Research Methodology Skill

This skill provides knowledge about the 5-phase research funnel methodology for systematic literature review and technical research.

When to Use This Skill

Use this skill when:

• •Conducting systematic literature reviews
• •Researching technical topics or academic papers
• •Needing to understand and compare multiple approaches to a problem
• •Creating research proposals or technical recommendations
• •The user mentions "research," "literature review," "paper analysis," or "technical survey"

The 5-Phase Research Funnel

Overview

The research funnel is a systematic methodology for transforming vague problems into actionable technical proposals:

Phase 1: DEFINE      → Refine problem into research question
    ↓                   Output: Keywords, search queries, constraints
Phase 2: DISCOVER    → Broad sweep to find papers
    ↓                   Output: 15-30 candidate papers
Phase 3: FILTER      → 10-second scan to select
    ↓                   Output: 3-5 selected papers
Phase 4: DEEP-DIVE   → Three-pass paper analysis
    ↓                   Output: Equations, pseudo-code, insights
Phase 5: SYNTHESIZE  → Comparison matrix & proposal
                        Output: Technical recommendation

Phase Details

Phase 1: Problem Definition

Goal: Transform a vague problem into a focused, answerable research question.

Key Activities:

• •Clarify what's actually being asked
• •Define scope boundaries (in/out of scope)
• •Generate keyword taxonomy:
  • •Primary keywords (core concepts)
  • •Secondary keywords (related terms, synonyms)
  • •Academic keywords (technical terminology)
  • •Exclusion keywords (filter irrelevant results)
• •Construct optimized search queries for different databases
• •Document constraints (date range, requirements, domains)

Success Criteria: The research question should be specific enough that you'll know when you've answered it.

Phase 2: Discovery

Goal: Cast a wide net to find all potentially relevant papers and resources.

Key Activities:

• •Execute search queries across multiple databases
• •Identify 15-30 candidate papers
• •Categorize by type (academic, preprint, blog, code)
• •Assign tracking IDs (UUIDs)
• •Note papers that need manual search

Search Sources:

• •Google Scholar: Comprehensive academic coverage
• •arXiv: Cutting-edge preprints
• •Semantic Scholar: AI-optimized academic search
• •GitHub: Code implementations
• •General web: Tutorials, discussions

Success Criteria: You have enough candidates to ensure you won't miss important work.

Phase 3: Filter

Goal: Rapidly evaluate candidates and select the most valuable papers for deep analysis.

Key Activities:

• •Apply 10-second scan to each paper
• •Score on relevance, depth, novelty, clarity, reproducibility
• •Select 3-5 papers ensuring diversity
• •Document rejection reasons

10-Second Scan Checklist:

1. •Title and abstract relevant?
2. •From reputable venue?
3. •Within date range?
4. •Accessible (not paywalled)?
5. •Something unique to offer?

Success Criteria: Selected papers collectively answer the research question.

Phase 4: Deep Dive

Goal: Thoroughly understand each selected paper through systematic analysis.

The Three-Pass Method:

Pass 1 - Bird's Eye View (5-10 min):

• •Read title, abstract, intro, headings, conclusion
• •Identify paper category and main claims
• •Decide if paper should stay in selection

Pass 2 - Detailed Understanding (30-60 min):

• •Read entire paper, skip proofs/equations initially
• •Mark key figures and tables
• •Note unfamiliar terms
• •Identify strengths and weaknesses

Pass 3 - Deep Analysis (1-2 hours):

• •Understand every equation and algorithm
• •Extract implementation details
• •Identify assumptions and limitations
• •Generate pseudo-code

Key Outputs:

• •Equation explanations (LaTeX → plain English)
• •Algorithm pseudo-code
• •Reproducibility assessment
• •Key takeaways and questions

Success Criteria: You could explain the paper to someone else and implement the key ideas.

Phase 5: Synthesis

Goal: Combine insights from all phases into actionable recommendations.

Key Activities:

• •Build comparison matrix across multiple dimensions
• •Identify patterns, trade-offs, gaps, trends
• •Develop multiple approach options
• •Make recommendation with roadmap

Comparison Dimensions:

• •Performance: accuracy, speed, complexity, scalability
• •Implementation: code available, language, dependencies
• •Resources: data requirements, hardware, time
• •Maturity: citations, implementations, production use

Proposal Structure:

1. •Executive summary
2. •Background and literature landscape
3. •Multiple approach options (conservative to innovative)
4. •Recommendation with justification
5. •Implementation roadmap
6. •Success metrics and risk mitigation

Success Criteria: The proposal is actionable and well-justified by the research.

Output Format

Each phase produces a Markdown file stored in ./research/<session-slug>/:

• •phase-1-define.md - Research question, keywords, search queries
• •phase-2-discover.md - Candidate papers list with metadata
• •phase-3-filter.md - Selected papers with scoring rationale
• •phase-4-deep-dive.md - Detailed analysis, equations, pseudo-code
• •phase-5-synthesize.md - Comparison matrix and technical proposal

Session Management

• •Session slug: Auto-generated from topic (URL-safe, kebab-case)
• •Session directory: ./research/<session-slug>/
• •Phase chaining: Each phase auto-detects previous phase output
• •Override: Use --session <slug> to specify session explicitly

Best Practices

For Effective Searches

• •Use Boolean operators: AND, OR, NOT
• •Use quotes for exact phrases
• •Include synonyms and related terms
• •Try both specific and broader queries
• •Check "Related Work" sections for citations

For Paper Selection

• •Balance classic/seminal works with recent innovations
• •Include papers with different approaches
• •Don't just pick highest-cited papers
• •Consider reproducibility and code availability

For Deep Analysis

• •Don't skip passes, even for familiar papers
• •Explain equations as if teaching someone else
• •Pseudo-code should be implementable
• •Note what surprises you

For Synthesis

• •Let data drive conclusions
• •Be honest about uncertainties
• •Make trade-offs explicit
• •Consider multiple audiences

Related Commands

• •/researcher:define - Start Phase 1
• •/researcher:discover - Execute Phase 2
• •/researcher:filter - Execute Phase 3
• •/researcher:deep-dive - Execute Phase 4
• •/researcher:synthesize - Execute Phase 5

References

See the reference files for detailed guidance:

• •references/search-strategies.md - Database-specific search techniques
• •references/paper-analysis.md - Detailed three-pass method
• •references/synthesis-templates.md - Templates for comparison and proposals