Nci Manipulation Analysis

For URLs

1. Use WebFetch to retrieve content from URL
2. Extract main article/post text
3. Proceed with text analysis workflow
4. Note source metadata (publication, date, author)
5. If triggers met: Use fact-checker agent to verify key claims

First Principles (Summary)

The NCI Protocol is grounded in these principles (see agents/perspective-generator.md for full version):

1. Evidence over authority - Evaluate patterns in content, not source reputation
2. Steel-man interpretation - Present strongest version of each perspective
3. Atomic decomposition - Break claims into smallest verifiable units
4. Source agnosticism - Apply identical standards regardless of source alignment
5. Bidirectional beneficiary analysis - Ask who benefits if believed AND if dismissed
6. Pattern vs. Intent - Focus on techniques; deep research evidence can inform motives

These principles ensure fair, consistent analysis across all content regardless of political or ideological alignment.

Workflow

Step 1: Input Processing

For direct text, record:

INPUT TYPE: Text
LENGTH: [word count]
CONTEXT PROVIDED: [any user context]

For URLs:

INPUT TYPE: URL
URL: [url]
Fetching content with WebFetch...
EXTRACTED: [article title, publication, date if available]

When processing URLs, also check:

• Publication reputation
• Author credentials (if available)
• Publication date and timeliness

Step 2: Score All 20 Categories

For each category, provide:

CATEGORY #[N]: [Name]
Score: [1-5]
Evidence: [Specific quotes/patterns from content]
Confidence: [LOW/MED/HIGH]

See references/categories.md for detailed category definitions and scoring criteria.

Detection signals to look for:

Step 3: Calculate Composite Factors

See references/scoring.md for weights.

COMPOSITE FACTORS:
─────────────────
Emotional Manipulation: [weighted avg of cat 1-5] → [1-5 scale]
Suspicious Timing:      [weighted avg of cat 6-8] → [1-5 scale]
Uniform Messaging:      [weighted avg of cat 9-11] → [1-5 scale]
Tribal Division:        [weighted avg of cat 12-14] → [1-5 scale]
Missing Information:    [weighted avg of cat 15-20] → [1-5 scale]

Step 4: Calculate Overall Score

OVERALL SCORE = Σ(composite_factor × weight × confidence)

Weights:
- Emotional Manipulation: 25%
- Suspicious Timing: 20%
- Uniform Messaging: 20%
- Tribal Division: 15%
- Missing Information: 20%

Scale 1-5 → 0-100: overall_score = (weighted_avg - 1) × 25

Step 5: Check Deep Research Triggers

After calculating scores, check if deep research is needed for claim verification.

Trigger Conditions (if ANY are met, proceed to verification):

DEEP RESEARCH CHECK:
─────────────────────
Overall NCI Score > 40?        [ ] Yes → Verify key claims
Suspicious Timing > 3?         [ ] Yes → Correlate events, timeline
Authority Issues (Cat 16) > 3? [ ] Yes → Verify credentials
Cherry-Picking (Cat 18) > 3?   [ ] Yes → Find omitted context
Historical Parallels > 2?      [ ] Yes → Research precedent campaigns

TRIGGERS MET: [N] → If > 0, proceed to verification

If Triggers Met:

1. Extract Key Claims: Identify 3-5 most impactful factual assertions

2. Invoke Claim Verifier: Use fact-checker agent or /decipon:verify

3. Apply Deep Research: Use ../deep-research/SKILL.md methodology

4. Track Results:

   CLAIM: [Statement]
   STATUS: [VERIFIED / PARTIALLY VERIFIED / UNVERIFIED / CONTRADICTED]
   SOURCE: [URL]
   CONFIDENCE: [1-100]
   NCI IMPACT: [How this affects scores]
   

5. Adjust Scores If Needed:

   • Verified claims → May reduce Authority Issues, Cherry-Picking scores
   • Contradicted claims → Increase relevant category scores
   • Document adjustments in final report

If No Triggers Met: Proceed directly to Step 6 (Perspective Generation).

Step 6: Generate Dual Perspectives

CRITICAL: Always generate BOTH interpretations.

MANIPULATIVE INTERPRETATION:
This content appears designed to [specific manipulation goal].
Key manipulation techniques detected:
- [Technique 1 with evidence]
- [Technique 2 with evidence]
- [Technique 3 with evidence]
Confidence: [X]%

LEGITIMATE INTERPRETATION:
This content may reflect [genuine intent/concern].
Supporting factors:
- [Factor 1]
- [Factor 2]
- [Factor 3]
Confidence: [Y]%

For perspective generation guidance, leverage the critique framework from the deep-research skill if available.

