Glossary: Choice
Choice is TypeSafe Jev’s unordered classification question: you supply a map of option keys to descriptions (or null), at most 255 options, and the model returns choice, probabilities, and confidence. Options are not ordered like a Score rubric. Jev does not emit a rationale. The map key you choose is echoed on the answer and is not used in inference — write the complete question in instructions.
This glossary entry is for builders who already heard of Jev / System One and need a citable definition before they treat Choice as one Noul per option. jev.pro is an independent unofficial field guide — not affiliated with TypeSafe AI. Confirm the primitive on docs.typesafe.ai. We do not sell or issue API keys.
How it shows up
Put type: "choice" on a key in questions for POST /v1/systemone. probabilities sum to 1. High-cardinality sets can use TypeSafe’s two-stage path (their description of jev behavior, not a latency SLA from this site). All questions share the same state (text/JSON only). Filter unused fields; jev-1.13 loses accuracy as distractors grow.
Confidence is derived from the distribution. Use a confidence threshold before you take the argmax; TypeSafe’s 0.5 / 0.9 examples are illustrations. Calibration is group-level under the RLCD vendor claim. Pin jev-1.13.0 once fitted; jev-latest can move.
Need “how much” → Score (2–10 levels, fractional score allowed). Need “is this true?” → Noul (no confidence). Need a value extracted from text → extract candidates in code, then Choice among them. Schema-safe ≠ factually correct: the model cannot invent a key outside your map, but it can pick the wrong winner. An intent Choice is a common product role, not a reserved field.
Contrast
Choice vs Score. Unordered closed set vs ordered rubric. Do not encode “how severe” as four unordered strings if you need a gradient.
Choice vs Noul. Choice is relative among the set. Separate Nouls ask whether each property is true in absolute terms — all can be low.
Choice vs intent. Intent is a job you assign to one Choice so code can route. Not every Choice is an intent.
Choice vs verdict. choice is a label. The verdict is the side effect after floors.
Common mistakes
- Treating Choice as independent Nouls that must sum to 1 across keys.
- Asking for a rationale string.
- Taking argmax with no confidence floor.
- Using Choice when you needed Score or Noul.
- Believing a closed map makes the winner true.
See also
FAQ
Is Choice the same as a Noul per option?
No. Choice is relative among the set. Separate Nouls ask whether each property is true in absolute terms — all can be low.
Does Choice emit a rationale?
No. Jev does not generate reasoning text. Log the distribution and your policy branch.
Can Choice invent an option?
Not outside your map. It can still pick the wrong winner.
One Choice per HTTP call?
No. Batch independent questions — fan-out is supported.
Limits
This is a jev.pro working definition. Choice is a TypeSafe API question type. We do not invent search volume. Schema-safe answers are not automatically factually correct. High-cardinality two-stage behavior is a vendor description, not an SLA from this site.
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.