Score is a structured question type that rates content against ordered descriptive levels that the user defines. It is best when the evaluator needs a graded assessment, not a binary one, and each level has a clear meaning. The model returns the chosen level, probabilities for all levels, and a confidence value.
What the score question type does
Score is a structured question type for graded evaluation. Instead of forcing a binary yes or no, it lets the user define ordered descriptive levels so the model can place content on a scale that matches the decision being made.
This makes score useful when the real need is comparative judgment, not simple classification. A rating scale can capture nuance such as weak, moderate, strong, or any other ordered set, as long as each level has a distinct meaning and a clear progression.
How score outputs should be read
The output is more than the chosen level. A well-formed score response also includes probabilities across all levels and a confidence value, which helps the caller understand how stable the judgment is and whether nearby levels were plausible alternatives.
That extra structure matters because a score is often used downstream by software or analysts who need to compare items, sort priorities, or trigger different actions based on threshold ranges. The confidence and probabilities are part of the interpretation, not decorative metadata.
When score is the right fit
Score works best when the evaluator needs a consistent rubric and the levels can be explained in advance. It is a strong choice for content review, risk grading, quality checks, and any workflow where relative degree is more useful than a binary pass or fail.
It is less useful when the task has no meaningful order, when the labels are vague, or when the evaluator cannot justify what separates one level from the next. In those cases, a score can look precise while actually hiding ambiguity.
Why score design matters
The quality of a score depends on the scale design. If the levels overlap, skip steps, or mix different concepts in the same rubric, the output becomes hard to compare and hard to trust. Good score design keeps the meaning of each level stable across use cases.
When the scale is well defined, score becomes a practical way to standardize judgment without flattening nuance. That is especially valuable in systems that need repeatable evaluation across many items, reviewers, or runs.
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
