builder / schema first
Decision Builder
Turn a plain-language decision into a Jev-ready schema.
Quick Guide: What is the Jev Schema Builder?
The Jev Schema Builder is a visual interface for constructing typed decision contracts. It guides engineers in transforming business rules into valid Jev Question Schemas (supporting Choice, Score, and Noul primitives). By formalizing prompt instructions and behavioral criteria, teams eliminate ambiguity before copying production-ready JSON into their applications.
schema / builder
Decision Builder
Turn a plain-language decision into a Jev-ready schema.
output / json
Generated schema
{
"state": "Customer: My package with tracking ID #92401 has been stuck at the transit hub for 6 days. Can I get a status update or cancel for a full refund?",
"questions": {
"routing": {
"type": "choice",
"instructions": "Route this customer support ticket to the best team.",
"criteria": {
"shipping": "Delivery delay, stuck tracking, or damaged parcel.",
"refund": "Explicit request for money back or cancellation.",
"billing": "Invoice questions, overcharge, or payment error.",
"other": "General questions or unrelated inquiry."
}
}
}
}Three Golden Rules of Decision Schema Design
Core architectural guidelines for building robust, high-precision decision criteria:
Mutually Exclusive & Collectively Exhaustive (MECE)
Guarantee candidate choices do not semantically overlap while covering all operational paths, preventing probability dispersion across ambiguous boundaries.
Objective Behavioral Criteria
Draft criteria definitions like standard operating procedures (SOPs). Ground decisions in factual signals (e.g. "requests invoice", "fatal crash logs") rather than subjective adjectives.
Mandatory Fallback Option
Always declare an `other` or `unknown` category. When messy real-world inputs diverge from expected scenarios, the model routes them cleanly into the triage bucket.
Frequently Asked Questions (FAQ)
Why are explicit criteria definitions necessary for each Choice key?
Unlike open-ended chat models that rely on semantic memory, Jev grounds its classification decisions directly in your criteria definitions. Precise criteria eliminate linguistic drift and align model probabilities with your exact workflow.
How should schemas handle ambiguous edge-case inputs?
Incorporate an explicit `other` or `review` option within the Choice map and monitor output confidence. When the top probability drops below your safety threshold (e.g., 0.80), divert the payload to manual review.
Can I bundle multiple decision questions into a single schema?
Yes. Jev accepts multiple parallel questions within a single `questions` dictionary (e.g. evaluating department routing and severity score together). The model evaluates them concurrently in a single forward pass.
How do I integrate the exported schema into production services?
Copy the compiled JSON payload and attach it to the `questions` property of your HTTP evaluation request. In TypeScript stacks, wrap the returned decision payload with zod schemas for end-to-end type safety.
How does Jev schema format compare to JSON Schema or OpenAPI?
Jev schemas are purpose-built for decision inference (System One models) rather than generic object validation, emphasizing semantic decision criteria over field regex constraints. They translate easily into standard TypeScript types.