Fact-Checking

Verification Layers

Layer 1: Internal Consistency

The AI checks if the article is internally consistent:

• Do dates, names, and locations make sense?
• Are there contradictory statements?
• Does the tone match the content?

Layer 2: Cross-Source Verification

When multiple sources cover the same story, the AI compares them:

def cross_reference(article, similar_articles):
    prompt = f"""
    Compare these articles covering the same incident:

    Article A: {article['summary']}
    Article B: {similar_articles[0]['summary']}

    Identify:
    1. Confirmed facts (present in ALL articles)
    2. Discrepancies (differences between articles)
    3. Unique claims (present in only one article)
    4. Confidence score (0-1) based on agreement

    Return as JSON.
    """
    return groq_extract(prompt)

Layer 3: Historical Pattern Check

The AI flags patterns that resemble known misinformation:

• Exaggerated numbers or claims
• Unnamed or vague sources (“sources say”)
• Emotionally manipulative language

Trust Score Calculation

Trust Score = 0.4 × AI_confidence
           + 0.3 × cross_source_agreement
           + 0.2 × source_reliability
           + 0.1 × historical_pattern_score

Range: 0-100

Published if: score > 70 (scraped news)
               score > 80 (citizen news, after community review)

Source Reliability Scoring

Source Type	Base Reliability
Major newspaper	0.9
Government release	0.95
TV news transcript	0.8
Citizen report (verified)	0.7
Citizen report (unverified)	0.3
Social media	0.2