General Research Workflow

Purpose

Execute systematic general-purpose research across history, mythology, literature, and non-ML domains using Red's (OSP) 6-phase evidence-based methodology with rigorous source evaluation and synthesis.

When to Use This Skill

Use this skill when:

• ✅ Researching historical events, mythological topics, or literary analysis
• ✅ Need to evaluate primary vs secondary sources
• ✅ Building evidence-based arguments with citations
• ✅ Topic requires source credibility analysis
• ✅ Have 6+ hours for thorough research

Do NOT use for:

• ❌ Academic ML research (use literature-synthesis instead)
• ❌ Quick fact-checking (<30 min)
• ❌ Literature reviews for academic papers (use deep-research-orchestrator)

Decision Tree: See references/decision-tree.md

Quick Reference

Step	Agent	Deliverable	Duration	Quality Gate
0	researcher	Wikipedia verification OR fallback plan	5-10 min	≥1 viable starting source
1	researcher	10+ citations from Wikipedia references	15-30 min	≥10 citations, ≥3 categories
2	researcher	20+ sources with metadata + relevance scores	1-2 hours	≥20 sources, ≥50% accessible
3	analyst	Classified sources with credibility/bias/priority scores	30-60 min	≥5 primaries, ≥80% credibility ≥3
4	researcher	Context profiles for 10+ sources, 3+ time periods	1-2 hours	≥10 contextualized, ≥3 periods
5	researcher	50+ notes, 20+ quotes with pages, 5+ cross-links	2-3 hours	All quotas met
6	coordinator	Evidence-based thesis + final report	1-2 hours	≥5 sources support thesis, validated

Agent Coordination Protocol

Sequential Execution

Each step passes deliverables to the next step. Do NOT proceed if Quality Gate fails.

Agent Roles

• researcher: Discovery, analysis, note-taking (Steps 0, 1, 2, 4, 5, Phase A-B of Step 6)
• analyst: Validation, classification, quality checks (Step 3, Phase C of Step 6)
• coordinator: Synthesis orchestration (Phase D of Step 6)

Memory MCP Tags

ALL stored data must include: WHO=[agent], WHEN=[timestamp], PROJECT=[research-topic], WHY=[intent]

Glossary

See references/glossary.md for complete definitions:

• Primary Source: Original documents/eyewitness accounts from the time period
• Secondary Source: Analysis/interpretation created after the events
• Credibility Score (1-5): Reliability based on expertise, venue, citations
• Bias Risk Score (1-5): Likelihood of systematic distortion
• WorldCat: worldcat.org - Global library catalog
• Google Scholar: scholar.google.com - Academic publication search

---

Step-by-Step Workflow

STEP 0: Pre-Flight Check (Gate 0)

Agent: researcher Goal: Verify Wikipedia article exists OR establish fallback plan

Procedure:

1. Search Wikipedia for research topic
2. IF article exists: ✅ Proceed to Step 1
3. IF NO article:
   • Try related/broader topics, alternative spellings
   • FALLBACK: Start with Google Scholar search instead
   • Extract ≥10 citations from Scholar results
   • Document: "No Wikipedia article, started with Google Scholar"
4. Check language accessibility:
   • Flag non-English sources for translation assessment
   • Document language limitation if proceeding without translations

Deliverable: Confirmation of viable starting point

Quality Gate 0: STOP if no viable sources. Escalate to user for topic clarification.

---

STEP 1: Wikipedia Mining

Agent: researcher Goal: Extract reference trail from Wikipedia

Procedure:

1. Read Wikipedia article for overview
2. Navigate to "References" section
3. Extract ALL citations with metadata:
   • ✅ Author(s) [REQUIRED]
   • ✅ Title [REQUIRED]
   • ✅ Year [REQUIRED]
   • ⚠️ ISBN/DOI [OPTIONAL]
4. Extract "Further Reading" citations
5. Categorize by type: Books, Papers, News, Websites
6. Store with Memory MCP tags

Example: See examples/wikipedia-citation-example.json

Deliverable: JSON array of 10+ citations

Quality Gate 1: STOP if <10 citations. Expand to related Wikipedia articles.

