Contradiction Reconciliation
How to handle conflicting information between sources, between Claude and Gemini outputs, and within research findings.
Overview
Contradiction Reconciliation
How to handle conflicting information between sources, between Claude and Gemini outputs, and within research findings.
Types of Contradictions
1. Factual Contradictions
Sources disagree on facts: dates, numbers, names, events.
Example: Source A says Luhmann created 90,000 cards; Source B says 70,000.
2. Interpretive Contradictions
Sources agree on facts but disagree on meaning or significance.
Example: Both sources agree handwriting activates more brain regions; they disagree on whether this matters for learning.
3. Methodological Contradictions
Different research approaches yield different conclusions.
Example: Laboratory studies show X; field studies show Y.
4. Temporal Contradictions
Information was true at one time but has changed.
Example: 2015 research showed X; 2023 research contradicts with new evidence.
5. Scope Contradictions
Something is true in one context but not another.
Example: Finding holds for students but not professionals; holds in US but not internationally.
6. Model Contradictions
Claude and Gemini return conflicting information.
Core Principle: Present, Don't Resolve
Do NOT:
- Artificially choose one side
- Smooth over genuine disagreements
- Present false consensus
- Hide contradictions from the author
DO:
- Present both positions clearly
- Show the evidence on each side
- Assess relative strength
- Let the author decide
Evaluation Framework
When sources conflict, assess:
1. Source Quality Comparison
| Factor | Source A | Source B |
|---|---|---|
| Type (primary/secondary) | ||
| Peer-reviewed? | ||
| Author credentials | ||
| Publication venue | ||
| Methodology rigor | ||
| Conflicts of interest |
Higher quality source gets more weight, but doesn't automatically "win."
2. Evidence Strength Comparison
| Factor | Source A | Source B |
|---|---|---|
| Sample size | ||
| Methodology type | ||
| Replication status | ||
| Recency | ||
| Direct vs. indirect evidence |
3. Nature of Disagreement
Is this disagreement about:
- Facts? → One is likely wrong; check sources
- Interpretation? → Both may be valid perspectives
- Scope? → Both may be right in different contexts
- Methods? → Different approaches may yield different valid results
Handling Specific Contradiction Types
Factual Contradictions
Step 1: Try to verify against primary/original sources Step 2: Check for transcription or citation errors Step 3: Look for more recent/authoritative sources Step 4: If unresolvable, present both with confidence assessment
Template for presenting:
Sources disagree on [specific fact]. Source A ([citation]) claims [X], while Source B ([citation]) claims [Y]. Source A appears more reliable because [reason]. However, this should be verified before publication.
Interpretive Contradictions
These are often legitimate differences in perspective.
Template for presenting:
Experts disagree on the significance of [finding]. [Expert A] argues [interpretation], citing [evidence]. [Expert B] counters that [alternative interpretation], based on [their evidence]. The author should consider which interpretation better serves the book's argument.
Methodological Contradictions
Different methods can yield different valid findings.
Template for presenting:
Laboratory studies (e.g., [citation]) show [X], while field research (e.g., [citation]) shows [Y]. This discrepancy may reflect [possible explanation: artificial conditions vs. real-world complexity / different populations / different measures]. Both findings may be valid in their respective contexts.
Temporal Contradictions
More recent doesn't automatically mean more correct.
Step 1: Understand why findings changed Step 2: Check if new research supersedes or just adds nuance Step 3: Consider if the change reflects real-world change or improved methods
Template for presenting:
Earlier research ([date], [citation]) found [X]. More recent work ([date], [citation]) suggests [Y]. The change appears to reflect [reason: new evidence / improved methods / changed conditions]. The author should cite [recommendation based on context].
Scope Contradictions
Both may be true in different contexts.
Template for presenting:
This finding appears context-dependent. Research shows [X] in [context A] ([citation]) but [Y] in [context B] ([citation]). The author should specify which context applies to their book's audience.
Handling Claude vs. Gemini Contradictions
When the two models return conflicting information:
Step 1: Identify the Exact Disagreement
What specifically do they disagree about? A fact? An interpretation? A source?
Step 2: Check Source Overlap
Did they cite the same or different sources?
- Same sources, different interpretations: Compare how each processed the source
- Different sources: Evaluate source quality on each side
Step 3: Look for Verification Flags
Did either model mark claims as [Retrieved] vs. [Training]?
- Retrieved information is generally more reliable
- Training knowledge may include hallucinations
Step 4: Attempt Third-Party Verification
For critical claims, verify independently:
- Search for the cited source directly
- Check if the claim appears in the source
- Look for additional corroboration
Step 5: Present the Conflict
If unresolvable through verification:
Template:
Claude's research indicates [X], citing [sources]. Gemini's research indicates [Y], citing [sources]. Key differences:
- [Difference 1]
- [Difference 2]
Assessment: [Which seems more reliable and why, or acknowledgment of genuine uncertainty]
Recommendation: [What the author should do—additional research, choose based on criteria, acknowledge uncertainty in the book]
When to Flag for Author Decision
Escalate to the author when:
- Both positions are defensible — No clear winner on evidence
- Contradiction affects thesis — The book's argument depends on which is true
- P1 evidence is contested — Critical claim has conflicting evidence
- New information challenges assumptions — Research contradicts what the book planned to argue
- Expert consensus doesn't exist — Genuine ongoing debate among authorities
Template for flagging:
⚠️ Author Decision Required
Contradiction: [Describe the conflict]
Position A: [Statement with evidence and sources] Position B: [Statement with evidence and sources]
My Assessment: [Your analysis of relative strength]
Impact on Book: [How this affects the chapter/thesis]
Options:
- [Option with implications]
- [Option with implications]
- [Option with implications]
Recommendation: [Your suggested path forward]
Documenting Contradictions
In the Chapter Research Summary, document significant contradictions:
Contradictions Section Format
| Topic | Position A | Position B | Resolution/Status |
|---|---|---|---|
| [Topic] | [Claim + source] | [Claim + source] | Resolved: [how] / Flagged for author / Presented as debate |
Using Contradictions in the Book
Contradictions aren't always problems—they can be opportunities:
For Credibility
Acknowledging disagreement shows intellectual honesty:
"Scholars debate whether [X or Y]..."
For Nuance
Contradictions often reveal complexity:
"In laboratory settings, [X]. In real-world conditions, [Y]. The difference suggests..."
For Engagement
Debates can be interesting:
"This remains an open question. [Expert A] argues... while [Expert B] counters..."
For Positioning
The author can take a stance:
"While some argue [X], the evidence for [Y] is stronger because..."
Quick Reference: Contradiction Response
| Contradiction Type | First Step | If Unresolved |
|---|---|---|
| Factual | Verify against primary sources | Present with confidence assessment |
| Interpretive | Check if both are valid perspectives | Present as legitimate debate |
| Methodological | Understand why methods differ | Present with context |
| Temporal | Check if later work supersedes | Present chronologically |
| Scope | Identify applicable contexts | Specify context in use |
| Model (Claude/Gemini) | Verify independently | Flag for author |
Use this guide when reviewing research that contains conflicting information, especially when Claude and Gemini outputs disagree.