CCAR-F scenarios › Multi-Agent Research System · Question 28 of 60
Scenario 3 · Multi-Agent Research System
You're building a research system on the Claude Agent SDK in which a coordinator agent delegates work to specialized subagents: one searches the web, one analyzes documents, one synthesizes findings, and one generates reports. Given a topic, the system produces a thorough report with citations.
Context Management & Reliability
Two credible sources give different figures for the same metric: an industry association reports 42% adoption and a government survey reports 31%. The document analysis subagent currently picks the figure from whichever source it considers more authoritative and passes only that value on. Reviewers say the final reports hide real disagreement. What should change?
Show answer and explanation
Answer: D — Have the document analysis subagent pass both values on, each annotated with its source and methodology, and let the coordinator decide how to reconcile them before synthesis.
Conflicting statistics from credible sources should be kept and explicitly annotated with attribution, leaving reconciliation to the coordinator instead of arbitrarily picking one (task 5.6). Averaging (a) invents a figure that neither source reported. Dropping contested metrics (b) removes important findings and hides the disagreement. A majority vote (c) is still arbitrary selection, and the synthesis agent isn't the agent that should do web searches.
More from this scenario
- The coordinator concatenates outputs from six subagents, about 60 pages, and passes them…
- Final reports often contain correct claims with missing or wrong citations. Traces show…
- The synthesis subagent's reports often include generic statements that appear in none of…
- For a broad topic, the coordinator delegates to four web search subagents covering…
- Users submit both broad questions ("How is climate policy affecting shipping costs?")…