Three Hours of Mastery, Then Go Build Something

For most of this site’s life I have written about what is wrong with school. That was the easy half. The harder half is to say what a school should look like instead, precisely enough that someone could build it and then be shown to be wrong.

Here is my attempt.

The shape of the day

A student arrives and spends the first block of the day, somewhere between two and three hours, on the core academics: reading, writing, math, science, history. Not in a room of thirty moving in lockstep, but each at their own point in the material, using software that will not let them advance until they have shown they have it. Ninety percent or better, and checked again later without warning.

An adult is in the room. Not lecturing. Watching for three things: who is stuck, who is coasting, and who is about to give up. Stepping in for those and almost nothing else. Alpha School calls these people guides rather than teachers, and the distinction is real. The software carries the explaining and the drilling. The human carries motivation, standards, and character.

When the block is done, it is done. The rest of the day belongs to real work: building something, selling something, writing for an audience, training for a sport, learning a trade. Not a simulation of work. The thing itself, with real customers and real consequences.

That is the whole design. Everything else is detail. But the details are where it turns out to be true or not, so let me take the three claims inside it one at a time.

Claim one: the fundamentals can be learned in a few hours a day

This is the claim most people flinch at, and it is the one with the best evidence.

In 1984 Benjamin Bloom published a short paper describing what happened when students were taught one-to-one by a tutor using mastery methods. The average tutored student performed better than 98 percent of students in a conventional classroom, an effect of two standard deviations (The 2 Sigma Problem). Bloom’s question was how to get that result without a tutor for every child, because no society can afford one.

For forty years the honest answer was: you cannot, quite. Later research put the number lower. Kurt VanLehn’s review of tutoring studies found human tutors averaging closer to 0.8 standard deviations, and the best intelligent tutoring systems within a rounding error of that (VanLehn, 2011). Kulik and Fletcher’s meta-analysis of fifty studies found a median effect for tutoring software of about 0.66 (Kulik & Fletcher, 2016). Smaller than Bloom’s two, but enormous by the standards of anything else in education, and it comes from software a decade older than what exists now.

The reason this translates into time is not mysterious. A conventional class moves at the pace of the room, and spends most of its hours on material that any given student either already knows or is not ready for. Mastery software spends every minute at the edge of what one student can do. I did the arithmetic on a school day elsewhere: once you subtract transitions, announcements, review for whoever was absent, and the half of every lesson pitched at someone else, a seven-hour day contains a small number of hours that are actually about you. The software does not add hours. It stops wasting them.

There is a version of this argument that goes too far, and I have made it. It says that since AI can do the arithmetic and write the essay, students no longer need to learn either. That is wrong, and the research is clear about why. Kirschner, Sweller, and Clark’s review of minimally guided instruction found that novices learn fundamentals best through explicit instruction and practice, not through discovery or delegation (Kirschner, Sweller & Clark, 2006). The point of the morning block is not to skip the fundamentals. It is to get through them faster, to genuine mastery, so that the afternoon becomes possible. You cannot build on what you do not have, and you cannot tell when the machine is wrong if you never learned to do it yourself.

Claim two: the freed time should go to real work

Give a fifteen-year-old four free hours a day and most schools would fill them with more school. The design above fills them with something a school has almost never offered: real stakes.

The precedent is older than the technology. Acton Academy, the microschool network Alpha grew out of, has run Children’s Business Fairs for years, where kids build a product, set a price, and sell to strangers in a parking lot. Founders School, which opened this fall, extends that into a four-year high school where the afternoon is a company: students sell before they build, validate before they scale, and aim at a real revenue target with the school’s own money attached to the promise. Whether the target is reachable is a fair question, and I come back to it below. Whether the shape is right is not, in my view, a close call.

The reason is motivation, and I wrote about the mechanics in Motivation Is the Product. A student who has to call a customer tomorrow morning finds a reason to finish the math today. The afternoon gives the morning its point. Separate the two and you have a faster version of the old school, which is better but not different. Keep them together and the fundamentals stop being an abstraction imposed by adults and start being the tools a kid needs for the thing they are actually trying to do.

Real work also teaches what school structurally cannot: rejection, ambiguity, deadlines nobody extends, the difference between finished and shipped. These are not soft skills. They are the skills, and the only way to learn them is to have something real go wrong.

Claim three: the adult’s job changes, and gets harder

The word “guide” makes some people nervous, because it sounds like a way to pay less for less. In the model I am describing it is the opposite. Remove content delivery from a teacher’s day and what is left is the hard part: knowing each kid, holding a standard without breaking a spirit, noticing the week before a student quits. Self-determination theory’s third need, after autonomy and competence, is relatedness (Ryan & Deci, 2000). Software has no answer for that need. A good adult does. The school should pay them like it matters, and judge them on whether the kids in front of them are getting better and staying in the game.

What is still unproven

I want to be careful here, because the schools running this model make claims that have not been checked by anyone outside them.

Alpha reports that its students grow at more than twice the national rate on standardized assessments and finish in the top few percent nationally. Those numbers come from the school’s own analysis of its own students, a selected group whose families chose an expensive alternative. No independent evaluation has been published. Founders School has a first class of twenty students and no graduates yet. A serious builder treats these figures as hypotheses to instrument, not facts to market. The peer-reviewed literature above is the reason to believe the model can work. The schools are the reason to believe it can be built. Neither is yet proof that it works at scale.

The critiques deserve an answer rather than a dismissal. Drill and kill: yes, the morning is drilling, and that is fine, because drilling the fundamentals to mastery is exactly what buys the afternoon, and the afternoon is the opposite of drill. The million-dollar target is unrealistic: probably, for most students, and I read it as a forcing function rather than a forecast. The number has to be large enough that the work has to be real. The commercial entanglements around these schools: a legitimate concern, and one that independent measurement would do more to resolve than any argument.

What I would build first

If I were building the software for a school like this, I would start with three numbers and one screen, and I laid them out in the previous essay: mastery checked later, time returned, and loops closed with the real world. The first screen a student sees in the morning should be designed to get them off it as fast as possible, having done something hard. The first screen a guide sees should be the short list of kids whose numbers slipped this week.

Everything after that is iteration. The design is not a secret, and it is no longer only a theory. A few schools are running it. The rest of us can watch, measure, and copy what holds up.