Skip to content

Jev directory

JEV core

A standard-library-only judgement primitive and harness for Jev: ask N items × K typed questions in bounded, fail-open requests, and consume answers through a closed boundary in code that refuses anything outside the set you enumerate.
Stars
1
Language
Python
Reading time
3 min
Published by
Community
Author
litshing
Last verified
Oct 2026

Highlights

  • The harness refuses any answer outside the enumerated set — the raw value is kept for audit, answer becomes None — and snapshots each consumed state.
  • Fails open: nothing raises, and a question it cannot validate is refused, not trusted; CJK input is gated as the vendor documents lower non-English accuracy.
  • Three CLI subcommands: ask (one state × K questions), score (N items × K questions, batched) and calibrate (derive a threshold from labelled scores).
  • Answer types: noul is a float in [0,1], score an int index into criteria, choice a criteria key; noul(answers, name) returns None for missing or NaN values.
  • Standard library only, with no virtualenv or dependencies; the key resolves from TYPESAFE_API_KEY or ~/.hermes/.env and is never printed, logged or written.

Quickstart

json
{
  "is_stale": {
    "type": "noul",
    "instructions": "Is this line obviously obsolete and safe to discard?",
    "criteria": {"true": "clearly obsolete", "false": "still useful"}
  }
}

Watch out

Copy reviewed against the truncated README on 2026-10-15; the use section cuts off after the answer-types paragraph. GitHub metadata lists no license, the consumers/ directory is not populated yet, and the usage-envelope spec the README references was not included in the digest.

Questions about this entry

Short answers drawn from the linked project or page.

From the Jev guides

Original write-ups on what this entry is for.

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai