CCAO-F practice question 3 of 10

A senior project coordinator has been asked to use Claude to produce an end-of-year report covering finance, operations, and customer outcomes. The three areas have separate data owners and separate source documents. Which decomposition approach is most likely to produce a strong report?

Full set: 138 questions
Answer options
A: Define the report goals, gather inputs per section, draft each section in its own prompt, then synthesize the executive summary from the drafted sections. Correct answer
B: Draft the executive summary first to anchor the report's narrative, then draft each section so it supports the summary already written.
C: Draft the three sections in parallel, each using a section-specific prompt, then concatenate and lightly edit the outputs into the final report.
D: Prompt Claude once with all source documents and request the full report, including the executive summary, in a single pass.
Correct Answer:
A. Define the report goals, gather inputs per section, draft each section in its own prompt, then synthesize the executive summary from the drafted sections.
Explanation

Complex generation tasks benefit significantly from prompt decomposition. By breaking the report into distinct sections and prompting Claude separately for each using specific source documents, the user minimizes context window overload and reduces the risk of hallucinations or cross-contamination of data. Synthesizing the executive summary last ensures it accurately reflects the finalized content of the individual sections. This approach aligns with established prompt engineering best practices for large-scale document generation, ensuring high fidelity to the distinct source materials.

Why Incorrect

B. Drafting a summary before the actual content often leads to forced narratives and factual inconsistencies with the underlying data.
C. Simply concatenating parallel outputs lacks the necessary synthesis and transitional flow required for a cohesive, professional report.
D. A single-pass prompt with massive, diverse context increases the likelihood of hallucinations, omitted details, and degraded output quality.

References

1. Anthropic Prompt Engineering Interactive Tutorial, Section: "Task Decomposition and Chaining Prompts".

2. "Prompt Programming for Large Language Models: Beyond the Few-Shot Paradigm" (Reynolds & McDonell, 2021), Section on task chaining. https://doi.org/10.48550/arXiv.2102.07350

3. Stanford University CS224N (Natural Language Processing with Deep Learning), Lecture notes on LLM Context Windows and Attention Mechanisms.

Scroll to Top