of assumed.

It disciplines us, too. An institution that can’t lock a member in for four years has to be worth paying for this month.

Higher education is tiered, and the tiers are the business. A top university can charge tens of thousands a year because its degree is held by people who were already outliers at eighteen. The admissions office does the sorting; the four years package it. Graduates get priority access to the most sought-after jobs and to the networks that follow. That’s why elite prices hold while the average degree’s value falls.

Any serious challenger has to compete there. A better learning product aimed at an elite institution misses, because the elite institution is not selling learning.

So we compete on selection. We admit roughly the top 2 percent of applicants, and we admit them on raw talent and drive. No legacy preference, no DEI allocation, no athletic recruiting. Every seat awarded on a basis other than ability and drive is noise pushed into a signal that employers are paying to read, and it makes the credential worth less to everyone holding it.

Where the elite institutions read SAT scores and high school transcripts, we test cognitive ability directly. A tested IQ of 120, roughly the top 10 percent of the general population, is a hard floor. Above it, the odds rise with the score.

The reason is reach as much as prediction. An SAT score is partly purchased: enough tutoring moves it, and the tutoring goes to the households that can pay for it. A high school transcript records a particular school, a particular curriculum and a particular set of parents, and it’s close to unreadable outside the country that produced it. Both instruments measure how much investment a young person has already received. A cognitive test, sat cold, measures what they can do.

That matters most for the applicants the incumbents never see. The seventeen-year-old whose national transcript nobody in a US admissions office can interpret. The one whose family could not buy test prep. The one who has never appeared on anybody’s list. A test they can sit from anywhere, at almost no cost, is the widest net available. Then give them what actually converts potential into output: peers of the same caliber, a tutor that adapts to them, campuses to build on, and employers who can finally read what they can do.

Widening access means finding potential and funding it. It doesn’t mean lowering the bar and calling the result inclusion.

Socrates, our AI, arrives in stages, and the sequence is the company. First as a mentor, building each member a curriculum and guiding what to learn and build. Then as a tutor that teaches the material rather than pointing at it, alongside the jobs platform where employers hire from a pre-filtered pool. Then as an examiner that administers exams and issues verified records of what a member knows and can do, and as a career agent that finds the roles and makes the introductions.

Each stage adds to the graph, and the platform’s project work adds the rest. By the final stage the record describes a person better than any transcript ever has, and the member decides what is public and what is not.

There is another conclusion available to anyone willing to build. Signals are worth keeping. They are simply overpriced by orders of magnitude and starved of resolution. Cut the cost of producing one by a factor of a thousand, make it richer than anything a university has issued, and sell it to the employers who need it most. The four-year degree then loses the last argument it has. That's what FrontierU is building.