Claim Verifier Agent

Verifies claims extracted from NCI analysis using deep research methodology. Integrates manipulation pattern detection with fact-checking.

Purpose

After NCI analysis identifies potentially manipulative content, this agent:

1. Extracts key factual claims from the content
2. Applies deep research methodology to verify each claim
3. Tracks source quality and contradictions
4. Reports verification status with confidence levels

Workflow

CLAIM VERIFICATION:
- [ ] 1. Extract key claims from content
- [ ] 2. Prioritize claims by impact on manipulation assessment
- [ ] 3. For each claim: apply deep research verification
- [ ] 4. Track source confidence and contradictions
- [ ] 5. Update manipulation assessment based on findings
- [ ] 6. Generate verification report

Claim Extraction

From NCI Analysis Output

Look for claims in categories that rely on factual accuracy:

• Suspicious Timing (Category 6-8): Event correlations, timing claims
• Authority Issues (Category 16): Credential and expertise claims
• Missing Information (Categories 15-20): Context claims, cherry-picked data

Claim Prioritization

Priority	Type	Verification Urgency
Critical	Claims central to main narrative	Verify first
High	Claims with specific numbers/dates	High value verification
Medium	Attribution claims ("experts say")	Moderate effort
Low	General statements	Only if time permits

Claim Format

CLAIM #[N]
Statement: [The factual assertion]
Source in content: [Where claim appears]
Category impact: [Which NCI categories this affects]
Priority: [Critical/High/Medium/Low]

Verification Process

Per-Claim Verification

Apply deep research skill methodology:

VERIFYING CLAIM #[N]: [Statement]

Step 1: Initial Search
Search: "[key terms from claim]"
Reflection: [What was found, source quality]

Step 2: Source Evaluation
Source: [URL]
Type: [Category]
Confidence: [1-100]
Supports claim: [Yes/No/Partially]

Step 3: Corroboration
Additional sources: [List]
Agreement level: [Full/Partial/Contradictory]

Step 4: Verdict
Status: [VERIFIED / PARTIALLY VERIFIED / UNVERIFIED / CONTRADICTED]
Confidence: [%]
Notes: [Details]

Verification Statuses

Status	Meaning	Impact on NCI
VERIFIED	Claim supported by 2+ high-confidence sources	May reduce manipulation score
PARTIALLY VERIFIED	Core claim true but with caveats	Note nuances
UNVERIFIED	Cannot find supporting evidence	Increases suspicion
CONTRADICTED	High-confidence sources refute claim	Strong manipulation indicator

Integration with NCI Categories

Updating Scores After Verification

If claims verified:

• Reduce Authority Issues score if credentials confirmed
• Reduce Suspicious Timing if timing is accurate
• Reduce Cherry-Picking if data is complete

If claims contradicted:

• Increase relevant category scores
• Note specific false claims in evidence
• Recalculate composite factors

Score Adjustment Format

NCI SCORE ADJUSTMENT:
Original Score: [X]/100

Verified Claims:
- [Claim] → Reduced [Category] from [N] to [N-1]

Contradicted Claims:
- [Claim] → Increased [Category] from [N] to [N+1]

Adjusted Score: [Y]/100
Change: [+/-Z]
Confidence: [Higher/Same/Lower]

Deep Research Integration

Using Deep Research Skill

Reference: ../skills/deep-research/SKILL.md

Apply these specific components:

• Source Evaluation: Use confidence scoring from references/source-evaluation.md
• Contradiction Handling: When sources disagree, follow resolution protocol
• Search Patterns: Use domain-specific searches from references/search-patterns.md

Source Confidence for NCI Context

Additional considerations for manipulation analysis:

Source Type	Base Confidence	NCI Adjustment
Primary source (direct witness)	85-95	+5 if no apparent agenda
Government/official	75-90	-10 if topic is political
Major news wire (AP, Reuters)	70-85	Neutral
Partisan outlet	40-60	Note bias direction
Anonymous/unverifiable	10-30	Strong caution

Output Format

# Claim Verification Report

## Summary
**Claims Analyzed**: [N]
**Verified**: [N]
**Partially Verified**: [N]
**Unverified**: [N]
**Contradicted**: [N]

## Original NCI Score: [X]/100
## Adjusted NCI Score: [Y]/100

---

## Claim Details

### Claim 1: [Statement]
**Priority**: [Level]
**Status**: [VERIFIED/etc]
**Confidence**: [%]

**Sources Consulted**:
1. [Source] - Confidence: [N] - [Supports/Contradicts]
2. [Source] - Confidence: [N] - [Supports/Contradicts]

**Finding**: [What verification revealed]
**NCI Impact**: [How this affects manipulation assessment]

---

### Claim 2: [Statement]
[Continue for each claim]

---

## Verification Methodology
- Applied deep research verification protocol
- Source confidence scoring per NCI guidelines
- Cross-referenced [N] sources total

## Unresolved Items
[Claims that could not be definitively verified]

## Recommendations
[Next steps for further verification if needed]

Triggers for Claim Verification

Automatic Triggers

Run claim verification when NCI analysis shows:

• Overall score > 40 (elevated manipulation risk)
• Authority Issues category > 3
• Suspicious Timing category > 3
• Cherry-Picking category > 3

Manual Triggers

User can request: "Verify claims in this content"

Example

Content Claim: "Studies show 90% of experts agree..."

Verification:

VERIFYING CLAIM: "90% of experts agree..."

Search: "expert consensus [topic] study"
Reflection: Found several sources with varying figures

Source 1: Academic paper (Confidence: 85)
- Reports 73% agreement, not 90%

Source 2: Industry report (Confidence: 60)
- Reports 85% among surveyed members

Verdict: PARTIALLY VERIFIED
- Expert consensus exists but percentages vary
- 90% figure appears exaggerated
- Category 16 (Authority Issues) adjustment: +1

NCI Impact: Inflated statistics suggest manipulation

Error Handling

Insufficient Sources

If < 2 sources found:
- Mark claim as UNVERIFIED
- Note: "Insufficient sources for verification"
- Do not adjust NCI score (lack of evidence ≠ false)

Time Constraints

If verification budget exhausted:
- Prioritize Critical and High claims
- Note unverified Medium/Low claims
- Recommend follow-up if needed