---

STEP 2: Source Discovery

Agent: researcher Goal: Locate actual sources, expand beyond Wikipedia

Procedure:

1. For EACH citation from Step 1:
   • Search WorldCat (worldcat.org)
   • Search Google Books
   • Classify as primary/secondary
   • Note accessibility: full text, preview, physical, unavailable
   • Note language
2. Expand using:
   • Google Scholar "Cited by"
   • "Related articles" suggestions
   • Author's other works
3. For each source record:
   • ✅ Title, author, year, type [REQUIRED]
   • ✅ Accessibility + language [REQUIRED]
   • ✅ Relevance note: "Addresses [question] via [primary evidence/analysis/context]" [REQUIRED]
   • ✅ Relevance score 1-5 [REQUIRED]

Deliverable: Source inventory with 20+ sources

Quality Gate 2: STOP if <20 sources OR <50% accessible. Continue discovery.

Exception: If obscure topic, proceed with ≥10 high-credibility sources, document limitation.

---

STEP 3: Source Classification

Agent: analyst (with researcher support) Goal: Classify sources using systematic rubrics

Procedure:

A. Primary vs Secondary:

• Primary: Original documents, eyewitness accounts, contemporary records, original data
• Secondary: Analysis of primaries, textbooks, biographies (written after)

B. Credibility Score (1-5) - Program-of-Thought:

Start: 3 (neutral)

ADD +1 for EACH:
✅ Peer-reviewed (academic journal, university press)
✅ Author has PhD/recognized expertise
✅ Cites primary sources, provides references
✅ Reputable institution

SUBTRACT -1 for EACH:
❌ Self-published or vanity press
❌ No author credentials
❌ No citations/references
❌ Known conflicts of interest

Final: 1-5 (capped)

C. Bias Risk Score (1-5):

Start: 2 (low bias)

ADD +1 for EACH:
⚠️ Advocacy organization affiliation
⚠️ Funding from interested party
⚠️ Strong ideological language
⚠️ Cherry-picked/one-sided presentation

Final: 1-5
(1=minimal, 3=moderate, 5=high bias)

D. Reading Priority (1-5):

Formula:
Priority = (Relevance × 0.4) + (Credibility × 0.3) +
           (Primary=+2, Secondary=0) + (Accessible=+1, Not=-1)

Bands:
5 = Read IMMEDIATELY
4 = Read soon
3 = Read if time
2 = Defer to end
1 = Skip unless critical

Flag sources:

• 💡 Priority 4-5: Immediate queue
• ⏸️ Priority 1-3: Defer
• ⚠️ Conflicting: Cross-check
• 🚨 Bias ≥4: Extra scrutiny

Example: See examples/source-classification-example.md

Deliverable: Classified inventory with scores

Quality Gate 3: STOP if <5 primaries OR <80% credibility ≥3.

Exception: If NO primaries available (ancient topics), proceed with ≥10 credibility ≥4 secondaries, document.

---

STEP 4: Contextual Analysis

Agent: researcher (with analyst verification) Goal: Understand sources in historical/cultural/scholarly context

Procedure: Start with priority 4-5 sources. For EACH:

A. Temporal Context:

• ✅ Publication date [REQUIRED]
• ✅ Major events at that time [REQUIRED]
• 💡 Time period's influence on perspective [ANALYSIS]

B. Cultural Context:

• ✅ Author's cultural/national background [REQUIRED]
• ✅ Intended audience [REQUIRED]
• 💡 Embedded cultural assumptions [ANALYSIS]

C. Historiographical Context:

• ✅ Position in scholarly debate [REQUIRED]
• ⚠️ Mainstream or controversial? [IF APPLICABLE]
• ⚠️ School of thought [IF APPLICABLE]

D. Translation Issues (if applicable):

• ⚠️ Translation reputation
• ⚠️ Known issues
• ⚠️ Multiple translations compared?

