CCDV-F Question 19

Community Discussion

Written by other learners. Comments can disagree with the marked answer — check the official documentation when in doubt.

• Chose A1

  aseline establishment: Before you can look for patterns (Option C), examine traces (Option D), or test solutions (Option B), you must first isolate the exact data points that are failing and compare them directly to their successful past counterparts to isolate the variables that changed.

• Chose D1

  **D. Examine traces of the refused requests to identify what triggers the refusal, whether input patterns, system prompt content, or other context.** The reasoning is that refusals are intermittent, so the first move is to look directly at what happened in the failing cases before forming any hypothesis. A trace shows the exact system prompt, conversation history, tool results, and model output for a refused request, which lets you see the actual trigger instead of guessing at it. Why not the others first: - **A and C** are useful, but they come after you have traces in hand. Comparing refused and answered requests (A) or looking for clusters across logs (C) only works well once you know which features of a request matter, and the traces are where you learn that. Aggregate analysis without first reading individual cases tends to chase the wrong variable. - **B** is a fix, not an investigation. Changing instructions before you understand the cause risks masking the problem, breaking something else, or leaving you unable to tell whether the change actually helped. A sensible sequence is D, then A and C to see how widespread the pattern is and what distinguishes refused from answered cases, then B to test a targeted fix once you have a hypothesis.

• Chose D4

  Investigate first by examining traces of the refused requests to see what actually triggers the refusal.