Choice is a structured question type that selects one option from a defined set of up to 255 candidates. It returns the selected option, a probability for each option, and a confidence value. This makes it useful when the evaluator needs to classify an outcome into mutually exclusive categories rather than answer yes or no.
What Choice Means in Structured Evaluation
Choice is a constrained response type that asks an evaluator to select one option from a closed set. It is used when the task is classification, not free-form generation, and when each candidate is meant to be mutually exclusive.
That structure matters because the output is not just the selected value. Choice also carries per-option probabilities and an overall confidence signal, which lets downstream logic compare alternatives and understand how decisive the model was.
How Choice Differs From Open-Ended Answers
Unlike a narrative answer, Choice forces the response into a predefined taxonomy. That makes it useful for routing, labeling, triage, and other workflows where consistency is more important than explanation.
The 255-candidate ceiling is part of the contract. It keeps the option space bounded so the evaluator can produce a stable winner and an interpretable probability distribution instead of an unbounded search over possible answers.
Probabilities, Confidence, and Decision Quality
The returned probabilities show how the evaluator distributed belief across the available options, while confidence summarizes how reliable the chosen selection is. Those values are especially useful when the application needs thresholds, escalation rules, or human review for ambiguous cases.
In practice, a high-confidence Choice result is easier to automate directly, while a low-confidence result often signals that the taxonomy may be too coarse, the prompt is underspecified, or the available evidence is genuinely mixed.
Where Choice Fits in Real Systems
Choice is most valuable in structured workflows such as content tagging, issue classification, policy routing, and moderation decisions. It is less suitable when the user needs explanation, synthesis, or a result that may validly contain more than one category.
Because the set is fixed, Choice works best when the categories are well defined and operationally meaningful. If the label set is noisy or overlapping, the output can look precise while still masking ambiguity in the underlying task.
Related resources from NHI Mgmt Group
Deepen Your Knowledge
Reviewed and updated by the NHIMG editorial team on September 25, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org