Create context profile (3-5 sentences per type)

Deliverable: Profiles for 10+ sources, 3+ time periods

Quality Gate 4: STOP if <10 contextualized OR <3 periods. Continue.

---

STEP 5: Comprehensive Note-Taking

Agent: researcher Goal: Extract claims, evidence, quotes with page numbers

Procedure: Read sources in priority order. For EACH source use template:

## SOURCE: [Title] - [Author] ([Year])
TYPE: [Primary/Secondary] | CREDIBILITY: [Score] | BIAS: [Score]

### ✅ KEY CLAIMS [REQUIRED - Min 2]
- Claim 1 (page X): "[quote or paraphrase]"
- Claim 2 (page Y): "[quote or paraphrase]"

### ✅ SUPPORTING EVIDENCE [REQUIRED]
- For Claim 1 (pages X-Y): [How supported? Data/sources/reasoning?]
- For Claim 2 (pages Y-Z): [How supported?]

### ✅ QUOTABLE PASSAGES [REQUIRED - Min 2 with pages]
- "Notable quote 1" (page X)
- "Notable quote 2" (page Y)

### ⚠️ CONTRADICTIONS [OPTIONAL - If detected]
- Conflicts with [Source B] on [point] (page X vs page Y)

### ⚠️ BIAS/AGENDA [OPTIONAL - If bias score ≥3]
- Observable patterns: [Examples]

### ⚠️ CROSS-REFERENCES [OPTIONAL - If relevant]
- Links to [Source C] on [topic]
- Supports/refutes [Source D]

Tag with: #primary-source or #secondary-source, #key-claim, #needs-verification, #high-confidence/#uncertain, #[topic-keywords]

Template: See examples/note-template.md

Deliverable: 50+ notes, 20+ quotes with pages, 5+ cross-links

Quality Gate 5: STOP if quotas not met. Re-read sources.

---

STEP 6: Synthesis & Dot Connecting

Agent: coordinator (with researcher + analyst support) Goal: Synthesize evidence into thesis and final report

Phase A - Pattern Recognition (researcher):

1. Review ALL Step 5 notes
2. Identify recurring themes (≥3 sources)
3. Find agreements (multiple sources, similar claims)
4. Find disagreements/contradictions
5. Map evidence chains: "Source A (p.42) → Source B (p.15) → Claim X"
6. Identify gaps (insufficient evidence)

Phase B - Thesis Formation (researcher):

1. CRITICAL: Let thesis EMERGE from evidence (NOT imposed)
2. Draft 1-2 sentence thesis that:
   • ✅ Makes clear argument
   • ✅ Supported by source evidence
   • ✅ Acknowledges scope/limitations
3. List supporting evidence (≥5 sources)
4. List counter-evidence
5. Identify limitations

SPECIAL CASE - Inconclusive: IF evidence too contradictory/insufficient:

• ✅ State "INCONCLUSIVE"
• 💡 Explain WHY
• 📋 Summarize competing interpretations
• 🚨 Document what sources would resolve ambiguity

Phase C - Validation (analyst):

1. Check for fallacies:
   • ❌ Circular reasoning
   • ❌ Confirmation bias
   • ❌ Unsupported leaps
2. Verify EVERY claim has citation + PAGE NUMBER
3. Flag unsupported assertions → return to researcher
4. Confirm ≥2 primaries cited (if available)
5. Assess strength:
   • ✅ STRONG: ≥5 sources, ≥2 primaries, logical
   • ⚠️ MODERATE: 3-4 sources OR 1 primary, minor gaps
   • ❌ WEAK: <3 sources OR no primaries → return

Phase D - Final Report (coordinator): Compile report using template:

# RESEARCH REPORT: [Topic]

## THESIS
[1-2 sentence evidence-based thesis]
OR [INCONCLUSIVE: Explanation]

## SUPPORTING EVIDENCE
1. [Claim 1]
   - **Source 1**: [Author, Year, p.X-Y] - [Quote/paraphrase]
   - **Source 2**: [Author, Year, p.Z] - [Quote/paraphrase]

