jevcore
Stdlib Python primitive for asking N items K typed questions in bounded, fail-open Jev requests, refusing out-of-set answers.
Overview
JEV core (litshing) is a Python library and CLI that provides a judgment primitive for TypeSafe's Jev: ask N things K typed questions each in bounded, cheap, fail-open requests. It includes a client, a cost guard, memoisation, calibration helpers and a protocol layer, all using only the standard library, with an offline test suite and one opt-in live smoke test.
Its JEV harness is a closed boundary in code around each decision. It refuses answers outside the enumerated set (a choice not among the criteria or a score index beyond the levels), keeping the raw value for audit but returning no answer; snapshots each decision's state as a fingerprint; marks every answer as unverified so the caller must verify; gates Chinese, Japanese and Korean text before sending because the vendor documents lower non-English accuracy; and fails open, never raising, refusing rather than trusting anything it cannot validate. A usage-envelope design document lists seven scenarios Jev must not be used for, the harness obligations and re-entry conditions. A publishing gate prevents files with local paths from being published.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Other | K caller-defined typed questions for each of N items | — |
How it works
- Define items and typed questions
- Client batches within cost and size bounds
- Harness validates answers against enumerated sets and fingerprints state
- Return unverified answers or refusals; caller verifies
Strengths
- Standard library only
- Refuses out-of-set answers instead of trusting them
- Documented usage envelope of forbidden scenarios
Caveats
- Blocks CJK input by design
- Personal tooling; install via symlink from a checkout
Getting started
Clone the repo, `ln -sf ~/jev-platform/bin/jev ~/.local/bin/jev`, then `jev --help`.At a glance
- Type
- Library
- Status
- prototype
- Language
- Python
- Runtime
- Library / CLI
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
A judgement primitive for TypeSafe \*\*System One / Jev\*\* — ask N things × K typed questions in bounded, cheap, fail-open requests — plus the \*\*JEV harness\*\*, the closed boundary in code that makes a Jev answer safe to consume. Standard library only.
logicrw/awesome-jev-projects
Frequently asked questions
What does the harness refuse?
Any answer outside the enumerated options or levels, keeping the raw value only for audit.
Why gate CJK text?
Because the vendor documents lower accuracy outside English.
Does it raise on errors?
No, it fails open and refuses what it cannot validate.
Related projects
Hack Club's lightweight LLM proxy, listed for adding a Jev forwarding endpoint behind its auth, limits and usage logging.
MCP serverproductionTypeScriptnpm library that rate-limits, spend-caps, caches and redacts every Jev request made in a loop.
LibrarybetaJavaScriptDeepSeek Harness Web GUI plugin giving the agent a jev_ask tool, usage tracking and an optional command guard with explicit caveats.
PluginbetaTypeScriptGo gateway that pools multiple TypeSafe accounts behind the native System One API with user keys, quotas and failover.
ServicebetaGoRust CLI for Jev that lints typed question sets offline, then sends them and prints probabilities, labels and scores as JSON.
CLIbetaRustOffline TypeScript token counter and request-cost estimator for Jev 1.13, a port of the reconstructed jevtok encoder.
PluginbetaTypeScript