How Trust Scores Work
Complete transparency into every scoring decision. No black boxes.
Reviews (35%)
Bayesian-adjusted, time-decayed ratings weighted by reviewer trust level, account age, and verification status. Recent reviews from trusted members carry more weight.
Authenticity (20%)
Anti-fraud analysis: review burst detection, duplicate content, IP clustering, vote brigading, rating volatility, and behavioral anomaly signals.
Identity (15%)
Domain age, WHOIS data, registration patterns, expiration risk, country mismatch, subdomain abuse, and ownership verification status.
Security (15%)
SSL validity and grade, HTTPS, HSTS, security headers (CSP, X-Frame-Options), DNS health, MX/SPF/DKIM/DMARC presence, blacklist status, and malware detection.
Transparency (10%)
Claimed profile status, owner responsiveness (response rate and speed), public business information completeness, and active engagement with community reviews.
Performance (5%)
Uptime percentage, average server response time, downtime history, and reliability trend signals.
Fraud Detection & Risk Score
In addition to the Trust Score (0–100), we calculate a separate Risk Score (0–100, higher = more dangerous) and Fraud Probability (0.0–1.0). These are computed independently and may cap the Trust Score when severe signals are detected.
Fraud caps are applied gradually: mild signals cap at 75, moderate at 55, severe at 35, and critical at 15. This prevents a single negative signal from completely destroying legitimate entities while ensuring confirmed scam sites cannot score highly.
Bayesian Rating Formula
where C=10 (prior confidence), M=3.5 (global mean), n=review count
Score Transparency
Every entity page shows "Why this score?" — a plain-language breakdown of every signal that contributed positively or negatively. Administrators can access the full evidence tree. Scores are versioned and recalculated when new reviews, security data, or scam reports arrive.