awesome-jev-survey
Evidence survey of Jev-style decision models: what the probabilities mean, when to act on them, where they fail, what open models release.
This repository is mostly documentation or write-up rather than runnable code.
Overview
awesome-jev-survey by Meng Luo reads the studies of TypeSafe's Jev and Jev-like typed decision models, audits open resources and traces every number to its source. Four questions guide it: what the probabilities mean, when software should act on them, where typed decisions fail, and what open implementations release.
It has an evidence map with one dot per study, findings, failure modes, an open-ecosystem audit, a searchable literature site and a survey PDF. Highlighted findings include a study where hosted Jev's AUROC dropped sharply after a name-rubric swap with zero type errors, a 37-dataset benchmark finding binary probabilities rank well but a fixed 0.5 threshold does not, and a pre-registered SCADA review gate. Numbers are as reported by authors; nothing was re-run and there is no unified leaderboard.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Other | What do Jev's probabilities mean and when should software act on them?(inferred) | — |
How it works
- Collect studies of Jev and Jev-like models
- Trace each number to its source
- Organize by meaning, action, failure and openness
- Publish map, site and PDF
Strengths
- Structured around practical questions
- Traces numbers to sources
- Searchable literature site
Caveats
- Nothing re-run; numbers as reported
At a glance
- Type
- Article
- Status
- research
- Runtime
- survey + website
- Has code
- yes
- License
- MIT (code), CC-BY-4.0 (content)
Use cases and integrations
How curated lists describe it
Awesome Jev: an evidence survey of Jev and Jev-like typed decision models…
wh000wh000/awesome-jev-liveAwesome Jev: an evidence survey of Jev and Jev-like typed decision models — calibration, selective control and open implementations, with a searchable literature site.
hellogumbo/awesome-jev
Frequently asked questions
Is this a leaderboard?
No, it is a survey; nothing was re-run.
What is a key finding on thresholds?
One cited benchmark found probabilities rank well but a fixed 0.5 threshold does not hold.
Is there a paper?
Yes, a survey PDF in the repository.
Related projects
Awesome JEV gallery: curated papers, open reproductions, benchmarks and independent evaluations for Jev and System One models.
ArticleproductionMarkdownList of research papers on Jev and the open models built in its shape, each checked against its arXiv page.
ArticleproductionCurated list of the open System One ecosystem: open models, independent benchmarks, calibration and constrained-decoding tools.
Articleproduction- cbjev★ 4
Self-hosted encoder model giving Jev-compatible choice, score and noul answers in one pass, pitched as a better-calibrated Laya successor.
ProjectresearchPython Joke decision model trained to answer wrongly with total confidence: 1.9% accuracy at 0.957 mean confidence, via a Jev-style API.
ProjectresearchPythonPre-registered field test of Jev, GLiNER2.5-Decide, Laya and other decision models against small LLMs on fresh September 2026 data.
ProjectresearchPython