Wellposed: lint Jev questions before trusting the answer

# wellposed

**Lint your jev requests before they come back confidently wrong.**

[](https://www.npmjs.com/package/wellposed) [](package.json) [](LICENSE)

Offline linter and agent skill for jev requests: 45 structural checks with no model call (missing
none-of-the-above options, broken state paths, wrong criteria shapes), plus jev-on-jev checks for what
structure cannot decide, with labelled corpora that score both layers.

```
  question "route_team"
    warn   Choice "route_team" has no "other"/"none of the above" option. If an input fits none of
           [billing, technical, account], jev must still pick one. On 31 real Choices given such
           inputs, 36% of the wrong answers came at confidence >= 0.9, past a confidence gate; adding
           "other" caught 90% and changed nothing on inputs that did fit.
           fix: Add e.g. {"other": "A case that fits none of the above"}

  question "agent_name"
    error  Question "agent_name" references `ticket.assigned_agent.name`, but state has nothing at
           "ticket.assigned_agent".
```

---

## What jev is

TypeSafe's jev is a model that doesn't write text. You hand it some data and a question, and it hands
back a typed answer your code can use directly — a true/false probability, a pick from a list, or a
rating. Three question types:

- **Noul** — a yes/no question. Returns "how likely is this yes," a number from 0 to 1.
- **Choice** — pick one option from a list you provide.
- **Score** — rate something on levels you define, like `can wait` → `needs attention today`.

You send it `state` (the stuff to look at) and `questions` (what to judge). It sends back answers.
No prompt engineering, no parsing JSON out of prose.

## The problem

jev is strict about **how** you write a request and completely relaxed