Kaushalya Kids Genius — Development Screener

A proof of concept for a parent-led developmental screening tool for children aged 0–6. A parent enters their child's details, answers around seventy short questions, and gets a report showing how each of six areas is developing, plus activities to do at home.

Built to demonstrate design, flow and experience. The milestone content is real and sourced from public-domain instruments, but it is a working placeholder — see Content and licensing.

Running it

npm install
npm run dev

No credentials, no database setup. Assessments persist to localStorage.

npm test          # 33 tests over the scoring engine and item bank
npm run build     # production build

To look at a finished report without answering seventy questions:

npx tsx scripts/seed-demo.ts

Paste the JSON it prints into localStorage under the key kaushalya.assessments.v1, then open /report/demo.

How the screening works

Three decisions carry the whole thing.

It tests three age bands, not one

Asking a 26-month-old only the 25–30 month questions tells you that something is wrong without telling you where the child actually is — which is the one thing a parent needs in order to act. So each domain is assessed across a window: the band below the child's age, the band at it, and the band above.

There are thirteen bands across 0–72 months, narrow in infancy where development moves fastest. Each band's upper bound is a CDC checkpoint age, so every item traces to a validated anchor.

It runs the basal rule to completion

If the child does not master the lowest band they were asked, we do not know where they are — the estimate is pinned to wherever we happened to start. So the assessment reaches down a band at a time until they master one, and up one band if they top out.

This matters more than it sounds. An earlier version only reached down on a near-zero score, which left a cliff: a child at 74% of the 19–24 month band scored a developmental age of 14 months, while one at 75% scored 24. Ten months on a one percent difference. tests/scoring.test.ts pins this.

It scores on a ratio, not a month gap

"Three months behind" is serious at nine months old and noise at five years. Each domain produces a developmental age, and the status comes from the ratio:

DQ = (developmental age in months ÷ age in months) × 100

DQ	Status	What it means
90+	On track	At or ahead of expectation
75–89	Emerging	A little behind; targeted play, recheck in 3 months
60–74	Needs focus	Meaningfully behind; daily activity, mention to the doctor
< 60	Worth a closer look	Suggest a developmental paediatrician

Two corrections the maths needs:

• Preterm babies are scored against corrected age up to 24 months. A baby born at 32 weeks is developmentally two months younger than their birth certificate says; skipping this misclassifies a lot of healthy children.
• Under four months, the ratio has a denominator small enough that one question swings it from 100 to 50, so the DQ is suppressed entirely and the report shows plain achievement instead.

One more rule worth knowing: a single weak domain is never averaged away. Five strong domains and one at 55 average to a comfortable 88, and that child needs attention — so the overall status is pulled down, and the report says which domain did it rather than showing a healthy number beside a cautious label.

Layout

app/                 landing → intake → assessment → report
components/          UI pieces and the hand-rolled SVG charts
content/             domains, age bands, 300 items, 108 activities  ← the data
lib/scoring.ts       the engine: pure, no I/O, fully tested
lib/narrative.ts     every sentence a parent can receive
lib/store.ts         persistence — the only file Supabase changes
supabase/migrations/ the schema this is shaped around
tests/               33 tests, incl. hand-worked example children

lib/scoring.ts is a pure function with no I/O, so it can be validated against hand-worked children before any family sees it. The example cases are typical, advanced, one weak domain, globally behind, preterm, and under four months.

Report wording is template-driven, never model-generated at runtime. This report tells some parents something frightening about their child, and every sentence they can receive should be reviewable in advance.

Content and licensing

Source	Items	Status
CDC Learn the Signs. Act Early. (2021)	157	US federal government, public domain
NIDCD/NIH hearing & communication checklist	41	US federal government, public domain
WHO motor milestone windows (2006)	3	Freely reproducible with attribution
Authored for this project	99	Vision strand, the 61–72 month band, gaps

ASQ-3 and Denver II are commercial licensed instruments. Their items are not used here and must not be pasted in. The Trivandrum Development Screening Chart (TDSC 0–6) is the closest validated Indian equivalent — 51 items, validated against Denver, 84.6% sensitivity and 90.8% specificity — and is worth licensing if Kaushalya wants an India-normed instrument. The bank is structured so its items could replace these directly.

What is placeholder and needs Kaushalya's input

• The six domains. These cover the four named in the brief (auditory, mobility, reactive, language and communication) plus vision and hand skills. If "reactive" means something specific in the Kaushalya framework, the item bank changes.
• The 61–72 month band. Past where the CDC checklists stop, authored from standard school readiness expectations. Needs review before use.
• The scoring thresholds. The 75% mastery cut and the 90/75/60 boundaries are defensible defaults from standard screening practice, but they are clinical judgements and should be confirmed by your child development lead.
• All 108 activities. Authored here as household-only, low-cost play. Replace with Kaushalya's own programme activities.

None of the above requires a code change. It is all data in content/.

Moving to Supabase

supabase/migrations/0001_init.sql is the schema this is already shaped around. The migration is:

1. Run the migration and seed content/ into the content tables.
2. Replace the six functions in lib/store.ts with Supabase queries.
3. Move scoreAssessment to a server action and persist domain_scores and results, so reports are immutable snapshots.

Two things are built in from the start because retrofitting them is painful: item text is stored as { en, hi, kn } JSON, and every assessment records the item bank version that produced it, so an old report still renders after the questions are rewritten.

Until then, reports live in one browser — a report link will not open on another device. That needs the database.

Known gaps

• No real share links or PDF file. "Save as PDF" uses the browser's print dialogue against a print stylesheet. A proper server-rendered PDF and a shareable URL both need the database.
• item_activities is unused. Activities are currently selected per domain at the child's developmental stage, which is already better than generic age advice — a four-year-old with 30-month language gets 30-month language play. Per-item mapping is the next refinement.
• No accounts or history. Progress between two reports tells a parent far more than any single report, and that is the most valuable thing to add next.
• English only, though the schema is ready for more.

---

This is a screening tool, not a diagnostic one. It is based on parent report and cannot diagnose any condition. Nothing in it should be presented to families as a diagnosis.