CCDV-F Prompt and Context Engineering Practice Question
A developer is tuning a customer-feedback classifier on the Anthropic API. The model currently mislabels sarcastic complaints as praise. The developer wants to improve accuracy using few-shot examples in the prompt. Which TWO practices should be applied? (Choose two.)
⚠ Common exam trap
The trap here is assuming that more examples or repeating one example improves classification, when the real gains come from covering the difficult cases with clearly delimited, correctly labeled examples.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Include examples that cover the difficult sarcastic cases alongside typical positive and negative examples.
Effective few-shot prompting targets the observed failure mode with representative, correctly labeled edge cases, and structures each example so the input-to-label mapping is unambiguous. Covering sarcastic cases fixes the specific boundary error, while delimiter tags prevent confusion between example text and labels. Volume, repetition, and post-target placement do not broaden the decision boundary and can introduce new biases.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Use the same example repeatedly so the model strongly memorizes the sarcasm pattern.
Why it's wrong here
Repeating one example biases the model toward that specific phrasing and can cause it to overfit, mislabeling other sarcastic complaints that differ in wording. It also wastes context budget that could cover more varied cases. Repetition does not broaden the decision boundary, so it fails to address the general sarcasm-detection problem the developer is trying to solve.
- ✗
Place all examples after the user's feedback text so the model reads the target first.
Why it's wrong here
Putting examples after the target text means the model encounters the classification task before seeing the pattern it should follow, weakening the guidance. Few-shot examples work best before the target so the model can condition on them. This ordering also risks the model treating the final example as the item to classify, producing incorrect labels.
- ✓
Include examples that cover the difficult sarcastic cases alongside typical positive and negative examples.
Why this is correct
Few-shot examples teach the model the decision boundary, so including sarcastic edge cases directly targets the observed failure mode. Examples that only show easy positives and negatives leave the model guessing on sarcasm. Covering the difficult cases, with correct labels, gives the model concrete evidence of how to classify them, which is the most direct way to reduce the mislabeling described.
- ✗
Provide as many examples as the context window allows, prioritizing quantity over diversity.
Why it's wrong here
Volume alone does not fix a boundary problem; hundreds of near-duplicate easy examples can reinforce the existing bias and waste context. Diversity and coverage of the failure mode matter more than raw count. This approach also crowds out instructions and can push relevant tokens into weaker attention positions, so it does not reliably improve sarcasm detection.
- ✓
Wrap each example in XML tags that separate the input text from its label.
Why this is correct
Structured tags such as input and label boundaries make the pattern unambiguous, so the model can distinguish example content from the classification target. This reduces confusion when example text itself contains words resembling labels. Clear delimiters are an Anthropic-recommended structuring practice and help the model generalize the input-to-label mapping rather than memorizing raw text.
About these practice questions
This CCDV-F question is part of Courseiva's 257-question bank — original exam-style content with full explanations and wrong-answer analysis, never real exam questions or exam dumps. Learn why practice questions differ from exam dumps →
JA
Written and reviewed by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
Last reviewed September 2026 · checked against the official Anthropic exam blueprint
This CCDV-F practice question is part of Courseiva's free Anthropic certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the CCDV-F exam.