jevlish

Typed Jev judgments for TypeScript business logic.

Documentation: jacobgoldfarb.github.io/Jevlish

Upgrading from an earlier release? See the changelog.

Use given to judge one value and from to filter or rank a collection. Exact checks remain TypeScript functions; semantic checks compile to Jev Noul, Score, and Choice questions.

import { given, means } from "jevlish";

const blocked = means<Ticket>("the customer cannot complete their task")
  .excluding("they can finish the task despite the inconvenience");

const decision = await given(ticket)
  .seenAs((t) => ({ subject: t.subject, body: t.body }))
  .when(blocked)
  .and((t) => t.status === "open")
  .do(escalate)
  .otherwise(leave)
  .whenUncertain(review);

npm install jevlish
export TYPESAFE_API_KEY=...

From a repository checkout, npm run example runs the complete support-triage example.

Expressions are lazy. Use .plan() to inspect the state and questions without sending them; use await or .run() to execute. Results keep uncertainty separate and carry the evidence behind each decision.

Node 20+. ESM only.

Start with the Introduction or Quickstart. The documentation covers filtering and ranking, scales, choices, batching, policy, evidence, runtime configuration, and testing.