5.6 Information Provenance & Multi-Source Synthesis

What You Need to Know

Information provenance — knowing where every claim comes from and how confident you should be in it — is the difference between a research system that produces trustworthy outputs and one that produces plausible-sounding fiction. The exam tests your understanding of how attribution survives (or dies) through multi-agent synthesis pipelines, how to handle conflicting sources, and how temporal context prevents false contradictions.

Structured Claim-Source Mappings

Every finding in a multi-agent research system must carry its provenance. This isn’t optional metadata. It’s the structural guarantee that the final output can be traced back to specific sources. Each finding must include:

Claim: The specific assertion being made
Source URL: Where the information was found
Document name: The title of the source document
Relevant excerpt: The specific passage that supports the claim
Publication date: When the source was published or data was collected

{
  "claim": "Global renewable energy investment reached $495 billion in 2023",
  "sourceUrl": "https://example.com/iea-report-2024",
  "documentName": "IEA World Energy Investment Report 2024",
  "relevantExcerpt": "Total investment in renewable energy technologies reached approximately $495 billion in calendar year 2023, representing a 17% increase over 2022.",
  "publicationDate": "2024-06-15"
}

The critical challenge is that attribution dies during summarisation. When a synthesis agent combines findings from multiple subagents, it naturally compresses and paraphrases. Without explicit instructions to preserve claim-source mappings, the synthesis produces statements like “Investment in renewable energy has grown significantly” — no amount, no source, no date.

Downstream agents must explicitly preserve and merge claim-source mappings through synthesis. This requires:

Subagents output findings in the structured claim-source format.
The synthesis agent is instructed to maintain these mappings when combining findings.
The final output includes inline citations or a structured reference section that traces each claim to its source.

Conflict Handling

When two credible sources report different statistics for the same measure, the synthesis agent faces a critical decision. The wrong approach — and the one the exam tests for — is to arbitrarily select one value.

Example: Source A reports 12% market growth. Source B reports 8% market growth. Both are credible publications.

Wrong approach: Select the more recent source, or average the values, or pick the one from the more authoritative publisher.

Correct approach: Annotate with both values and full source attribution. Let the consumer decide.

Market growth estimates vary by source:
- **12% growth** — IEA World Energy Report (published June 2024, using 2023 calendar year data)
- **8% growth** — Bloomberg NEF Annual Review (published March 2024, using July 2022–June 2023 data)

The difference may reflect different reporting periods and methodological approaches.

This preserves the full picture. The consumer can see both values, understand the sources, and make their own judgement about which is more relevant to their needs. Arbitrarily selecting one value destroys information and presents a false certainty.

Temporal Awareness

Different publication dates explain different numbers. That’s not a contradiction. It’s temporal context, and it has to be preserved.

Consider two sources:

Source A (published 2023): reports 8% growth
Source B (published 2024): reports 12% growth

Without publication dates, these look contradictory. With dates, they tell a story: growth accelerated from 8% to 12% over the measured period. The “conflict” is actually a trend.

Require publication/data collection dates in all structured outputs. This isn’t housekeeping; it’s what makes correct interpretation possible. Without temporal context, valid trends get misread as data quality issues, and the synthesis agent may incorrectly flag or suppress findings that are actually consistent.

Subagents must include these dates in their structured outputs. The synthesis agent must preserve them through the merging process. And the final output must present them alongside the data they describe.

Content-Appropriate Rendering

Different types of content demand different presentation formats. The exam tests whether you understand that synthesis should not flatten everything into a uniform format:

Financial data → Tables. Numbers, comparisons, and trends are most readable in tabular format. Forcing financial data into prose paragraphs makes it harder to compare values and spot patterns.

Year	Investment ($B)	Growth (%)
2021	366	12%
2022	423	16%
2023	495	17%

News and current events → Prose. Narrative context, cause-and-effect relationships, and chronological developments read naturally as paragraphs.

Technical findings → Structured lists. Architectural patterns, API specifications, and configuration options are clearest as bulleted or numbered lists with clear hierarchy.