## LIMITATIONS & CAVEATS
- **Limitation 1**: [Description]
- **Limitation 2**: [Description]

## COUNTER-EVIDENCE OR ALTERNATIVES
- **Alternative 1**: [Description] - [Source, Year, p.X]

## PRIMARY SOURCES REFERENCED [N total]
1. [Title] - [Author], [Year] - [1-sentence significance]

## SECONDARY SOURCES REFERENCED [N total]
1. [Title] - [Author], [Year] - [1-sentence contribution]

## METHODOLOGY NOTES
- Total sources: [N]
- Primaries: [N] | Secondaries: [N]
- Time periods: [List]
- Perspectives: [Regions/cultures]
- Research duration: [X hours]

## RECOMMENDATIONS FOR FURTHER RESEARCH
- [Additional sources needed?]
- [Unanswered questions?]

Example: See examples/final-report-example.md

Deliverable: Complete research report

Quality Gate 6 (FINAL): GO:

• ✅ Thesis supported by ≥5 sources OR "INCONCLUSIVE" explained
• ✅ ≥2 primaries cited (OR exception documented)
• ✅ NO unsupported claims
• ✅ ≥1 limitation acknowledged
• ✅ Analyst validation passed

NO-GO (return to Phase B):

• ❌ Unsupported claims
• ❌ Logical fallacies
• ❌ <5 sources for thesis
• ❌ No limitations (overconfident)

---

Red's Research Principles

This workflow embeds Red's (OSP) methodology:

Principle	Implementation
"Trust No One"	Step 3: Systematic credibility + bias scoring
"Context is Everything"	Step 4: Temporal/cultural/historiographical analysis
"Thesis from Evidence"	Step 6: Thesis EMERGES, not imposed. "INCONCLUSIVE" option
"Wikipedia is a Gateway"	Step 1: Mine references, not final authority. Gate 0 fallback
"Primary Sources Matter"	Step 3: ≥2 primaries required (or exception)
"Page Numbers Save Lives"	Step 5: All quotes/claims MUST have page refs

See references/red-methodology.md for full explanation.

---

Success Metrics

Quantitative

• ✅ ≥20 sources in inventory
• ✅ ≥5 primary sources (OR exception documented)
• ✅ ≥80% sources credibility ≥3
• ✅ ≥50 notes captured
• ✅ ≥20 quotes with page numbers
• ✅ ≥5 cross-source links
• ✅ Thesis supported by ≥5 sources (OR "INCONCLUSIVE")
• ✅ ≥2 primaries cited (OR exception)
• ✅ 6-10 hours duration

Qualitative

• ✅ Context explained for ≥10 sources
• ✅ Biases identified in ≥3 sources
• ✅ Thesis emerges from evidence
• ✅ All claims have citations + pages
• ✅ ≥1 limitation acknowledged
• ✅ Alternatives acknowledged
• ✅ NO logical fallacies

---

Error Handling

Failure	Gate	Resolution
No Wikipedia article	0	Google Scholar fallback
<10 citations	1	Related articles, alt terms
<20 sources	2	Different discovery methods
<50% accessible	2	Prioritize accessible, document
<5 primaries	3	Continue OR document exception
<80% credibility ≥3	3	Return to Step 2
Non-English sources	0, 2	Flag for translation OR document
Contradictory evidence	6B	"INCONCLUSIVE" option
Logical fallacies	6C	Return to Phase B
Unsupported claims	6C	Add sources OR remove claims

---

Integration

Before: Use intent-analyzer if research question vague During: Can parallel literature-synthesis for ML components After: Use academic-reading-workflow for deep reading

---

Process Visualization

See general-research-process.dot for complete workflow diagram showing all steps, gates, and decision points.

---

Red's Research Principles: "Trust No One, Context is Everything, Thesis from Evidence, Wikipedia is a Gateway, Primary Sources Matter, Page Numbers Save Lives"