fact-verify

Fact Verification

Verify factual claims in security documents against authoritative sources. Uses Perplexity MCP when available; falls back to direct web queries when it is not.

Process

Step 1: Detect Claim Type

• CVE verification: if CVE-ID present or claim_type=cve
• Event investigation: if investigation document or claim_type=event

Step 2: Extract Verifiable Claims

CVE claims:

• CVSS score (verify against NVD)
• EPSS score (verify against FIRST API)
• KEV status (verify against CISA catalog)
• Affected versions (verify against vendor advisory)
• Patched versions (verify against vendor advisory)
• ATT&CK technique IDs (verify against attack.mitre.org)

Event claims:

• IP ownership/ASN (verify via WHOIS/BGP)
• Geolocation claims (verify via IP geolocation services)
• Threat intelligence associations (verify via threat feeds)
• Protocol/port combinations (verify standard assignments)

Step 3: Verify Each Claim

Before verifying any claims, determine research path: try calling any mcp__perplexity__* tool with a simple test. If it fails (unknown tool, MCP error), use WebSearch/WebFetch for ALL claims. Do not retry Perplexity or stop. Announce which path.

For each claim:

1. If Perplexity is available: query with specific verification question
2. If Perplexity is not available: use WebSearch or WebFetch to query authoritative APIs directly:
   • NVD API for CVSS: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=<CVE-ID>
   • FIRST API for EPSS: https://api.first.org/data/v1/epss?cve=<CVE-ID>
   • MITRE ATT&CK for technique IDs: https://attack.mitre.org/techniques/<T-ID>/
3. Cross-reference with authoritative source
4. Record: claim, source, verification result (Verified/Unverified/Incorrect), research method (perplexity|web-search)
5. For incorrect claims: document the correct value and source

Step 4: Generate Report

For each claim, output:

• Claim text
• Source cited
• Verification status
• Authoritative source and value (if different)
• Impact of discrepancy (if any)

Trusted Sources

NVD, CISA KEV, FIRST EPSS, MITRE ATT&CK, vendor security advisories, GitHub advisories.

Security Considerations

• Validate CVE ID format before querying
• Sanitize ticket IDs to prevent injection
• Detect private IPs (cannot verify externally)
• Flag unverified claims clearly