The Average Student Doesn't Exist

The Average Student Isn't There

Walk into a Class 5 classroom in a rural government school in India and ask the children to read a passage from their Class 2 textbook. According to ASER 2024, fewer than half of them can. Even by Class 8, roughly three in ten still cannot read a Class 2 text fluently, which means a meaningful share of the class is operating six full grades below the chapter their teacher is paid to deliver. The teacher in front of the room does not have the option to teach Class 2 today. She has Class 5 chapters, a syllabus deadline, and forty children whose actual reading levels span six grades.

This is the central, unspoken problem of mass schooling. The curriculum is paced for one student, and that student does not exist.

The fighter pilot who wasn't there

In 1952, the U.S. Air Force could not figure out why so many of its pilots were crashing planes that were, on paper, perfectly engineered. A young researcher named Lt. Gilbert Daniels measured 4,063 pilots across ten body dimensions like chest, sleeve, and thigh, with the goal of updating cockpit specs to fit "the average pilot." Then he asked a question nobody had bothered to ask. How many of the 4,063 pilots were actually average on all ten dimensions?

Zero. Not a single one. Even when he loosened the bar to the middle 30% on just three of those ten dimensions, fewer than 4% of pilots qualified. The cockpit had been designed for a man who did not exist, and every real pilot was paying for it. The Air Force responded by mandating adjustable seats, controls, and helmets. Performance improved and the mishap rate fell.

Todd Rose retells this story in The End of Average and argues, persuasively, that almost every system designed for "the average person" is making the same mistake the Air Force made, school very much included. A child who is on the curriculum's expected level in attention, vocabulary, prior knowledge, and pace, all on the same day, is roughly as rare as a perfectly average pilot.

The 2-Sigma Problem

In 1984, the educational psychologist Benjamin Bloom published a result so striking it has its own name. He called it the 2-Sigma Problem. Comparing students taught in conventional 30-to-1 classrooms against students taught one-to-one and held to mastery on each unit, Bloom found that the tutored students performed two standard deviations better. The average tutored student, in other words, outperformed roughly 98% of students from the conventional class. Bloom himself described the gap as the most striking thing in the data, and used it to set a challenge to the field. Find a group method that matches one-to-one tutoring.

That challenge is the closest thing pedagogy has to a Millennium Prize. Forty years on, no classroom-only intervention has come close to closing it. The few that have moved the needle share a single feature. They stop pretending all the students are at the same place.

The pacing trap

Every academic year, on the first day of every chapter, a teacher begins the lesson. Some students walked in with the prerequisites in place. Some are missing two chapters' worth of foundations from last year. A few already understand the material from a tutor or a sibling. The lesson starts anyway.

Two things happen, and both are well-documented in cognitive science. The students without prerequisites hit what John Sweller called cognitive overload. Working memory is pinned by sub-skills the lesson assumes are automatic, leaving no capacity for the new idea. They are not learning the chapter, they are pretending to. The students who are ahead operate below their zone of proximal development, Lev Vygotsky's term for the band between what a learner can do alone and what they can do with help, where almost all real learning happens. They are not learning either, they are waiting.

The middle of the class gets a partial signal. Then the chapter ends, the test happens, and a fresh set of prerequisites quietly fails to lock in for the next chapter. Compounded over a school year, the bottom of the class falls further behind on every chapter that depends on the last one. By Class 8, the curriculum and the actual classroom are speaking different languages.

ASER's Beyond Basics report, which surveyed 14 to 18 year olds in rural India, makes that disconnect concrete. About a quarter of teenagers cannot fluently read a Class 2 text in their regional language, and well over half cannot correctly do a 3-digit by 1-digit division problem, a skill expected by Class 3 or 4. These are not children who failed once. They are children who were taught for a decade by a system that never met them where they were.

We already know what works

The research on the alternative is not new and not controversial. It is just rarely implemented at scale.

