jev-samples
Runnable Jev samples: README checks, AGENTS.md readiness, PR triage, MovieLens next-watch ranking and issue triage, with measured runs.
Overview
jev-samples is a set of small, self-contained runnable projects showing how to use Jev: you send state and questions that carry their own answer options, and branch on the typed value that comes back. readme-check asks five questions about a README using all three primitives and explains every number. agents-md-readiness scores an AGENTS.md or CLAUDE.md against the agents.md format with seventeen questions and a weighted readiness score, shipping scored runs for eleven repositories and a Go port for CI.
pr-triage decides what evidence each open pull request needs before merging: rules settle drafts, failing branches and sent-back PRs for free, and the rest get one call of twelve questions whose answers become effort and consequence scores that pick one of five routes; on eighteen Airflow PRs, rules settled ten and the other eight cost $0.001. It ships a Go port and a Claude Code skill. next-watch ranks MovieLens titles by putting twenty candidates in one Choice; on 100 members Jev put the hidden title first 25.5% of the time versus 13.2% for genre matching, for $0.010 total. issue-triage covers open issues.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which of these twenty titles will this member watch next? | — |
| Score | How much effort and consequence does this pull request carry?(inferred) | — |
| Yes / no | Does this AGENTS.md cover this element of the agents.md format?(inferred) | yes · no |
How it works
- Pick a sample and point it at your own files or repository
- Rules settle easy cases at no cost
- One Jev call asks the sample's questions
- Code turns answers into routes, scores or rankings
- Compare with baselines where provided
Strengths
- Realistic, runnable examples with measured results
- Rules-first design keeps costs low
- Go ports and a Claude Code skill
Caveats
- Samples, not maintained products
- Small evaluation sets
- Requires a TypeSafe key
Getting started
See each sample's folder under samples/ for run instructions.At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- Sample scripts (some Go ports)
- Has code
- yes
- Stars
- 1
Use cases and integrations
How curated lists describe it
Runnable samples for Jev, TypeSafe AI's System One model. Send state and questions carrying their own answer options, then branch on the typed value that comes back.
wh000wh000/awesome-jev-liveRunnable samples for Jev, TypeSafe AI's System One model. Send state and questions carrying their own answer options, then branch on the typed value that comes back.
hellogumbo/awesome-jev
Frequently asked questions
What kinds of samples are included?
README and AGENTS.md checks, PR triage, issue triage and a MovieLens next-watch ranker.
How cheap is PR triage?
On 18 Airflow PRs, rules handled 10 for free and the other 8 cost $0.001.
How good is the ranking sample?
Jev ranked the hidden title first 25.5% of the time versus 13.2% for genre matching.
Related projects
Independent 52-page Chinese report on Jev that checks vendor claims and reproduces 50 Chinese tests, with traceable data.
ProjectresearchPythonBenchmark of Jev on 159 community-reconstructed questions from Indonesia's SNBT 2025 university entrance exam, all responses published.
BenchmarkresearchPythonLocal SvelteKit workbench and CLI for organizing, editing, running and comparing Jev experiments defined as JSON.
CLIbetaTypeScript'Vitest for LLM outputs': CLI running YAML/JSON eval suites on Jev for RAG faithfulness, hallucination and agent safety in CI.
CLIprototypeTypeScriptLearning lab for DeepEval with Jev as the judge: replayed evaluations, adjustable decision rules, release gates and drift alerts.
ProjectprototypePythonJev-compatible open System One engine with a Claude Code-style terminal REPL showing live probabilities, plus an experiment harness.
ProjectprototypePython