Research pipelines

Research pipelines sample questions

Five original sample questions with answers and explanations for the research pipelines domain of the Claude Certified Architect - Foundations exam. They are written by CyberSkill and kept separate from the mock question bank, so the answers are shown here. This is an unofficial study aid and is not affiliated with or endorsed by Anthropic.

1. A research agent must gather facts from eight independent web sources and produce one synthesis. None of the sources depend on each other. Which dispatch pattern stays fast without flooding the coordinator context?

A. Read all eight sources into the coordinator context, then write the synthesis in a single pass.
One context holding eight full sources crowds out the synthesis and risks truncation, and it serializes all the reading.
B. Dispatch eight sub-agents in parallel, each returning a short structured summary with citations, then synthesize from the summaries. (correct)
Independent work runs concurrently, and compact per-source summaries keep the coordinator context small enough to reason over all eight.
C. Process the sources one at a time in a single agent, appending each full page to the running prompt.
Sequential reading is slow, and the growing prompt of full pages is the exact context bloat to avoid.
D. Pick the two sources that look most promising and ignore the rest to save tokens.
Dropping six sources trades coverage for cost and undercuts the point of a research pass.

Why: Independent sources are a fan-out: run them in parallel and return compact, cited summaries so the coordinator can synthesize without holding every full page.

2. Two sub-agents return conflicting figures for the same metric, each with moderate confidence. Before the coordinator writes the final answer, the best move is to:

A. Average the two numbers and move on.
Averaging conflicting facts invents a figure that neither source supports.
B. Take whichever sub-agent answered first.
Order of arrival says nothing about which figure is right.
C. Run a focused check that re-fetches the metric from the primary source and resolves the conflict before synthesizing. (correct)
A targeted verification against the source of record settles the conflict with evidence rather than a guess.
D. Include both numbers in the final answer and let the reader decide.
Passing an unresolved contradiction downstream moves the problem to the user instead of solving it.

Why: When sub-agents disagree, resolve the conflict against the primary source before synthesizing, rather than averaging, guessing, or shipping the contradiction.

3. You are designing how sub-agents report findings so the final research output can be audited later. Each finding should travel with:

A. Only the claim text, to keep messages short.
A bare claim cannot be checked or trusted after the fact.
B. The claim plus a reference to its source (URL or document id and location). (correct)
Carrying the source with each claim makes the synthesis auditable and lets a reviewer or a later run verify any line.
C. The full raw page the claim came from, inline in every message.
Inlining whole pages bloats context; a reference keeps the source reachable without the bulk.
D. A confidence score and nothing else.
Confidence without a source is an opinion no one can verify.

Why: Attach a source reference to every finding so the final answer is auditable and re-checkable, without inlining whole pages.

4. A research agent keeps spawning follow-up searches and the run is not converging. The most reliable way to prevent an endless loop is to:

A. Let it continue until it naturally stops.
A loop that is not converging has no natural stop, so this burns budget indefinitely.
B. Give the task an explicit budget and a coverage check, and stop once the questions are answered or the budget is spent. (correct)
A defined budget plus a coverage test gives the run a clear, checkable stopping point.
C. Cut the run off at a random time.
An arbitrary cutoff may stop mid-finding or far too late; it is not tied to the work.
D. Add more sub-agents so it finishes sooner.
More agents widen the search and can deepen the loop rather than end it.

Why: Bound research with an explicit budget and a coverage check so the run stops when the questions are answered or the budget is reached, not whenever it happens to halt.

5. Three sub-agents searched overlapping territory and several findings repeat across their reports. Before synthesis, the coordinator should:

A. Concatenate all three reports verbatim into the final answer.
Verbatim concatenation ships the duplicates and pads the output with repetition.
B. Keep the first report and discard the other two unread.
Discarding two reports unread loses the unique findings they hold.
C. Merge the reports, collapse duplicate findings, and keep one cited instance of each. (correct)
Merging and de-duplicating preserves every distinct finding once, with its source, and drops the noise.
D. Ask the user to remove the duplicates.
De-duplication is the agent job here, not work to push back onto the user.

Why: Merge overlapping sub-agent reports and collapse duplicates to one cited instance each, keeping every distinct finding without the repetition.

Sample questions for the other domains

Extraction pipelines sample questions Customer support agents sample questions Code exploration sample questions What the research pipelines domain covers

Practice research pipelines for real

These five are a taste. The full free mock has 15 research pipelines questions among its 60, under a 120-minute timer and scored against the 720 pass line, with an explanation on every option.

Drill this domain (15 questions) Take the full free mock