Idea	Year	Finding	Implication
Vygotsky, Zone of Proximal Development	1978	Learning happens just above current ability, with scaffolding	Pace the lesson to the learner, not the calendar
Bloom, 2 Sigma Problem	1984	One-to-one mastery learning around +2 SD vs. conventional class	Mastery and responsiveness are where the gain is
Sweller, Cognitive Load Theory	1988	Overloaded working memory blocks new learning	Fix prerequisites before introducing complexity
Banerjee, Banerji, Berry, Duflo et al., TaRL	2017	Grouping children by level, not grade, sharply raises learning at low cost	Level beats grade. Evidence is from Indian classrooms.

Pratham's Teaching at the Right Level (TaRL) deserves particular attention because the evidence comes from this country, not from a Massachusetts charter school. In a series of randomised trials run with J-PAL across multiple Indian states, students were briefly grouped by what they could actually do rather than what grade they were in, and taught accordingly for an hour a day. Reading and arithmetic outcomes improved by large, durable margins, at a cost the public system could afford. India's National Education Policy 2020 explicitly names competency-based learning as a direction of travel and frames the closing of learning gaps as a priority. The diagnosis is in the policy. The instrument to deliver it, at the scale of every student in every classroom, is what is missing.

"Personalised" is doing a lot of work

A great deal of edtech labels itself personalised. Most of it is not. A homepage that shows different videos to different students is recommendation. A quiz that hands out the next question based on the last answer is adaptive sequencing only if it knows which sub-skill the question tests, which sub-skills it depends on, and which of those the student has actually mastered.

Personalisation is a measurement problem before it is a content problem. You cannot teach a student you have not measured, and you cannot measure a student without a model of the subject finer-grained than "Chapter 6: Light." The honest version of personalisation needs three things working together. The first is a knowledge graph that names the sub-skills inside a chapter and the prerequisite edges between them. The second is a per-student belief, updated after every response, about how likely it is that the child has mastered each of those sub-skills. The third is a planner that picks the next thing to do for this student, today, given those two.

It is more work than a recommendation widget. It is also the thing the research has been pointing at for forty years.

How we are approaching it

We built Clevriq around this idea. Each chapter in the syllabus is decomposed into a small graph of the sub-skills that compose it, with explicit prerequisite edges. Every response a student gives, whether in practice, a doubt, or a test, updates a Bayesian estimate of mastery on each sub-skill they touched, and an item-response model picks the next question to maximise information about what they actually know. Where there is a gap, the day's plan reprioritises to fix it before moving on. The chapter on the calendar still anchors the work. The order of the work inside the chapter is the student's, not the syllabus's.

None of this is magic, and we make no claim to have closed Bloom's two-sigma gap. We do think the right unit of personalisation is the sub-skill, not the chapter, and that the right place to start is by measuring honestly what each child knows. The system has been teaching the average student for a long time. The average student isn't there.

---

References

1. Bloom, B. S. (1984). The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring. Educational Researcher, 13(6), 4-16.
2. Daniels, G. S. (1952). The "Average Man"? Wright Air Development Center, Technical Note WCRD TN 53-7.
3. Rose, T. (2016). The End of Average: How We Succeed in a World That Values Sameness. HarperOne.
4. Vygotsky, L. S. (1978). Mind in Society: The Development of Higher Psychological Processes. Harvard University Press.
5. Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257-285.
6. ASER Centre. Annual Status of Education Report (Rural) 2024. Pratham, January 2025.
7. ASER Centre. ASER 2023: Beyond Basics, Rural Youth Aged 14 to 18. Pratham, January 2024.
8. Banerjee, A., Banerji, R., Berry, J., Duflo, E., Kannan, H., Mukerji, S., Shotland, M., and Walton, M. (2017). From Proof of Concept to Scalable Policies: Challenges and Solutions, with an Application. Journal of Economic Perspectives, 31(4), 73-102.
9. Ministry of Human Resource Development, Government of India. National Education Policy 2020. New Delhi, 2020.