Questions, answered plainly

As studies, with their method attached. A result published without its method is an assertion with a number attached, and a reader has no way to separate it from a well-designed study except by trusting us. So every study we run will be published with its method, its sample, and the conditions under which it would have come out otherwise.

Each research direction on this site also states the finding that would end it, and its status — sustained, revised or retired — is recorded here with the date it changed. A reader can follow every direction as it moves, including when the outcome goes against us.

This is PrepGraph’s research, in PrepGraph’s own voice. PrepGraph is an AI learning company, working in India and the United States, from school to first job. It sells learning software to students, families and schools, and runs a research programme on how learners remember and build skills.

What makes it research rather than marketing is a property you can check: every position here could be argued against, and every direction states the finding that would end it. Judge it on that. If a section contains no position a competent, sceptical school leader could reject, tell us.

Two reasons, and only one of them is competitive. The other is that a mechanism described loosely becomes a claim nobody can test. Naming a technique tells a reader nothing about whether it survives a section of fifty, a syllabus that must be finished by March, and a teacher with four minutes between periods.

A method stated properly arrives with its assumptions and its failure modes, and we publish methods only in that form, inside a study. So judge this work on whether the problems are stated correctly: that is harder to fake than a diagram.

The work can fail, and says how. Every direction on this site states the condition under which we would abandon it, and when that condition is put to the test the outcome is recorded here with its date, including an outcome that ends the direction. A direction without such a condition would be an essay; every direction here carries one.

Mostly in where the difficulty is thought to lie. The common claim treats personalisation as a delivery problem, as though the hard part were getting the right material in front of the right student and the knowing were already solved. We think the difficulty is in knowing what a student understands, and that what is done with that knowledge afterwards is the easier half. The argument for that, and the argument about what the mark can and cannot carry, is made in full on the position page rather than restated here.

It is also worth saying that the state is not absent from this problem, which a page like this can imply by never mentioning it. National assessment surveys, the competency-based direction in national policy, reform of board question papers and continuous assessment inside schools are all attempts at the same input. Our position is not that they do not exist but that they arrive at the wrong grain and the wrong moment for a teacher: a survey reports on a system, a reformed paper reports on a year, and the decision that matters is taken on a Monday morning about the children who did not follow Friday’s lesson. That is a position someone in policy can disagree with, and we would welcome the gap closing from the public side.

The PrepGraph Council for Responsible AI & School Innovation – India is a forum on responsible AI in Indian schools, convened and funded by PrepGraph. It certifies nothing and endorses no vendor, PrepGraph included, so it never appears as a badge on our products.

Its terms, and how institutions engage with it, are published on its own site at council.prepgraph.com, and that text governs.

Product data is handled under the contract and privacy policy of the platform. Research involving students is a separate matter, with terms of its own: it begins only after an ethics review has cleared it and the school has agreed in writing to what is being asked and what it will cost in pupil, teacher and instructional hours, and a school can withdraw at any time.

The terms on which the commercial platform holds a school’s data are set by the contract a school signs with PrepGraph. That contract is the document that binds us there, and it is not a research authorisation.

A direction that cannot fail is not research, so each one carries a condition under which we would abandon it. These are the general ones. The specific falsifier belongs with the direction it applies to. When one of these is settled, its outcome is reported here with its date, including an outcome that goes against us.

The unit of interest is a teacher’s decision, not a teacher’s performance. A system that displaces the teacher’s judgement also degrades the measurement, because the teacher is the only party who sees the student in the room and can tell a gap in understanding from a bad morning.

An account detailed enough to be useful to a teacher can also be aggregated into a ranking of teachers by whoever holds it, whatever it was built for, and a stated intention is not a safeguard against that. A school should ask any supplier, in writing and before signing, whether teacher-level aggregation of such an account is prohibited by the contract, who inside the school would be able to produce it, and what happens to that ability when the head of school changes. An answer that describes only what the supplier intends should be read as a no.

An account of what one learner has shown, kept for her by an AI tutor rather than in a teacher’s head, and kept beyond the lesson, the term and the teacher who saw it: synthetic, because it is made; memory, because it lasts. It is not a claim about students’ intellect, and nothing here ranks anyone as more or less intelligent.

The distinction that matters is between knowing that a student scored sixty-two in a unit test and knowing which idea is missing and what should happen in the next lesson. Whether the useful unit is the student, the topic, or something narrower again is an open question, and we treat it as one. PrepGraph’s tutor already keeps a memory a learner can read and erase; how such an account is kept true, carried and checked is the research.

The constraints are not the setting, they are the research. Sections of forty to sixty, a syllabus that must be completed before a board examination, shared devices, an attainment spread inside one room that would be treated as several classes elsewhere, and a medium of instruction that is often a second language at home.

A method that assumes one device per child, unbounded time and a common first language is not a method here; it is a method somewhere else that someone would have to argue is portable. Portability is an empirical claim rather than a background assumption, and it cuts both ways: nothing developed here should be assumed to travel outward either.

altme.org is a separate initiative on a person’s rights over their own memory. The asymmetry is deliberate and follows from what each thing is for. Work that asks the public to inspect, adopt or participate in it owes openness as the price of asking. Work that sits inside a commercial product has two honest options, to publish a method properly with its evidence or to describe nothing at all, and we have taken the second here. The middle option, describing a mechanism impressively without exposing it to test, is the one that produces the material everybody complains about.

School leaders deciding whether any of this deserves their attention, teachers who want to see what is being claimed on their behalf before it arrives in their classroom, and researchers and policy people who will judge the framing rather than the software. It is also for our own staff: every sentence here is one they can hold the company to.

If you are looking for PrepGraph’s products: the company and its tutor are at prepgraph.com, courses and certificates at courses.prepgraph.com, hiring at talent.prepgraph.com, the teacher programme at academy.prepgraph.com, and the school programme at partners.prepgraph.com.

Yes. The most useful thing a school brings is not access but constraint: the question that keeps recurring in a staffroom, the thing that fails in the second week of every new system, the reason last year’s initiative was quietly dropped. That conversation is open now, on the terms below.

Use the contact route on this site for research correspondence. Council matters go to the Council directly at council.prepgraph.com.

A message that names the constraint you are working under, the size of the class, and the decision you are trying to make will get the most useful reply. How our systems are built is the one thing we do not discuss. A question about how well something works gets the answer this site gives throughout: outcomes are published as studies, with their method attached.