jev-engineering
34-page paper and kit on typed decision systems for agent workflows, separating observation, judgement, policy, action and verification.
This repository is mostly documentation or write-up rather than runnable code.
Overview
Jev Engineering: Typed Decision Systems for Reliable Agent Workflows is a technical review and implementation framework by Av1dlive dated 19 September 2026. It proposes splitting agent workflows into observation, model judgement, deterministic policy, authorised action and outcome verification, with Jev's typed decisions as the judgement step.
The repository includes an illustrated 34-page PDF, an editable manuscript with references, eight diagrams, a dependency-free advisory claim-checking demo, an example native request, fictional practice cases and an integration prompt for coding assistants. The demo's 30 offline checks test validation and policy behaviour on synthetic responses; an optional live mode prints advisory results only. The README states it is not official, not peer reviewed and not a benchmark, that no live requests were made for the paper, and that demo thresholds are uncalibrated.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this claim supported by the evidence?(inferred) | — |
How it works
- Observe and build a state
- Ask Jev typed questions for judgement
- Apply deterministic policy and authorise any action
- Verify the outcome
Strengths
- Clear architecture separating judgement from policy and action
- Offline self-test demo with no dependencies
- Ready-made integration prompt
Caveats
- Not peer reviewed or official
- Thresholds uncalibrated; no live measurements
Getting started
`node jev-workflow-demo.mjs --self-test` (Node 20+, no install).At a glance
- Type
- Project
- Status
- research
- Language
- JavaScript
- Runtime
- Node demo script
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Jev Engineering: Typed Decision Systems for Reliable Agent Workflows. Paper, diagrams, and companion examples by Av1dlive.
logicrw/awesome-jev-projects
Frequently asked questions
What is the paper about?
A framework for using Jev's typed decisions inside agent workflows with separate policy, action and verification.
Does it report benchmark results?
No, it attributes others' figures and made no live requests.
Can I run anything?
Yes, an offline claim-checking demo with 30 self-tests.
Related projects
Small Python demos of Jev: parallel Noul ticket checks with measured speedup, and a three-way Choice citation and claim checker.
AppprototypePythonLearning lab for DeepEval with Jev as the judge: replayed evaluations, adjustable decision rules, release gates and drift alerts.
ProjectprototypePythonEducational code where a local LLM voices a game NPC and Jev classifies player messages and checks every line for spoilers.
ProjectprototypePythonNext.js site with six live Jev demos, each compared side by side with Gemini or Claude on latency, cost and agreement.
DemobetaTypeScriptTypeScript demos for testers: Jev triages CI test failures, gates test-agent commands and RAG answers, across gateway, Ollama and Foundry.
LibraryprototypeTypeScriptOfficial TypeSafe cookbook recipes for guardrails, re-ranking, RAG passage classification, citation checks and more with Jev.
Articleproduction