Forcing all content into a single format — all tables, or all prose, or all lists — degrades readability and comprehension. The synthesis agent should select the appropriate rendering format based on the content type.

Attribution Preservation Through Multi-Step Synthesis

In a multi-agent pipeline, attribution must survive every step:

Research subagent collects findings with claim-source mappings.
Analysis subagent evaluates findings and adds assessment, preserving original mappings.
Synthesis subagent combines findings from multiple agents, merging mappings.
Report generation produces the final output with inline citations.

At each step, there is a risk of attribution loss. The most common failure point is step 3, where the synthesis agent combines and paraphrases findings without carrying the source mappings forward. The synthesis agent’s prompt must explicitly require that every claim in its output is traceable to a specific source.

Reports should include explicit sections distinguishing well-established findings from contested ones, preserving original source characterisations and methodological context. A finding supported by three independent sources is different from a finding based on a single report, even if both are presented with equal confidence in the text.

Completing Analysis with Conflicts Intact

When document analysis encounters conflicting values, the analysis agent must complete its work with the conflicts included and explicitly annotated. It should not resolve the conflict — that decision belongs to the coordinator or the consumer.

{
  "field": "annualRevenue",
  "conflictDetected": true,
  "values": [
    {
      "value": "$4.2M",
      "source": "Annual Report 2023",
      "context": "Audited financial statements, fiscal year ending December 2023"
    },
    {
      "value": "$3.8M",
      "source": "SEC Filing Q4 2023",
      "context": "Preliminary unaudited figures, calendar year 2023"
    }
  ],
  "possibleExplanation": "Difference may reflect audited vs preliminary figures and fiscal vs calendar year reporting periods"
}

The coordinator can then decide how to handle the conflict: present both values, investigate further, or escalate to a human analyst.

Exam Traps

Practice Scenario

A multi-agent research system produces a synthesis report on market trends. Two credible sources report different growth rates: Source A reports 12% growth (2023 data) and Source B reports 8% growth (2024 data). The synthesis agent currently selects the more recent value. What is the correct approach?

A. Always use the most recent source, since the later publication date makes it the more reliable reflection of current market conditions
B. Average the two values and report 10% growth, with a footnote recording the variance between the two sources
C. Annotate both values with source attribution and publication dates, letting the consumer decide how to interpret the difference
D. Flag the conflict and escalate it to a human researcher for resolution before including either of the two figures in the final report

Answer & explanation

Correct: C

A — This silently discards valid historical context. The difference may reflect an actual trend showing growth acceleration, not a data quality issue.
B — Averaging conflicting statistics is mathematically misleading and destroys the temporal context that explains the difference between the two values.
C — This preserves provenance, temporal context, and both data points. The consumer can see that different dates explain different numbers and interpret accordingly.
D — This is unnecessary — the values are not contradictory when temporal context is considered. Proper annotation with dates resolves the apparent ambiguity.

Sources

Claude Certified Architect Foundations Exam Guide — Domain 5, Task Statement 5.6 — Anthropic
Anthropic Multi-Agent Research Patterns — Anthropic
Anthropic Prompt Engineering — Citation and Attribution — Anthropic

Exam Simulator

Five exam-style multiple-choice questions on Information Provenance & Multi-Source Synthesis. Pick an answer, then open the explanation.

Question 1

A. Always use the most recent source as it reflects the latest data
B. Annotate both values with their sources and their publication dates in full
C. Average the two values and report 10% growth with a note about source variance
D. Flag the conflict and escalate to a human researcher for resolution before including in the report

Answer & explanation

Correct: B

A is wrong: Silently discards valid historical context. The difference may reflect an actual trend (growth deceleration), not a data quality issue.
B is correct: Preserves provenance, temporal context, and both data points. The consumer can see that different dates explain different numbers and interpret accordingly.
C is wrong: Averaging conflicting statistics is mathematically misleading and destroys the temporal context that explains the difference.
D is wrong: Unnecessary escalation — the values are not contradictory when temporal context is considered. Proper annotation with dates resolves the apparent ambiguity.

