Catch contract errors before spending a single API credit
Most classification integration bugs happen before the model ever runs: duplicate label names, missing Idempotency-Key headers, labels over 60 characters, or mixing top-level labels with typed questions in the same JSON body (which returns 400 ambiguous_input). Validating those constraints locally shortens integration to one clean POST request.
Switch between a flat label array (2 to 12 mutually exclusive outcomes) and typed multi-question mode (up to 4 independent choice, noul yes/no, or ordered score questions in a single request). Both modes consume exactly one decision credit on completion and refund automatically on confirmed provider failures.
Single-label vs typed multi-question schemas compared
When a workflow only needs to pick one destination queue, a flat label list keeps the payload minimal. When a single message needs orthogonal judgements — such as checking whether a post is a bug report while simultaneously scoring urgency and assigning an owner — combining them into one typed questions array avoids three separate network round-trips.
| Request mode | Domain constraint | Returned fields | Credits per call |
|---|---|---|---|
| Single labels | 2 to 12 unique strings (max 60 chars) | choice, confidence, probabilities, needsReview | 1 credit |
| Typed choice | 2 to 12 named criteria per question | answers[id].choice, confidence, needsReview | 1 credit (up to 4 Qs) |
| Typed noul | Explicit true and false definitions | answers[id].probability (0–1), needsReview | 1 credit (up to 4 Qs) |
| Typed score | 2 to 7 ordered rubric levels | answers[id].level, confidence, needsReview | 1 credit (up to 4 Qs) |
Move from local schema check to a server-side API key
Once your JSON payload validates cleanly above, test one live call in the browser playground or mint a revocable server-side key in your workspace. Prepaid credits start at $29 for 1,800 one-off credits or $39 for the 30-day Builder plan (3,500 credits), with zero auto-renewal.
Try a live decision in Playground · Compare Top-up and Builder plans
JSON decision schema questions
Does testing a schema in this builder spend credits?
No. Validation, response simulation, and curl/TypeScript code generation run deterministically in your browser with zero network calls and zero credits spent.
Can I combine choice, noul, and score in one API call?
Yes. Pass 1 to 4 entries in the questions array; Jev evaluates all of them in a single call for one credit and sets top-level needsReview to true if any answer falls below your threshold.
