The School of One Model
School of One in one page
- What it is: A maths-class model where every student receives a personalised daily lesson playlist, matched to their current level and learning gaps.
- Where it started: New York City public middle schools, launched 2009 with the city’s Department of Education. Founder: Joel Rose.
- The daily playlist mix: Each student’s day combines several modalities. Teacher-led instruction in a small group. Independent practice on a computer. Peer tutoring. Virtual tutoring with a remote instructor. Project work.
- What the algorithm does: Every night, software reviews each student’s exit-slip results from that day and assigns them the next set of activities. Students who got it move on. Students who missed it get a different activity on the same skill the next day.
- Why it matters: It breaks the assumption that every student in a class moves at the same pace through the same material on the same day.
A traditional maths class moves the whole room through the same topic at the same pace. The students who already understood the topic from last week are bored. The students who missed last week’s foundation are lost. The teacher, who has time for one explanation, picks the middle and accepts losing both ends.
The School of One asks what would happen if every student in the room got a different lesson each day, chosen specifically for what they needed to learn next. The answer is hard to deliver by hand. With algorithms, real-time data, and a mix of teaching modalities, it becomes possible at the scale of a real classroom.
How the day works
A School of One classroom does not look like a traditional one. The room is split into stations. Each student walks in, checks their daily playlist on a screen, and goes to the station the playlist tells them to start at.
A typical playlist mixes several modes within a single class period.
Live teacher-led instruction. A small group of students who need direct teaching on the same skill that day meets with the teacher in one corner of the room. The teacher leads a short focused lesson. Group size is small enough that the teacher can see every student.
Algorithmic practice. Students work alone on a computer through an adaptive practice tool. The tool adjusts difficulty based on how the student is doing in real time. Wrong answers trigger easier follow-up questions; right answers escalate to harder ones.
Peer-to-peer. Two students work together on a problem, taking turns explaining their reasoning. The pairings are chosen so that one student has just learned the skill the other is now learning.
Virtual tutoring. A student joins a video call with a remote tutor who walks them through a topic. This was an unusual choice in 2009 and is now common.
Project-based work. Students work on a longer task that ties several skills together. This is the integration step from earlier in the day’s playlist.
At the end of the class, every student takes a short exit ticket. The results feed the algorithm overnight, which sets the next day’s playlist.
Live teacher instruction, algorithmic practice, peer-to-peer, virtual tutoring, and project-based work.
The point is variety in one class period. A student who is bored or stuck in one mode often does well in another. Mixing modes also gives students different kinds of support and practice in the same daily session.
What the algorithm actually does
The technical core is a recommendation system tuned for learning rather than entertainment.
Every student’s mastery on every skill in the curriculum is stored as a number. Each exit ticket updates several of these numbers. The algorithm uses the updated numbers to pick, for each student, the next set of activities most likely to move that student forward.
Three rules shape the recommendation.
Match difficulty to readiness. A student who is just on the edge of mastering a skill gets practice on that skill. A student who is solid on it gets the next harder one. A student who missed a prerequisite gets sent backward to fill the gap before moving on.
Vary the mode. A student who has had three days of computer practice gets a teacher-led group session or a project. The algorithm tries to keep modes balanced over time.
Balance the room. The algorithm also has to fit the teacher’s capacity. A small-group teacher-led session can only hold so many students per day. The algorithm matches students to sessions partly on need and partly on capacity, so the schedule is feasible.
No piece of this is exotic. The underlying technique is similar to the recommendation engines that drive online streaming, with the success metric changed from watch time to skill mastery.
What changed when this was tried at scale
The first School of One classrooms launched in New York City public middle schools in 2009. The model was funded by the city Department of Education with private foundation support and run as a pilot in a handful of schools.
Independent evaluations of the early years gave mixed results. Some studies found significant gains in maths achievement compared with comparable classrooms; others found smaller gains or none. The implementation was hard. The data systems were new. Teachers had to learn a new way of running a classroom. The space had to be configured for parallel stations rather than a single front.
In 2011, the model spun out into a nonprofit called New Classrooms, co-founded by Joel Rose and Chris Rush, which now offers similar adaptive models under the name Teach to One. Schools across the United States have piloted versions of it.
The lesson is not that any one product is the answer. It is that the basic idea (a daily personalised playlist, generated from real-time data, delivered through multiple modes in the same room) is now feasible. The cost has dropped, the tools have improved, and the implementation has been tested. A school that wants to break the one-pace classroom now has a working pattern to copy.
What the model gets right
Three structural choices in the School of One model break the limits that the standard classroom imposes.
Pace is per student, not per class. Each student moves at the pace their own data justifies. A student who masters a topic in two days moves on; a student who needs ten days takes ten days. The class no longer waits for the slowest student or skips ahead past the slowest student.
Feedback is daily, not termly. Every student is checked every day. The teacher and the algorithm both see who is stuck and on what, in time to act on it the next morning.
Modality is mixed within one room. A student who needs a teacher today gets one. A student who is ready for solo practice does that. A student who works best by explaining to a peer does that. The classroom is not stuck with one delivery method per period.
These are exactly the things the brick-and-mortar classroom could not deliver. The School of One does not abandon the building; the students still gather in a physical room. It uses the room differently.
What the model still struggles with
The model is not finished. Three problems are still open.
Content quality. The algorithm is only as good as the activities it can recommend. Building a deep, high-quality library of activities for every skill at every difficulty level is expensive and ongoing. Most subjects do not yet have the kind of content that maths has.
Subjects beyond maths. Maths is the easiest subject to model this way because skills are well-defined and prerequisites are clear. Subjects with messier structure (history, literature, art) are harder to break into discrete masterable units, and trying too hard can flatten what makes the subject interesting.
Teacher role. The model demands a different kind of teacher: one who can lead small sessions, monitor data, coordinate with a system, and accept that they are not the only source of instruction in the room. Recruiting and training for this is a substantial change from the standard teacher prep programme.
The point is not that the model is perfect. It is that the standard classroom does not solve any of these problems either; it just hides them behind a single average lesson. A model that exposes the problems also gives a way to start fixing them.
Same-pace, same-feedback-timing, same-modality.
- Pace is per student, not per class.
- Feedback is daily, not termly.
- Modality is mixed within one room, not one mode per period.
The classroom still exists. What changes is how the time inside it is used.
Common misreadings
The School of One is not a model where every student stares at a screen all day. The screen time is one mode among several, and a substantial fraction of the day is still live human teaching.
It is not a model that replaces teachers. It changes what teachers spend their time on. Less broadcast lecturing, more small-group teaching and data review.
It is not a finished product. The first schools that ran it had real implementation problems, and not every pilot has produced strong test-score gains. The argument for it is that even the early implementations broke constraints the standard classroom could not break, and the model has been improving.
And it is not the only model in this space. Khan Academy, Knewton, Carnegie Learning’s Mika, ALEKS, and many others apply variations of the same adaptive-pacing idea. The School of One is the most cited because it tried it inside a public school system at meaningful scale.
How was this article?