About
JevOps.wiki is an independent, human-curated field guide to the Jev ecosystem: what is built, what works, and what the independent tests actually found. It is not affiliated with TypeSafe AI.
Why this exists
The Jev ecosystem went from zero to more than a thousand indexed projects in under two weeks, and roughly 93,000 GitHub stars. The directories track everything, which is their strength and their problem: a thousand entries is not a reading list, it is a wall. And almost nothing publishes the operational layer: which patterns transfer, which numbers are self-reported, which thresholds survived contact with real traffic, and where the documented failure modes actually bite.
This wiki is that layer. It is deliberately small: the best projects by category, the patterns with evidence behind them, the independent studies with what to do about each finding, and a getting-started path that skips the mistakes.
Method
Who maintains it
An independent builder active in the Jev community. The work behind this site is the same work offered commercially: evaluating whether Jev fits a product, designing the question sets, calibrating the thresholds, building the evaluation harness, and writing the report that tells you where the system will and will not hold.
How to get listed
The wiki is curated manually. If you are building something on Jev, send it through the submit form: name, link, and one line on what it proves. No account, no email, and it lands on the maintainer's phone for review.
What gets picked: projects that prove something, a capability, an economics result, or a failure mode worth knowing. Numbers are labeled self-reported unless a third party measured them. What does not: dead links, thin demos, and duplicates. Getting listed in the community directories (awesomejev.com, awesome-jev-usecases, awesome-typesafe-jev, jev.directory) also helps, since that is often where curation starts.
Work with me
If you are building something Jev-powered, these are the engagements that help most.
Production review
A full pass over your decision pipeline: question design, state assembly, thresholds, fallbacks, logging, and the failure modes you have not hit yet.
Evaluation harness
Labeled sets, accuracy and calibration measurement, threshold tuning, and a rerunnable suite you own, so model version bumps do not surprise you.
Question-design audit
The highest-leverage hour in this stack. Atomicity, escape hatches, criteria conflicts, option order, and the decomposition that fixes accuracy.
Build support
From prototype to production: architecture for confidence-gated flows, gate design for agents, or a second pair of eyes on a launch.
Contact
For production reviews, evaluation harnesses, and question-design audits: mozarchbtw@gmail.com. A short note about what you are building and where it hurts is enough.
Credits
- Community directories: awesomejev.com, awesome-jev-usecases, awesome-typesafe-jev, jev.directory.
- Official: TypeSafe AI, documentation, workflow evals, and the published model jaggedness pages.
- Independent studies referenced on the rigor page, by their authors: the KoBBQ calibration audit, jev-certify, jev-orderby-bench, Janus, the action-gate study, JevBench, and Jevals.com.
- Every project linked on this site belongs to its authors. Headline numbers belong to their measurements.
Last updated 30 September 2026.