Structured Research & Information Gathering

Capability: Systematic research and information synthesis across multiple sources with hypothesis-driven approach.

Based on: Anthropic Claude 4.5 Best Practices for Research and Information Gathering

Core Principles

Claude 4.5 demonstrates exceptional agentic search capabilities and can find and synthesize information from multiple sources effectively. This skill provides a structured framework for complex research tasks.

When to Use

•Complex research questions requiring multiple sources
•Information synthesis across large corpora
•Hypothesis validation tasks
•Competitive analysis or market research
•Technical deep-dives requiring source verification
•Any research where confidence calibration matters

Methodology

1. Define Success Criteria

Before starting research, establish:

•What constitutes a successful answer?
•What level of confidence is required?
•What sources are authoritative?
•What contradictions need resolution?

2. Structured Research Approach

<structured_research_protocol> Search for information in a structured way:

•
Develop competing hypotheses
- •Generate 2-4 initial hypotheses about the answer
- •Don't commit to one hypothesis too early
- •Track evidence for and against each
•
Track confidence levels
- •Assign confidence scores (0-100%) to each hypothesis
- •Update confidence as you gather evidence
- •Document reasoning for confidence changes
- •Improve calibration through self-critique
•
Maintain research state
- •Create/update research-notes.md with findings
- •Use hypothesis-tree.json for structured hypothesis tracking
- •Provide transparency into research progress
•
Iterative self-critique
- •Regularly critique your approach
- •Ask: "Am I searching the right places?"
- •Ask: "What assumptions am I making?"
- •Ask: "What contradictory evidence am I missing?"
•
Source verification
- •Verify information across multiple independent sources
- •Check source authority and recency
- •Note when sources conflict
- •Prefer primary sources over secondary </structured_research_protocol>

3. Research State Files

hypothesis-tree.json (Structured):

json

{
  "research_question": "What is the best approach for X?",
  "hypotheses": [
    {
      "id": 1,
      "statement": "Approach A is optimal",
      "confidence": 65,
      "evidence_for": [
        "Source 1: Performance benchmarks show 2x improvement",
        "Source 2: Industry standard in 2025"
      ],
      "evidence_against": [
        "Source 3: High complexity cost",
        "Source 4: Limited tool support"
      ],
      "sources": ["url1", "url2", "url3", "url4"]
    },
    {
      "id": 2,
      "statement": "Approach B is optimal",
      "confidence": 35,
      "evidence_for": ["..."],
      "evidence_against": ["..."],
      "sources": ["..."]
    }
  ],
  "contradictions": [
    "Source 1 vs Source 3 on performance metrics"
  ],
  "open_questions": [
    "What is the actual implementation cost?"
  ],
  "last_updated": "2026-01-04T21:30:00Z"
}

research-notes.md (Unstructured):

markdown

# Research: [Question]

## Session 1 - 04 Gen 2026

### Search Strategy
- Started with official docs
- Expanded to community forums
- Checked academic papers for validation

### Key Findings
- Finding 1: [summary] (Source: X)
- Finding 2: [summary] (Source: Y)
- Contradiction found: X says A, Y says B

### Confidence Evolution
- Initial: 50% Hypothesis 1, 50% Hypothesis 2
- After docs: 70% H1, 30% H2 (reason: benchmark data)
- After forums: 65% H1, 35% H2 (reason: implementation challenges)

### Next Steps
- Verify performance claims with independent benchmarks
- Check for recent updates (last 6 months)

Research Workflow

Phase 1: Setup (5-10%)

•Define research question clearly
•Establish success criteria
•Generate initial hypotheses (2-4)
•Identify key sources to check

Phase 2: Initial Sweep (20-30%)

•Search official documentation
•Check authoritative sources
•Build initial hypothesis confidence
•Note contradictions and gaps

Phase 3: Deep Dive (40-50%)

•Investigate contradictions
•Verify claims across sources
•Update hypothesis confidence
•Self-critique: "Am I missing something?"

Phase 4: Synthesis (20-30%)

•Resolve contradictions if possible
•Document unresolved questions
•Provide final answer with confidence level
•List key sources and reasoning

Quality Checklist

Before completing research:

• Multiple independent sources checked (minimum 3)
• Contradictions identified and investigated
• Confidence levels calibrated (not just 0% or 100%)
• Primary sources cited where possible
• Source recency verified (especially for technical topics)
• Competing hypotheses considered
• Self-critique performed: "What could I be wrong about?"
• Open questions documented
• Research state saved for future reference

Examples

Example 1: Technical Decision

Question: "Should we use REST or GraphQL for our API?"

Approach:

•
Hypotheses:
- •H1: REST is better (60% initial)
- •H2: GraphQL is better (40% initial)
•
Research areas:
- •Performance benchmarks
- •Developer experience
- •Tooling maturity
- •Our specific use case requirements
•
Sources:
- •Official docs (REST and GraphQL)
- •Independent benchmarks (2025)
- •Team experience surveys
- •Production case studies
•
Outcome:
- •Final: 70% REST, 30% GraphQL
- •Reasoning: Our use case (simple CRUD) favors REST
- •Caveat: GraphQL better if requirements change to complex querying

Example 2: Market Research

Question: "What are the top 3 competitors in space X?"

Approach:

•
Generate initial list from:
- •Industry reports
- •Market cap data
- •Social mentions
•
Validate through:
- •Customer reviews
- •Feature comparisons
- •Growth metrics
•
Cross-reference:
- •Multiple market research firms
- •Direct competitor analysis
- •Customer surveys
•
Result:
- •Top 3 with confidence levels
- •Evidence for ranking
- •Emerging competitors to watch

Anti-Patterns to Avoid

Don't:

•Commit to first hypothesis without testing alternatives
•Use single source as authoritative
•Ignore contradictory evidence
•Present 100% confidence without overwhelming evidence
•Skip self-critique step
•Leave research state undocumented

Do:

•Generate multiple competing hypotheses
•Verify across independent sources
•Update confidence iteratively
•Document reasoning and gaps
•Self-critique regularly
•Save research state for transparency

Integration with Other Skills

Works well with:

•Strategic Analysis: Use for competitive research
•Architecture: Validate technology choices
•Code Review: Research best practices
•Performance: Benchmark validation

Metrics

Track research effectiveness:

•Time to answer
•Number of sources checked
•Confidence calibration accuracy
•Contradictions resolved vs unresolved
•Open questions identified

Advanced Techniques

Hypothesis Tree Expansion

•Start with 2-4 top-level hypotheses
•Branch into sub-hypotheses as needed
•Prune low-confidence branches early

Confidence Calibration

•Track predictions vs outcomes
•Adjust confidence based on evidence strength
•Use Bayesian updating for iterative refinement

Source Authority Ranking

•Primary > Secondary > Tertiary
•Recent (2025+) > Older (2020-2024) > Ancient (<2020)
•Official > Independent > Anecdotal
•Quantitative > Qualitative (where appropriate)

Remember: The goal is not just to find AN answer, but to find the RIGHT answer with appropriate confidence and transparency about uncertainty.