CCAR-F - Context Management & Reliability - Section 5.6

Preserve information provenance and handle uncertainty in multi-source synthesis.

Source attribution is lost at summarisation steps unless claim-to-source mappings are carried through and merged by the synthesis agent. Conflicting statistics from credible sources get annotated with their attribution rather than silently resolved to one value, and publication or collection dates must travel in structured output so a temporal difference is not read as a contradiction. Reports should separate well-established findings from contested ones and render each content type in its natural form.

claim-source mappingsattribution loss during summarisationconflict annotationtemporal data in structured outputwell-established versus contested findings

Practice question for this objective

Free sampleContext Management & Reliabilityhard

In a multi-agent research system, two web search subagents return different values for the same quantity: one retrieved source puts the sector's installed capacity at 4.2 gigawatts and another puts it at 6.9 gigawatts. The synthesis subagent silently selects one value and the report presents it as settled. A quality review of 40 reports finds 17 in which a quantity that the retrieved sources disagreed about was rendered as a single uncontested figure, and two clients have since queried a number that the desk could not defend. Which change most effectively addresses this?

  • AInstruct the synthesis subagent to prefer the most recently published source whenever two retrieved figures for the same quantity differ, and to discard the older value at the point where it detects the conflict.
  • BAsk the synthesis subagent to attach a self-reported confidence score to each figure it emits, so that a reader can judge for themselves which of the two retrieved values is more likely to be the accurate one.
  • CAdd a dedicated fact-check subagent that re-runs a web search for every numeric claim in the finished draft and replaces any figure that it cannot immediately reproduce from a freshly retrieved source.
  • DExtend the synthesis contract so that when sources disagree on the same quantity the aggregated record retains each value with its own source, and the report renders the disagreement explicitly beside the claim. Correct
Annotate conflicting sources in the record and surface the disagreement, rather than letting synthesis resolve it silently into one figure. Multi-source synthesis fails safely only when disagreement survives aggregation. A pipeline that reduces two supported values to one produces output indistinguishable from output backed by unanimous evidence, so the reader cannot tell a settled figure from a contested one and the desk cannot defend either. Retaining both values with their sources makes the state of the evidence visible at the point of use.

Why A is wrong: It is tempting because recency is often a reasonable tie-break, but it still resolves the conflict invisibly and can be wrong, since the newer source may measure a different scope or restate an older estimate.

Why B is wrong: It is tempting because a score looks like an uncertainty signal, but a self-reported number is not calibrated against the evidence and it still hides the second value, which is the fact the reader actually needs.

Why C is wrong: It is tempting because verification sounds strictly safer, but it is over-engineered relative to the evidence and re-introduces the same defect, since the checker also picks one figure and discards the disagreement.

Why D is correct: Correct: the fault is that a real conflict in the evidence is collapsed before the reader sees it, so the fix is to make conflict a first-class part of the record and render it rather than resolve it silently.

See more CCAR-F practice questions, answers explained.

More in this domain

Back to all Context Management & Reliability objectives, or the CCAR-F cert hub.

Examworthy is not affiliated with or endorsed by Anthropic. Original, blueprint-aligned practice material only.