Research Synthesizer Companion

An agent that combines findings from multiple research sources into a coherent, unified narrative, reconciling conflicting information, identifying cross-cutting themes, and producing synthesis documents that are more insightful than any individual source.

When to Use This Agent

Choose Research Synthesizer when:

• Combining findings from multiple researchers into a unified view
• Reconciling conflicting data from different sources
• Identifying patterns and themes across diverse research streams
• Creating coherent narratives from fragmented research findings
• Producing insight documents that connect dots between separate analyses

Consider alternatives when:

• Conducting individual research tasks (use a research analyst)
• Formatting synthesized findings into reports (use a report generator)
• Fact-checking specific claims (use a fact-checker agent)

Quick Start

Copy
# .claude/agents/research-synthesizer.yml
name: Research Synthesizer
model: claude-sonnet-4-20250514
tools:
  - Read
  - Write
  - Glob
  - Grep
prompt: |
  You are a research synthesizer. Combine findings from multiple
  sources into coherent narratives. Reconcile conflicts, identify
  patterns, and produce insights that emerge from cross-referencing
  diverse research streams. The synthesis should be greater than
  the sum of its parts.

Example invocation:

Copy
claude --agent research-synthesizer "Synthesize the three research
  reports in research/findings/ covering technical feasibility,
  organizational readiness, and financial impact of our proposed
  migration. Create a unified assessment."

Core Concepts

Synthesis Process

Collect → Map → Compare → Reconcile → Integrate → Narrate
   │        │       │          │           │           │
 Gather   Theme   Find      Resolve    Weave       Build
 all      each    overlaps  conflicts  threads     coherent
 findings finding  & gaps   & gaps     together    story

Conflict Resolution Framework

Conflict Type	Resolution Approach
Data disagreement	Examine methodologies, report range
Definitional	Clarify terms, reframe with precise definitions
Temporal	Note time difference, present chronological evolution
Scope	Identify different contexts, specify applicability
Perspective	Present both views with their reasoning

Synthesis Output Structure

Copy
## Synthesized Finding: {Title}

### Convergent Evidence
{What multiple sources agree on — strongest claims}

### Divergent Perspectives
{Where sources disagree and why}

### Emergent Insights
{New understanding from combining findings}

### Confidence Assessment
{Overall confidence based on evidence quality and agreement}

### Implications
{What this synthesized finding means for the decision at hand}

Configuration

Parameter	Description	Default
input_sources	Number of research streams to synthesize	2-5
conflict_strategy	How to handle contradictions	Analyze and document
theme_identification	Identify cross-cutting themes	true
confidence_scoring	Rate confidence per synthesis	true
output_format	Synthesis document format	Markdown
emergent_insights	Surface new insights from combination	true

Best Practices

1. Start by mapping all findings to the original research questions. Before synthesizing, create a matrix: research questions as rows, source findings as columns. This reveals which questions are well-covered, which have conflicting answers, and which have gaps. The matrix guides synthesis effort toward areas that need reconciliation rather than areas where sources already agree.

2. Treat conflicts as data, not problems to resolve. When Source A says the market is growing and Source B says it's shrinking, the interesting question is why they disagree. Different methodologies, time periods, market definitions, or geographic scopes usually explain the conflict. Documenting the disagreement and its causes provides more insight than arbitrarily choosing one position.

3. Look for emergent insights that no single source contains. The value of synthesis isn't just summarizing—it's finding connections that appear only when you hold multiple findings together. "Source A shows declining traditional IT spending" plus "Source B shows rising cloud spending" plus "Source C shows flat total IT budgets" together tell a story of technology budget reallocation that none describes individually.

4. Weight evidence by source quality and independence. Three sources citing the same original study provide less confidence than three independent studies reaching the same conclusion. Evaluate how many truly independent data points support each synthesized finding. A finding supported by multiple independent high-quality sources deserves higher confidence than one supported by many sources all tracing to a single origin.

5. Write the synthesis for the decision-maker, not the researchers. Researchers care about methodology and nuance. Decision-makers care about implications and recommended actions. The synthesis should translate research language into decision language: "This means we should..." not "The data suggests a correlation between..." Provide clear takeaways that connect synthesized findings to the business context.

Common Issues

Synthesis is just concatenation—findings from different sources listed sequentially. True synthesis requires reorganizing by theme, not by source. Instead of "Source A found X, Source B found Y," write "On the topic of cost, Sources A, B, and C converge on a range of $2-4M, with the variation explained by different assumptions about migration timeline." Theme-based organization forces genuine integration.

Conflicts are hidden or glossed over to create a false sense of consensus. Readers lose trust when they discover contradictions the synthesis didn't acknowledge. Report conflicts transparently: present both positions, explain why they differ, and state which position is better supported and why. When you can't determine which position is correct, say so. Intellectual honesty strengthens the synthesis.

Synthesis loses important nuance from individual sources. The purpose of synthesis is to elevate the signal above the noise, but sometimes the noise contains important qualifications. Preserve key caveats, limitations, and edge cases even as you integrate findings. Use the main text for the synthesized view and footnotes or appendices for nuances that specific audiences may need.