Jev design checklist
This design checklist is an independent production gate, not a rewrite of TypeSafe’s narrative how to build page. Open-cluster pages are not official and not endorsed. Confirm facts on docs.typesafe.ai. We do not sell keys.
This page is for builders crossing from curious to production-minded Jev use. jev.pro is an independent unofficial field guide — not affiliated with TypeSafe AI. Cover intents; do not clone rival IA.
Open-cluster pages are independent field-guide notes. Replicas and third-party interfaces mentioned anywhere on jev.pro are not Jev and not endorsed. This is not a TypeSafe playbook.
Before you pick primitives
- Write the action space (including abstain) in one enum.
- Mark which actions are irreversible vs read-only (risk-scaled floors later).
- List judgments a knowledgeable person could make in a few seconds — one judgment per question.
- Move counts, dates, money compares to code.
- Decide what must stay on-prem (then Jev may be the wrong door).
If you cannot write the action space, you are not ready for Choice / Score / Noul. Choice is an unordered set (max 255; add other / none if incomplete). Score is 2–10 ordered levels with a fractional expected score — do not treat score as dollars. Noul is P(yes); 0.5 means uncertain, not “medium.” Confidence is peakedness of a Choice/Score distribution; Noul has none.
Atomic questions are composed in code. One POST /v1/systemone can carry many questions. First-party HTTP is that POST plus GET /v1/models. Do not invent extra evaluation routes.
If you are replacing a mega-prompt or a frozen classifier, read the migration guide before you lock this checklist. Shadow mode first; no big-bang irreversible auto-act.
State and questions
- Sketch the filtered
stateobject; name fields you will backtick. - Choose Choice / Score / Noul per action, not “one Score for everything.”
- Add
other/noneon every Choice that can be wrong. - Align
instructionswithcriteria(no inverted true/false). - Plan a cascade + confidence abort if the taxonomy is deep (vs classifiers).
- Generation (replies, emails) is explicitly out of scope for Jev.
Criteria are descriptions a person could apply. Structured instructions may be objects or arrays; refer to field paths with backticks. See when to use Jev if the work is generation, exact math, or a regulated attestation.
Product / GEO gates (this site’s bar)
- One primary intent per URL you will publish internally.
- Vendor latency and $ labeled as vendor claims.
- Schema-safe ≠ correct is written next to any “can’t hallucinate” quote.
- Official docs link: https://docs.typesafe.ai
- No “we issue keys” language in your console copy either.
Plan the decision trace at design time, not after the first incident. Architectures that skip traces: anti-patterns. Next: implementation checklist, then launch checklist. Hub: Open. Path overview: Jev playbook.
What this page does not claim
- Not official TypeSafe design documentation.
- Not a complete use-case map.
- Completing the boxes does not certify the system.
FAQ
Is this TypeSafe’s how-to-build page?
No. Official narrative stays on docs.typesafe.ai. This is an independent gate list.
Do I design one Score for the whole ticket?
No. One judgment per question. Weight in code.
When do I read the migration guide?
When you already have a prompt or classifier you must dual-run. Design the action space here first, then migrate.
Disclaimer
This is an independent unofficial site and is not affiliated with TypeSafe AI; official documentation is available at https://docs.typesafe.ai. Never treat jev.pro as TypeSafe official documentation. We do not sell, issue, or proxy API keys.