Three evidence sources
- Editor review — 40%: percentage of matching valid comparison criteria, averaged over one latest valid approved assessment per editor author. Each qualifying assessment needs at least 3 criteria with a boolean match and nonempty aspect, advertised and delivered findings. Empty comparisons earn no fallback score.
- Customer reviews — 35%: one latest approved review per distinct customer; average (rating − 1) × 25. Evidence-checked badges do not change this weight.
- Internet reputation — 25%: latest approved report only. Positive = 100, neutral = 50, concern = 0; average valid tones within each normalized external registrable domain, then weight each domain equally. Store-owned domain and subdomains, missing/invalid tones are excluded.
No evidence, no score
Minimums: 1 valid editor author assessment, 3 distinct customers and 3 distinct external registrable domains. If any minimum is missing, no score; weights are never redistributed. Domain boundaries use the Public Suffix List, so pages/subdomains do not inflate samples. Distinct domains are not proof of independent ownership or control; ownership requires separate review.
Official, versioned records
40% editor + 35% customer + 25% internet − upheld penalties, clamped to 0–100 and rounded to a whole number. A new policy starts unapproved. Evidence changes invalidate the saved official result until a moderator recomputes it. Historical results remain in the audit/history; demo numbers are illustrative only.
Confidence, samples and date
Every score is shown with its confidence level, sample counts and assessment date.
Badges
- Editor Tested — we made a real test purchase.
- Positive Record — consistent positive evidence.
- Mixed Experiences — evidence points both ways.
- Documented Concerns — specific problems backed by evidence. Not a verdict of fraud.
- Identity verification is separate and never changes a score.
Penalties and appeals
Only upheld reports count. Duplicate normalized ad URLs count once per store (largest applicable penalty); reports without a URL are separate cases. Each case is capped at 10 points, total at 30. Under appeal or dismissed reports never count. Every decision and policy activation is audited.
AI assistance
AI drafts use provider-retrieved public citations only. An admin verifies identity and each source before approving. Private Instagram/Facebook comments are not covered. AI does not award trust/scam badges or decide guilt.