Question 2

A synthesis agent combines findings from three research subagents. The final report states ‘renewable energy investment has grown significantly’ without citing any specific source or figure. What went wrong?

A. The synthesis agent paraphrased findings without preserving claim-source mappings
B. The research subagents did not find specific figures
C. The report is appropriately summarised for executive readability
D. The synthesis agent correctly abstracted away unnecessary detail

Answer & explanation

Correct: A

A is correct: Attribution dies during summarisation. The synthesis agent compressed findings without preserving the structured claim-source mappings. The original might have been ‘$495 billion in 2023 (IEA World Energy Report 2024)’ which became ‘has grown significantly’.
B is wrong: The question implies the subagents produced detailed findings. The problem is at the synthesis step.
C is wrong: Removing specific figures and sources is not summarisation — it is information destruction. Executive summaries should still cite sources and include key figures.
D is wrong: Dropping attribution and figures is not abstraction. The synthesis agent must explicitly preserve claim-source mappings through the merging process.

Question 3

A multi-agent research pipeline has four steps: (1) research subagent collects findings, (2) analysis subagent evaluates findings, (3) synthesis subagent combines findings, (4) report generation. At which step is attribution most commonly lost?

A. Step 1 — research subagents often fail to record sources
B. Step 2 — analysis subagents strip source information during evaluation
C. Step 4 — report generation formats remove inline citations
D. Step 3 — synthesis subagents combine and paraphrase without carrying source mappings forward

Answer & explanation

Correct: D

A is wrong: If research subagents use structured claim-source mappings, sources are captured at collection. Proper schema design addresses this.
B is wrong: Analysis subagents typically add assessment to findings, not strip them. The original mappings should pass through.
C is wrong: Report generation is mechanical formatting. If citations exist in the synthesis output, they will be formatted into the report.
D is correct: Step 3 is the most common failure point. The synthesis agent naturally compresses and paraphrases, and without explicit instructions to preserve claim-source mappings, attribution is lost during the merge.

Question 4

A synthesis report needs to present three types of content: quarterly investment figures by sector, a narrative about regulatory changes, and a list of API integration patterns. How should these be rendered?

A. All as prose paragraphs for consistent formatting throughout the report
B. All as structured tables for easy scanning and comparison
C. Investment figures as tables, regulatory narrative as prose, API patterns as structured lists
D. All as bullet-pointed lists for maximum readability

Answer & explanation

Correct: C

A is wrong: Forcing financial data into prose makes it harder to compare values. Forcing technical patterns into prose obscures structure.
B is wrong: Narrative regulatory changes read unnaturally as tables. Tables are for numerical comparisons, not cause-and-effect storytelling.
C is correct: Content-appropriate rendering: financial data as tables (numbers, comparisons), news/narrative as prose (cause-and-effect, chronology), technical findings as structured lists (specifications, patterns).
D is wrong: Bullet lists work for technical findings but are wrong for financial comparisons (need tables) and narrative context (need prose).

Question 5

An analysis agent encounters two documents reporting different annual revenue figures for the same company: $4.2M from the annual report and $3.8M from an SEC filing. What should the analysis agent do?

A. Complete the analysis with both of the values annotated
B. Flag the conflict and stop analysis until a human resolves it
C. Use the annual report figure as it is more authoritative
D. Average the two figures and report $4.0M as the estimated revenue

Answer & explanation

Correct: A

A is correct: Complete the analysis with conflicts intact and explicitly annotated. Include both values, their sources, and a possible explanation (audited vs preliminary, fiscal vs calendar year). The decision on how to handle the conflict belongs to the coordinator or consumer.
B is wrong: Stopping analysis is disproportionate. The conflict can be annotated and the analysis completed. The coordinator or consumer can decide how to handle it.
C is wrong: Arbitrarily selecting one value destroys information. The annual report uses audited figures; the SEC filing uses preliminary unaudited figures. Both are valid.
D is wrong: Averaging conflicting figures is mathematically misleading and destroys the source context that explains the difference.