Research Validate

This checks for:

• Generic academic title patterns (hallucination indicator)
• Future publication years
• Entries with neither URL nor DOI
• Anachronistic citations

Note: URL reachability checking is optional (add --check-urls flag for thorough validation, but it's slow).

Step 3: Credibility Scoring

For each source, run the credibility evaluator:

python3 ~/.claude/skills/research/scripts/source_evaluator.py "<url>" "<title>" "<date>"

Record scores. Flag any source with score < 40 as verify — these need extra scrutiny.

Step 4: Field Coverage (Comparison Mode Only)

If the research type is comparison, validate each entity JSON against fields.yaml:

python3 ~/.claude/skills/research/scripts/validate_json.py data/<entity>.json fields.yaml

Report coverage percentage per entity. If any entity has < 70% coverage, flag it for re-research.

Step 5: Cross-Entity Consistency (Comparison Mode Only)

Read all entity JSONs and check for contradictions:

• If entity A says "supports feature X" but entity B says "only B supports feature X"
• If numerical values seem inconsistent (e.g., wildly different date formats)
• Report any contradictions found

Step 6: Write Results

Write validation.json to the output directory:

{
  "citation_check": {"total": N, "suspicious": N, "pass": true/false},
  "credibility_scores": [{"url": "...", "score": N, "recommendation": "..."}],
  "field_coverage": {"entity_name": 0.85},
  "contradictions": ["description of contradiction"],
  "overall_pass": true/false
}

Step 7: Report Summary

Tell the user:

• Number of suspicious citations
• Number of low-credibility sources
• Field coverage per entity (if comparison)
• Any contradictions found