Methodology - How We Research and Rate Cards

Data first, prose second

Every card entry is a structured fact set: KYC level and trigger, card network and kind, custody model, native vs conversion top-up coins, the full fee stack (issuance, monthly, top-up, FX, ATM, inactivity), limits, cashback terms with caps, regions, wallet support, operator/issuer background, incidents and sources with dates. Reviews are AI-drafted from that fact set only - the model may not invent numbers, fees or history. Drafts are fact-checked against the database before publication, and every fact row carries a verification date.

Where facts come from

Provider pages fetched live: pricing and fee pages, cardholder terms, supported-coin documentation, help centers and issuer disclosures. Independent leads from review roundups and news coverage are marked as leads - never treated as proof. Community research (Reddit, forums) covers complaint patterns and user experience. Quality is flagged per card: verified, reported or unverified. Unverified facts are labeled in reviews, never presented as confirmed.

Scoring

Editorial scores (0-10) are weighted: fees 22%, reliability 18%, privacy 14%, limits 12%, rewards 12%, availability 12%, UX 10%. Fees and reliability lead deliberately - a card is a cost instrument with counterparty risk, and a generous bonus never offsets a broken fee structure or a vanishing operator. User ratings are shown separately, smoothed with a statistical prior, and never mixed into editorial scores.

Freshness

Card programs change constantly - fees move, regions open and close, programs die (see the archive). We run daily reachability checks on every provider domain, re-verify facts on a 90-day cycle and regenerate reviews when facts change. Every review shows its version, generation date and fact-verification date.

Independence

Commercial relationships never influence scores, ranks or the dead-cards archive. Providers can submit factual corrections with sources via contact; we verify and correct within 7 days. We do not remove warnings or archive entries in exchange for anything.