What we can point to, and what we can’t yet
The evidence this tool builds on, the study that could prove it wrong, and what a first pilot has to measure.
Classroom Sims is live at classroomsim.in and no classroom has used it. We have no monitoring data, no adoption data and no learning-gains data of our own. The district report in the pilot film uses demonstration figures. Everything below is either other people’s evidence about the mechanism, or what we intend to measure.
The evidence we build on
The mechanism is instruction re-targeted to what students actually know. Classroom Sims supplies the diagnosis; the teacher does the targeting.
Teaching at the right level
Randomized trials of Pratham’s model in four Indian states. Reading gains were 0.15 SD in Haryana, where government teachers taught it in a dedicated hour with government supervisors, and 0.70 SD in Uttar Pradesh, where volunteers ran in-school learning camps. Teacher-led arms in Bihar and Uttarakhand, with no such mechanism in the day, showed nothing. The mechanism is instruction re-targeted to what students actually know.
Diagnosis, then targeting, by software
Mindspark, Delhi middle-school students in Classes 6–9, allocated by lottery: 0.37 SD in mathematics and 0.23 SD in Hindi after 4.5 months. Its engine diagnoses each student’s level and then targets content. Scaled into Rajasthan government school timetables, it produced 0.22 SD in mathematics after 18 months, with gains proportional to time on the platform — and each school needed a laptop lab and a paid lab in-charge (Muralidharan & Singh, 2025, NBER 34205). Classroom Sims hands the diagnosis to the teacher and leaves the targeting to them, at a fraction of that cost.
Formative assessment, generally
The one formal meta-analysis puts the effect of checking understanding during instruction at 0.20 SD — 0.17 in mathematics and 0.09 in science — mostly in high-income classrooms (Kingston & Nash). The often-quoted 0.4–0.7 SD comes from Black & Wiliam’s narrative review. We treat this as supporting evidence, not the load-bearing claim.
Targeted instruction as a “great buy”
The World Bank / FCDO / UNICEF / USAID panel rates targeting instruction by learning level among the most cost-effective interventions available to low- and middle-income countries.
The nearest study is a null, and it defines the bet
Diagnostic feedback reports and low-stakes monitoring were randomized across 200 rural Andhra Pradesh primary schools. Teachers in treatment schools worked harder when observed — and after a year their students did no better on independently administered tests. That feedback came once, about a month after a start-of-year test, at the student level, with no action attached. Classroom Sims differs on all three margins: every lesson, at the class level, with a reteach suggestion. Whether that difference produces use is the first thing a pilot must measure — and if it does not, nothing downstream matters.
What a first pilot measures
Four questions, in the order the theory can fail. No incentives are offered to teachers.
How misconceptions are detected
What the tool records, and what it never records
Diagnostic MCQs
Each question has four options. Wrong options are written to a named misconception, so choosing one maps to a specific conceptual error rather than a score. 497 NCERT topics, Classes 1–12, six languages. Items and their misconception mapping get subject-expert review before any pilot.
Response-level records
Every answer is stored as the option chosen, whether it was correct, and the misconception it maps to — by seat number or hand count, never by name. That is what makes distractor analysis possible.
No PII, ever
No student names, emails or phone numbers are collected. District and school views aggregate by topic and grade and create no per-teacher score. Districts own their data.
For researchers
De-identified, response-level data from the first pilot will be available for research under a data agreement, with ethics approval held by the research partner. We have no research collaborations to claim yet and are looking for one. If you study learning in Indian classrooms, write to us.
hello@classroomsim.in