jev.proField guide · curious → Pro

Open · Last updated

Jev decision templates

Templates below are unofficial starters. They are not TypeSafe defaults. Add other/not_stated. Pin the model once you tune.

This page is an independent field-guide note. jev.pro is not affiliated with TypeSafe AI. Confirm primitives on docs.typesafe.ai/primitives. We do not sell keys. This is not a TypeSafe playbook.

Open-cluster pages are independent field-guide notes. Replicas and third-party interfaces mentioned anywhere on jev.pro are not Jev and not endorsed.

Choice — intent

instructions: “Which single intent fits this ticket?”
criteria: refund / status / cancel / other — one sentence each.
Code: abort if confidence < your floor or choice == other.

Choice is an unordered set, max 255. Add other / none if the list is incomplete. Returns choice, probabilities, and confidence. Do not threshold Choice like a Noul. Prefer cascades over a 200-way blob.

If you are extracting these enums from a mega-prompt, dual-run with the migration guide before you auto-route.

Score — severity rubric

Levels 1–5 as descriptions, not “1 means 1.” Do not interpolate a dollar or minute value from probabilities.

Score accepts 2–10 ordered levels and returns a fractional expected score. Keep arithmetic in code. Do not treat score as dollars. Confidence is peakedness of the Score distribution.

Noul — screen

instructions: a yes/no you would ask a colleague. Threshold in code. No confidence field — use bands and review in the middle.

Noul is P(yes). 0.5 means uncertain, not “medium.” Compose atomic questions in code; one POST /v1/systemone can carry this Noul next to the Choice and Score above. First-party API is that POST plus GET /v1/models.

TEMPLATE_INTENT = {
    "type": "choice",
    "instructions": "Which intent fits this message?",
    "criteria": {
        "refund": "Wants money returned for a purchase.",
        "status": "Asks where an order is.",
        "other": "None of the above.",
    },
}

Fill criteria as descriptions a person could apply. Structured instructions may be an object or array; backtick field paths such as ticket.messages[0].text. Advanced JSON criteria are documented — keep them readable.

Naming

Question ids are for you. Official: they are not the model-facing wording. Put the entire question in instructions. Ids like q1 make traces unreadable — use intent, policy_ok, severity.

Name the action space in the same file as the floors. Persist a decision trace that includes the template SHA. Policy-shaped siblings: policy templates, workflow examples. Wire-up: implementation checklist. Ship: launch checklist. Scope: when to use Jev. Dump-truck and oracle shapes: anti-patterns.

What this page does not claim

FAQ

How many options?

Docs: Choice max 255. Prefer cascades over a 200-way blob.

JSON criteria?

Advanced structure is documented — keep it readable.

Can I ask Jev to average three Scores in one question?

No. Three Score questions, weights in code.

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.

Open-cluster pages are independent field-guide notes. Replicas and third-party interfaces mentioned anywhere on jev.pro are not Jev and not endorsed. Hub: Open. Siblings: policy templates, workflow examples, choice score noul. Canonical: https://docs.typesafe.ai.

Sources