Factcheck

Overview

Factcheck extracts every verifiable claim from articles and social media posts, verifies them against authoritative sources, and applies minimal corrections while preserving the author's voice. It uses the Chain of Verification (CoVe) method: formulate a specific search query, search authoritative sources, fetch and read the actual page, then cross-reference with a second independent source for critical claims.

Workflow

1. Setup — read the source file, create backups, set up a working directory.
2. Claim Extraction — identify every factual claim with type (stat, date, name, tech, science, legal, quote) and importance level (critical, high, medium, low).
3. Verification — process claims by importance. Critical claims require 3+ sources, high claims need 2+, medium need 1+, low are spot-checked. Each claim gets a verdict: ACCURATE, INACCURATE, OUTDATED, MISLEADING, UNVERIFIABLE, or NEEDS_CONTEXT.
4. Consistency Analysis — check for internal contradictions, ripple effects from corrections, and temporal consistency.
5. Apply Corrections — minimal changes to fix facts, with inline HTML comments citing sources for the author's reference.
6. Review and Finalize — present changes for approval before modifying the original file.

Source Hierarchy

Official product pages, API documentation, official blog posts, GitHub repos, peer-reviewed papers, reputable news outlets. Social media posts are never treated as authoritative sources.

Key Rules

• Fix facts, not style — the author's voice is sacred
• Always back up the original before modifying
• Every correction must cite a source — no hallucinated URLs
• Keep the article in its original language
• Note "as of [date]" for facts that may change over time