clarity-judge

What it is

A community-built TypeScript app that evaluates supplied writing against separate named checks—hedging, em dashes, clarity, filler, tone, passive voice, and actionability—showing each verdict, confidence, and supporting sentence instead of one opaque score.

How it uses Jev

Each enabled check becomes a separate typed question submitted in one primary batch through the app's /api/judge endpoint. Jev returns a verdict, labeled answer confidence, and a relevant sentence; results show raw probabilities in Details. A 70% threshold flags issues locally without new model calls.

Primitives:choicescore

Technique worth stealing

Run each writing concern as its own named check with its own verdict and confidence, rather than collapsing everything into a single overall score.

Try it

pnpm install --frozen-lockfile then pnpm dev; without a key the app uses labeled demo mode.

View on GitHub