Step 7: Output Report

Standard Format (Markdown):

# NCI Analysis Report

## Content Summary
[Brief description of analyzed content]

## Overall Score: [0-100] [severity indicator]
Confidence: [X]%

## Composite Factors
| Factor | Score | Confidence |
|--------|-------|------------|
| Emotional Manipulation | [X.X]/5 | [%] |
| Suspicious Timing | [X.X]/5 | [%] |
| Uniform Messaging | [X.X]/5 | [%] |
| Tribal Division | [X.X]/5 | [%] |
| Missing Information | [X.X]/5 | [%] |

## Key Findings
[Top 3-5 manipulation indicators with evidence]

## Claim Verification (if deep research triggered)
| Claim | Status | Confidence | Source |
|-------|--------|------------|--------|
| [Claim 1] | [VERIFIED/etc] | [%] | [URL] |
| [Claim 2] | [Status] | [%] | [URL] |

**Score Adjustment**: [Original] → [Adjusted] ([+/-N] due to verification)

## Perspectives
### If Manipulative
[Manipulative interpretation]

### If Legitimate
[Legitimate interpretation]

## Category Details
[Expandable section with all 20 category scores]

## Methodology
NCI Protocol v1.0 - Pattern-based manipulation detection
Deep Research: [Yes/No] - [N] claims verified

Severity Indicators (NCI Protocol v1.0):

• 0-25: [·] Low manipulation risk
• 26-50: [!] Moderate - some concerning patterns
• 51-75: [!!] High - strong manipulation patterns
• 76-100: [!!!] Severe - overwhelming manipulation signs

Integration with Deep Research

This plugin includes the deep-research skill for fact-checking and claim verification. Reference: ../deep-research/SKILL.md

Automatic Triggers

Deep research is recommended when NCI analysis shows:

Workflow Integration

NCI + DEEP RESEARCH WORKFLOW:
─────────────────────────────
1. Complete NCI analysis (Steps 1-6)
2. Check trigger conditions
3. If triggered:
   - Extract key factual claims
   - Apply claim-verifier agent
   - Use deep research methodology
   - Update scores based on findings
4. Generate final report with verification status

Using the Claim Verifier

After NCI analysis, invoke the claim-verifier agent:

• See ../agents/claim-verifier.md for verification workflow
• Uses source evaluation from ../deep-research/references/source-evaluation.md
• Applies critique framework from ../deep-research/references/critique-framework.md

Verification Commands

Source Evaluation Integration

When assessing sources during NCI analysis, apply confidence scoring:

See ../deep-research/references/source-evaluation.md for detailed scoring.

Contradiction Handling

When sources disagree during verification:

1. Note the contradiction explicitly
2. Apply confidence scoring to each source
3. Research additional sources to resolve
4. If unresolved, present both perspectives in report

See ../deep-research/references/critique-framework.md for resolution protocol.

Examples

See references/examples.md for historical case studies including:

• Nayirah Testimony (1990) - Score: 88
• Tobacco Industry Campaign - Score: 82
• Modern examples with full category breakdowns

When to Use

Use NCI Analysis:

• Content claiming urgent action needed
• Viral stories with strong emotional triggers
• Content creating clear us-vs-them dynamics
• Stories suspiciously timed with political events
• Content from unknown or questionable sources

Don't Use:

• Simple factual lookups (use fact-checking)
• Opinion pieces clearly labeled as such
• Personal correspondence
• Fiction/entertainment

References

• references/categories.md - All 20 category definitions with detection signals
• references/scoring.md - Scoring methodology, weights, and quantitative formulas
• references/examples.md - Historical case studies for calibration
• references/vocabulary.md - Detection vocabulary lists
• references/guidance.md - Actionable tips per factor

User Interaction

Use the AskUserQuestion tool at key decision points throughout the NCI analysis workflow:

When to Use

• Input ambiguity: Content type, source, or format unclear
• Score interpretation: Borderline scores requiring next-step guidance
• Deep research triggers: Confirmation before escalating to claim verification
• Perspective balance: High disagreement requiring synthesis direction
• Output preferences: Format, detail level, or focus area selection

Example Invocations

Content type clarification (Step 1):

Content could be satire, opinion, or news
→ Use AskUserQuestion tool:
  Question: "Content type affects analysis interpretation. What is this?"
  Options:
  - "News reporting (analyze as factual claims)"
  - "Opinion/editorial (note subjective framing)"
  - "Satire (analyze technique, not truth claims)"
  - "Unclear - analyze as presented"

Deep research trigger (Step 5):

NCI Score: 52, Authority Issues: 4, Cherry-Picking: 3
→ Use AskUserQuestion tool:
  Question: "Analysis triggers suggest claim verification. Proceed with deep research?"
  Options:
  - "Yes, verify key claims (recommended for score > 40)"
  - "No, pattern analysis is sufficient"
  - "Show me which claims would be verified first"

Perspective disagreement (Step 6):

Manipulative confidence: 68%, Legitimate confidence: 42%
→ Use AskUserQuestion tool:
  Question: "Perspectives diverge by 26 points (High). How to present?"
  Options:
  - "Full synthesis with dominant perspective highlighted"
  - "Present both with equal emphasis"
  - "Focus on what verification would clarify"

Output format (Step 7):

Analysis complete, ready for output
→ Use AskUserQuestion tool:
  Question: "How should I format the NCI report?"
  Options:
  - "Full markdown report with all categories"
  - "Summary with key findings only"
  - "JSON for programmatic use"
  - "Information Nutrition Label visual format"

Benefits of Interactive Analysis

• Clearer user intent: Reduces misinterpretation of ambiguous requests
• Appropriate depth: User controls triage vs deep analysis
• Informed escalation: User decides when to invest in verification
• Transparent trade-offs: Options present clear choices with implications