Citation Validator

📋 Reference: See .claude/shared/constants/source_quality_ratings.md for full details.

• A: Peer-reviewed RCTs, systematic reviews, meta-analyses
• B: Cohort studies, clinical guidelines, reputable analysts
• C: Expert opinion, case reports, company white papers
• D: Preprints, conference abstracts, blogs
• E: Anonymous, biased, outdated, broken links

Safety Limits

• Max claims to validate: 200 per session
• Timeout per URL check: 5 seconds (reduced from 10s)
• Max parallel URL checks: 5
• Cache validated URLs: 7 days TTL

Token Optimization

📋 Reference: .claude/shared/constants/token_optimization.md

Batch Validation: Process 20 citations at a time

URL Checking: Cache results for 7 days in data/citation_cache.json

Skip Re-validation: If URL checked within 24 hours

Context Budget: 30k tokens max

Error Handling

📋 Reference: .claude/shared/constants/error_codes.md

Common Errors:

• E102: URL not accessible (404) → Search for archived version
• E401: Hallucination detected → Remove claim or find citation (penalty: -2 points)
• E402: Source quality too low → Search for higher-quality sources

Hallucination Penalty: -2 points per occurrence (reduced from -5)

Validation Process

1. Claim Detection: Identify all factual claims
2. Citation Presence: Check each claim has citation
3. Completeness Check: Verify all required elements
4. Quality Assessment: Assign A-E rating
5. Accuracy Verification: Use WebSearch/WebFetch to verify
6. Hallucination Detection: Flag unsupported claims
7. Chain-of-Verification: Extra scrutiny for critical claims

Quality Score

Target: ≥ 8/10

• Citation coverage (0-3 pts)
• Completeness (0-2 pts)
• Accuracy (0-3 pts)
• Source quality (0-2 pts)
• Hallucination penalty (-5 pts each)

Examples

See examples.md for validation scenarios.

Detailed Instructions

See instructions.md for complete validation methodology.