CCARF Practice Question

Paper 4 · Question 24

D4 Prompt Engineering & Structured Output · Paper 4

A moderation classification tool includes a `detected_patterns` array where the model lists specific patterns it identified (e.g. 'repeated slur targeting ethnicity'). Validation checks whether the detected patterns match the assigned category, and on mismatch retries with feedback like 'You classified this as spam but detected patterns of hate speech targeting ethnicity, please re-evaluate.' What is the primary benefit of the `detected_patterns` field in this workflow?

• A.It lets the system auto-correct the category by overriding the model's classification with a rule-based pattern match
• B.It provides an auditable evidence trail showing which specific content features drove each moderation decision
• C.It lets validation detect reasoning inconsistencies and feed back targeted errors the model uses to self-correct.
• D.It increases classification accuracy by forcing the model to identify specific patterns before assigning a category

Show answer & explanations

Correct answer: C

A: The system does not override the model's classification. It feeds back the inconsistency and asks the model to re-evaluate. The model may correct its category or its patterns — the point is self-correction, not rule-based override.

B: Auditability is a secondary benefit. The primary benefit in this validation/retry workflow is enabling the system to detect and feed back reasoning inconsistencies for self-correction.

C: The detected_patterns field externalises the model's reasoning into structured data that can be programmatically validated against the conclusion. When patterns and category are inconsistent, the specific error feedback ('you detected hate speech patterns but classified as spam') gives the model actionable information to self-correct. This is the core benefit in a retry-with-error-feedback workflow.

D: While structured reasoning fields can improve accuracy, this framing misses the retry workflow context. The primary benefit is enabling programmatic validation of reasoning consistency and targeted feedback for self-correction.

Related Prompt Engineering & Structured Output questions

More: Paper 4 · All practice questions