# MyChild Engine — Validation Study (Plain-Language Summary)

> Why MyChild is back, and what the validation numbers mean in plain language. 71,939 synthetic profiles, Cohen's kappa 0.906. Not clinical validation.

Canonical HTML page: https://www.mychildapp.in/study-simple · Full study: https://www.mychildapp.in/study.md

## Founder note

"I built the first version of this engine in 2014, when I was 16. Launched the app in 2016. It screens for developmental delays based on CDC milestones, WHO guidelines, and Red Flags indicators. I ran the startup for a couple years, then wound it down. Almost nobody is still working on this. The few who are aren't open-sourcing it. With what AI can do now, I decided to bring it back. I'm not raising money or restarting the company. I'm open-sourcing the engine so any platform that reaches parents can use it, improve it, and help push it toward clinical validation. The real goal is to get this validated and give it away." — Harsh Songra

## The numbers so far

- **Cohen's Kappa 0.906** — agreement between engine and evaluator beyond chance; above 0.8 is "almost perfect" in clinical research
- **Sensitivity 91.4%** — when there's actually a concern, the engine catches it 91.4% of the time
- **Specificity 98.8%** — when a child is developing normally, the engine correctly says so 98.8% of the time (1.2% false alarms)
- **71,939 profiles tested** — synthetic developmental profiles, not real children

## How we tested it

Phase 1: 1,747 hand-verified profiles (kappa 1.000 by construction — validates logic on clear-cut cases). Phase 2: 70,192 adversarial profiles designed to break the engine (kappa 0.852, specificity 98.0%; most of the 781 disagreements were cautious under-flags). Ground-truth labels came from an AI clinical evaluator, not licensed clinicians — this is not clinical validation.

## How it compares

ASQ-3 (82–97% sens / 83–93% spec), M-CHAT-R/F (85–95% / 93–99%), PEDS (74–79% / 70–80%) — all clinically validated with real children. MyChild Engine (91.4% / 98.8%) — synthetic data only. Directional comparison, not equivalence.

## Reproduce it

```bash
npm install mychild-engine
npx mychild-engine validate --profiles data/synthetic-profiles.json --format markdown
```

---

Pages: [Home](/index.md) · [About](/about.md) · [Study](/study.md) · [Study Data](/study-data.md) · [Docs](/docs.md) · [RBSK](/rbsk.md) · [Privacy](/privacy.md) · [Terms](/terms.md) · [Contact](/contact.md)