Source Reliability Bundle

Last modified by Robert Schaub on 2026/02/08 08:12

FactHarbor Source Reliability Bundle

FactHarbor uses configurable source reliability scores to assess the credibility of sources during fact-checking. These scores come from Media Bias/Fact Check (MBFC), a scientifically validated media bias database.

Quick Start

Option 1: Use Sample Bundle (50 sources)

A sample bundle with 50 common sources is included:

# Copy sample bundle to project root
copy source-bundle-sample.json source-bundle.json

# Add to .env.local
echo FH_SOURCE_BUNDLE_PATH=./source-bundle.json >> apps\web\.env.local

Option 2: Full MBFC Database (9,500+ sources)

For comprehensive coverage, fetch the full MBFC database via RapidAPI:

# 1. Get RapidAPI key from https://rapidapi.com/mbfcnews/api/media-bias-fact-check-ratings-api2

# 2. Set environment variable and run loader
set RAPIDAPI_KEY=your_key_here
npx ts-node apps\web\src\lib\mbfc-loader.ts ./source-bundle.json

# 3. Configure FactHarbor to use the bundle
echo FH_SOURCE_BUNDLE_PATH=./source-bundle.json >> apps\web\.env.local

Configuration

Environment Variables

 Variable  Description  Default
 FH_SOURCE_BUNDLE_PATH  Path to source bundle JSON file  None (no scores)
 RAPIDAPI_KEY  RapidAPI key for MBFC loader  None

Score Interpretation

 Score Range  MBFC Rating  Interpretation
 0.90 - 0.99  Very High  Wire services, fact-checkers (Reuters, AP, FactCheck.org)
 0.80 - 0.89  High  Quality journalism (BBC, NPR, Economist)
 0.70 - 0.79  Mostly Factual  Generally reliable with some issues
 0.50 - 0.69  Mixed  Verify claims independently
 0.30 - 0.49  Low  Unreliable, often misleading
 0.05 - 0.29  Very Low  Conspiracy, fake news, pseudoscience

Bundle Format

The source bundle is a JSON file with this structure:

{
 "version": "1.0.0",
 "generated": "2026-01-01T00:00:00.000Z",
 "sourceCount": 50,
 "provider": "Media Bias/Fact Check",
 "providerUrl": "https://mediabiasfactcheck.com",
 "sources": {
   "reuters.com": 0.95,
   "apnews.com": 0.95,
   "bbc.com": 0.85,
   "foxnews.com": 0.30
  },
 "metadata": {
   "reuters.com": {
     "name": "Reuters",
     "factual": "very high",
     "bias": "least biased"
    }
  }
}

Why MBFC?

Media Bias/Fact Check was chosen because:

# Scientifically Validated: Studies show "almost perfect" inter-rater reliability when compared to independent fact-checking datasets
# Comprehensive: 9,500+ sources rated
# Transparent Methodology: Clear criteria for bias and factual reporting ratings
# API Available: Programmatic access via RapidAPI
# Regularly Updated: Ratings are reviewed and updated

Design Decisions

No Hard-coded Defaults

FactHarbor does not include hard-coded source reliability scores. All scores must come from the configured bundle. This ensures:

  • Transparency: All scoring is traceable to the bundle
  • Configurability: Organizations can use their own ratings
  • No Hidden Bias: No implicit assumptions about source reliability

Sources Without Scores

If a source is not in the bundle, it returns null (unknown reliability). The analyzer will still use the source but won't apply a reliability weight.

Updating the Bundle

To refresh the bundle with latest MBFC data:

set RAPIDAPI_KEY=your_key_here
npx ts-node apps\web\src\lib\mbfc-loader.ts ./source-bundle.json

Consider scheduling this monthly to capture MBFC rating updates.

References