What it is
A local web tool for developers evaluating TypeSafe AI's Jev model: paste a state, define typed questions, and see each answer with its full probability distribution plus the raw request/response bytes.
How it uses Jev
The Express backend calls the TypeSafe HTTP API directly with a state and a questions object. Jev returns per-question answers with probability distributions; the UI renders noul as a 0→1 probability with a 0.5 midpoint line, choice as selected option plus per-option probabilities, and score as a probability-weighted mean with tier distribution.
Primitives:choicescorenoul
Technique worth stealing
Skip the official SDK and call the HTTP API raw, so the UI shows exactly the bytes sent and received.
Try it
Node.js 20+; npm install, write TYPESAFE_API_KEY and PORT to .env.local, npm start, open localhost:5180.
Evidence
Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.
- Jev-centricyes0.96
- Shows a System One patternyes0.85
- Handles uncertaintyno0.05
- Measuredno0.05
- Runnableyes0.97
- Worth recommendingunclear0.60
- Model replicano0.03
- Problem scopescore on a 0–2 scale1.00
- About Jevyes0.98
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.