adopting-jevlisted
Install: claude install-skill shiftynick/jev-axi
# Adopting Jev in a project
Jev is a "System One" model: it answers typed questions about a piece of text with calibrated
probabilities in roughly 0.1 to 0.5 seconds, for about $0.042 per million input tokens, with
output tokens free. Three question types: **noul** (probability of yes), **choice** (one of up to
255 options), **score** (a position on a 2 to 10 level scale). All questions in one request see the
same state and run in parallel, so asking thirteen instead of one cost 12x less than thirteen
calls in TypeSafe's own test.
It never generates text, explains itself, does arithmetic, or reads images. Independent write-ups
put it at roughly comparable accuracy to frontier models on snap judgments, sometimes lower:
treat it as **cheaper and faster, not smarter**. That framing decides where it belongs.
Your job with this skill: find the places in the user's project where a judgment is being made
expensively or crudely, design the Jev version, and set up an honest comparison before anything
switches over. Don't install, rewrite, or enable anything unprompted.
## The live docs are the source of truth
This skill's prices, limits, model names, and benchmark numbers were true when it was written and
go stale. Before designing an integration or quoting a number to the user:
1. Read the index at `https://docs.typesafe.ai/llms.txt`. It lists every page.
2. Fetch pages as Markdown by adding `.md` to the path
(`https://docs.typesafe.ai/primitives/score.md`). It is smaller an