Jev Suitability Test
Six questions decide whether your task is Jev-shaped (TypeSafe System One) or belongs to a different tool. Everything runs locally.
Suggested starting primitive for this task: Choice — one option wins from a fixed taxonomy; confidence gates the routing.
Judgement, not creation
Is the AI deciding something rather than writing, summarizing, or composing content?
Bounded answer space
Can every possible answer be defined in advance?
One focused judgement
Can it be expressed as a single narrow question, not a multi-step plan?
Context-contained
Can everything needed be placed in the state you send?
Fast human judgement
Could a knowledgeable expert answer in ~5 seconds given the context?
Machine-consumed result
Will software use the answer directly to branch, route, or gate?
Probably not a Jev task
Too few criteria fit. If the task needs open-ended output, research, or multi-stage reasoning, use the right tool instead: an LLM for prose, a reasoning model for analysis, deterministic code for arithmetic, a human for strategy and risk.
The decision table
| Need | Prefer |
|---|---|
| Exact calculation, lookup, known rule | Code |
| Quick semantic judgement, classification, scoring, verification | Jev |
| Complex multi-stage reasoning, research | Reasoning LLM |
| Writing prose, generating code, explanations | LLM |
| Objectives, strategy, acceptable risk | Human |
Shorthand: code calculates, Jev judges, LLMs reason and create, humans determine objectives and risk. Concept credit: TypeSafe AI's System One guidance.
About the Jev Suitability
The most common Jev mistake is sending it the wrong kind of task — one that needs generation, research, or multi-stage reasoning rather than a fast bounded judgement. TypeSafe's guidance distills into six criteria: judgement not creation, bounded answer space, atomic single question, context-contained state, fast-human judgement, and a machine-consumed result.
Answer the six for your task and get a scored verdict: 5–6 yes means excellent System One fit with a suggested primitive (Noul, Choice, or Score); 3–4 means decompose — extract the narrow judgements and hand the rest to code or a reasoning model; 0–2 means use a different tool entirely. The decision table at the bottom places Jev against code, LLMs, reasoning models, and humans.
How to use
- Describe the task you are considering.
- Answer the six criteria — the explanation under each updates.
- Read the verdict: Jev-shaped, decompose, or use another tool.
- Follow the link to the playground to prototype the Jev parts.