What it is
An open lab for using Jev, TypeSafe's System One decision model, correctly next to Claude Code. It documents measured bugs, a patch, a hard fallback with alerts, and where typed decisions belong.
How it uses Jev
Jev answers a typed choice question about which model tier to use for each fresh turn, given the session state. Its probabilities feed a policy with a cache guard and confidence threshold, then the local proxy selects haiku, sonnet, or opus. A health gate asks one real decision before the session opens.
Primitives:choice
Technique worth stealing
Use Jev as a decision layer with a health gate, not a chatbot; measure in transcripts and billed dollars.
Try it
Clone repo, set JEV_API_KEY, run ./install.sh jev, then use jev command.
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-centricyes0.83
- Shows a System One patternno0.37
- Handles uncertaintyunclear0.57
- Measuredno0.15
- Runnableno0.28
- Worth recommendingunclear0.55
- Model replicano0.05
- Problem scopescore on a 0–2 scale1.09
- About Jevyes0.99
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.