Jev launch checklist
This launch checklist is a set of independent unofficial gates for putting Jev in production. TypeSafe’s models page says rate limits are dynamic; plan 429s. Confirm current ids and price on docs.typesafe.ai/models. 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. Confirm facts on docs.typesafe.ai.
Open-cluster pages are independent field-guide notes. Replicas and third-party interfaces mentioned anywhere on jev.pro are not Jev and not endorsed.
Model and cost
- Production pin is a versioned id, not only
jev-latest. - Cost model uses vendor $0.042 / MTok input (confirm) × measured tokens; no invented volume.
- Alert on
usage.input_tokensp95 and on 429 rate.
Quality
- Offline harness passed on a held-out week of tickets.
- Floors documented per action (confidence-gated routing pattern).
- Handoff queue has an SLA and a trace payload.
- Jaggedness cases in QA: math, dates, injected instructions, huge state.
Channel
- Runbook for your door (OIDC expiry, marketplace 402, CF binding).
- Fail closed: if Jev is down, safe default — not “call GPT to guess the enum.”
Trust / SEO-GEO (if you publish)
- Unofficial disclaimer + https://docs.typesafe.ai
- Perf/pricing labeled vendor claims
- Schema-safe ≠ factual correctness
- English public copy if you follow this site’s language rule
Pre-flight that is easy to skip
Before you flip the flag, a human should answer out loud:
- Pin. Production request uses a versioned id (
jev-1.13.0or current), not onlyjev-latest.response.modelis logged. - Floors. Each irreversible action has a fitted τ and an owner. Noul 0.5 is not a floor — official yes/no equally likely.
- Fail-closed. Timeout / 529 / retry exhaustion →
reviewor deny. Official SDKs retry 429/529 only. No “call GPT to guess the enum.” - Key. From TypeSafe’s console or a marketplace. Not from jev.pro. We do not sell, issue, or proxy keys.
- Gold. Held-out week, policy actions scored, door + pin + SHA dated.
- HITL. Queue has SLA and a trace payload; UI does not invent “Jev’s rationale.”
- Jaggedness. QA includes math, dates, injected instructions, huge state — expect weakness, not a hotfix prompt.
Cost model uses vendor $0.042 / MTok input (confirm models) × measured tokens. Rate limits are dynamic per TypeSafe’s models page — plan 429s. Pair with implementation checklist (wire) and design checklist (intent). Operate after launch: playbook.
When not to use this page’s advice
Do not treat this as TypeSafe’s official launch checklist (theirs is the quick start). Do not skip fail-closed because the demo was green. Do not launch the same floors on SemIf/OpenJev under a Jev banner — not Jev and not endorsed. Do not publish vendor multiples as your launch metrics. Schema-safe ≠ the first week will be correct.
See also
FAQ
Is this TypeSafe’s launch checklist?
No. Ours. Theirs is the official quick start.
Do I need GSC for an internal tool?
No. The trust bullets apply if you publish pages.
Can I launch on jev-latest?
You can. You should not once floors exist.
Will you give us a key for go-live?
No. Keys come from TypeSafe or a documented marketplace.
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.