In which scenario is a deterministic evaluation more appropriate than a model-graded one?

Cis correct —deterministic evaluationsare best suited for cases where thecorrect output is known and fixed, allowing for binary or rule-based validation.

Examples include:

Exact matches (e.g., status: “Approved”)

Regex pattern checks

Structured JSON outputs

Correct field extraction (e.g., invoice number = INV-2023-0021)

UiPath supportsdeterministic evaluationusing logic like:

"Output equals Expected"

"Contains X and Y"

"JSON schema is valid"

This is distinct frommodel-graded evaluations, which are used when outputs areopen-endedorqualitative(e.g., summarization, sentiment, tone). These require LLM-based grading to assess whether the output is “good enough” even if it varies slightly.

Option A and B refer tosubjective assessmentsbetter suited formodel-graded scoring.

D implies feedback-driven quality, again requiringflexible interpretation, not deterministic checking.

Deterministic methods offerspeed, clarity, and automationin validation — ideal for tasks where there'sonly one right answer.