# Case Studies and Research Evidence | SwaVid

Classroom validation case studies and research notes from SwaVid: what we tested with 150+ students, what teachers confirmed, and the honest limits of

Canonical: https://www.swavid.com/case-studies

Source: https://www.swavid.com/case-studies

# Evidence, with the limits attached.

## Validated with 150+ students, 85% teacher-confirmed accuracy

## Why diagnosis-first is the only correct architecture for AI tutors

## Learning before answering

## How SwaVid chooses the next question

Everything here is classroom validation or published learning science, clearly labelled as one or the other. No invented statistics, no testimonials, no efficacy claims. That constraint is the point.

Findings and limitations appear in the same report, on every page.

The full report from our school validation cohort: 365 submissions reviewed, real exhibits, what we changed, and what the evidence cannot claim.

The argument from published learning science: Bloom’s 2-sigma problem, the prerequisite structure of knowledge, and three decades of knowledge-tracing research.

A shorter visual thesis: why a useful tutor maps the learner and the prerequisite before it starts talking.

The reproducible method: the NCERT prerequisite graph, the evidence update, the candidate set, and the loss function that picks what to ask next.

## Key Links

- [Case study · Classroom validation Validated with 150+ students, 85% teacher-confirmed accuracy The full report from our school validation cohort: 365 submissions reviewed, real exhibits, what we changed, and what the evidence cannot claim. Read](https://www.swavid.com/case-studies/apple-global-school)
- [White paper Why diagnosis-first is the only correct architecture for AI tutors The argument from published learning science: Bloom’s 2-sigma problem, the prerequisite structure of knowledge, and three decades of knowledge-tracing research. Read](https://www.swavid.com/research/why-diagnosis-first)
- [Research note Learning before answering A shorter visual thesis: why a useful tutor maps the learner and the prerequisite before it starts talking. Read](https://www.swavid.com/research/learning-before-answering)
- [Research method How SwaVid chooses the next question The reproducible method: the NCERT prerequisite graph, the evidence update, the candidate set, and the loss function that picks what to ask next. Read](https://www.swavid.com/research/how-swavid-chooses-the-next-question)