deep-research

name	deep-research
description	Structured deep research methodology — systematic investigation with source evaluation and confidence levels

Deep Research

A systematic approach to investigating complex questions with rigor and transparency.

Research Workflow

1. Define the Question

• State the core question clearly and specifically
• Identify sub-questions that must be answered
• Define scope boundaries (time period, geography, domain)
• Set expected output format and depth

2. Initial Search (Breadth)

• Cast a wide net with 3-5 varied search queries
• Use SearXNG for aggregated results:

  curl -s "http://searxng.selfhosted.svc.cluster.local:8080/search?q=QUERY&format=json" | jq '.results[:10] | .[] | {title, url, content}'
  

• Scan titles and snippets — identify promising sources
• Note recurring themes, key authors, and organizations
• Flag contradictions early

3. Source Identification & Evaluation

• Collect 10-20 candidate sources
• Evaluate each against the source quality tiers (below)
• Prioritize primary sources over secondary
• Check publication dates — freshness matters for evolving topics

4. Deep Dive (Depth)

• Read primary sources thoroughly
• Extract key claims with supporting evidence
• Note methodology for any studies cited
• Cross-reference claims across multiple sources
• Document disagreements between sources

5. Synthesize

• Organize findings by sub-question
• Identify consensus vs. contested points
• Assess overall confidence level
• Note gaps — what couldn't be answered?

6. Cite & Deliver

• Use the output template (below)
• Include inline citations [1] with full references
• State confidence levels explicitly
• Separate facts from analysis/opinion

Source Quality Tiers

Tier 1 — Primary Sources (Highest trust)

• Original research papers (peer-reviewed)
• Official government data and reports
• Court documents and legal filings
• First-party documentation (official docs, RFCs, specs)
• Direct interviews or statements from involved parties
• Raw datasets from authoritative sources

Tier 2 — Strong Secondary Sources

• Major news outlets with editorial oversight (Reuters, AP, NYT, WSJ)
• Industry analyst reports (Gartner, Forrester — note potential bias)
• Technical publications (ACM, IEEE, USENIX)
• Established trade publications (Ars Technica, The Register)
• Books by recognized domain experts
• Reputable think tank publications

Tier 3 — Supporting Sources (Use with caution)

• Blog posts from domain experts (verify credentials)
• Conference talks and presentations
• Wikipedia (use for orientation, then follow citations)
• Community forums (Stack Overflow, Reddit — check vote counts)
• Social media posts from verified experts
• Podcasts and interviews

Do Not Cite

• Anonymous blog posts without verifiable claims
• Content farms and SEO-optimized listicles
• Social media rumors without primary source backup
• Outdated information (>2 years for fast-moving fields)
• Sources with clear undisclosed conflicts of interest

Confidence Levels

Rate each finding:

Level	Label	Meaning
★★★★★	Confirmed	Multiple Tier 1 sources agree; no credible dissent
★★★★☆	High confidence	Strong evidence from reliable sources; minor gaps
★★★☆☆	Moderate	Credible sources support it, but some uncertainty or conflicting info
★★☆☆☆	Low confidence	Limited evidence; based on Tier 2-3 sources or extrapolation
★☆☆☆☆	Speculative	Educated guess; insufficient evidence to confirm

Output Template

# Research: [Question]

**Date:** YYYY-MM-DD
**Researcher:** [Agent name]
**Confidence:** [Overall level]
**Scope:** [What's included/excluded]

## Executive Summary
[3-5 sentence overview of key findings]

## Key Findings

### Finding 1: [Title]
**Confidence:** ★★★★☆
[Detailed explanation with inline citations [1][2]]

### Finding 2: [Title]
**Confidence:** ★★★☆☆
[Detailed explanation]

## Analysis
[Your synthesis — what do the findings mean together?]

## Gaps & Limitations
- [What couldn't be determined]
- [Sources that were unavailable]
- [Areas needing further investigation]

## Recommendations
1. [Next steps or actions based on findings]

## References
1. [Author, "Title," Source, Date. URL]
2. [Author, "Title," Source, Date. URL]

Research Anti-Patterns

• Confirmation bias — Don't just search for evidence supporting your hypothesis
• Source laundering — Multiple outlets citing the same original source ≠ multiple sources
• Recency bias — Newer isn't always better; foundational works matter
• Authority bias — Experts can be wrong; check their evidence, not just credentials
• Scope creep — Stay focused on the original question; note tangents for future research

Usage

# Create a structured research output file
bash scripts/research-template.sh "What is the impact of AI on cybersecurity hiring?" "security, AI, careers"

Notes

• Time-box research: 80% of value comes from 20% of the effort
• Always state what you DON'T know — uncertainty is information
• Save raw search results and source material for reproducibility
• Update research documents when new information emerges