Clarity Judge
Writing checked on separate named axes, each with its own verdict.
# Clarity Judge Evaluate writing against **separate, named checks** with TypeSafe AI's Jev: hedging, em dash overuse, clarity, filler phrases, tone, passive voice, and actionability. Inspect each verdict, confidence signal, and supporting sentence rather than relying on one opaque overall score. This is an **independent community project**, not an official TypeSafe product. It evaluates supplied writing; it does not generate a rewrite or establish that a passage is factually correct. [Contributing](CONTRIBUTING.md) · [Agent guide](AGENTS.md) · [TypeSafe API reference](https://docs.typesafe.ai/api) ## Live demo **[judge.jev.works](https://judge.jev.works)** runs this app as a public demo. It carries **no server API key**, so every visitor sees deterministic simulated results, labeled as demo mode. Those show how the interface behaves, not Jev's measured performance, and nothing a visitor does there reaches TypeSafe. For real verdicts, open the key dialog and paste your own TypeSafe key. It is held in that browser's localStorage, travels only as a request header to the deployment's own `/api/judge` endpoint, and is never displayed again or stored by the deployment. Remove it from the same dialog when you are done, particularly on a shared machine. The deployment has no server key on purpose. A public URL configured with one lets any visitor spend its credits, so this one leaves it unset and asks each person to bring their own. ## Run it locally Use the pnpm version pinned in [package.json](package.json), currently `10.34.5`, and keep `pnpm-lock.yaml` as the only dependency lockfile. The manifest declares Node.js `>=20`; use a version supported by the installed Next.js dependency as well. Node.js 22+ is a practical development baseline. ```sh git clone https://gi