How we test and score

Principles

Evidence over opinion

Every signal comes from a public source we link, our own analysis of that data, or a test purchase we recorded. Reviews count, but never on their own.

No paid placement

Ranking order comes only from the published score. No site can pay to be listed, ranked or removed.

Same rubric for everyone

Every site in a category is scored on the same signals and weights. Each score records the methodology version that produced it.

Right of reply

Before a negative finding goes live, the site gets 7 days to respond. Its reply is published next to the finding.

Corrections in public

Score changes and factual corrections are logged on the site's page with the date and reason.

The test journal

1. 1We buy the category's standard product from a dedicated test account, paying from our own wallet. The site isn't told it's being tested.
2. 2The full checkout is recorded. We log the advertised price and the final amount paid, so hidden fees and conversion spreads show up.
3. 3We timestamp every status change until delivery, then check what arrived against what was listed.
4. 4We send a scripted question to support and, where the category allows it, request a refund on a second order.
5. 5We check again at 30 and 90 days for clawbacks, revoked keys, recovered accounts or bans.
6. 6Screenshots and clips are published with account details, wallet addresses and order IDs blacked out. Unredacted originals are kept privately for disputes.

What the score is made of

How the score is calculated

Each signal is scored 0–10 against a fixed rubric. A site's Trust Score (0–100) is the weighted average of its signals, where each signal's weight is its pillar's category weight shared across that pillar's signals. Anything we haven't been able to check counts as the average across all rated sites, so missing data neither helps nor hurts. Risk flags then subtract points: −15 for an active review-platform warning, up to −15 for a sudden collapse in traffic, up to −10 for a very new domain and −5 for a tiny audience. Size and age earn nothing extra. Signals past their refresh window count at half weight, and expired ones don't count at all.

A score is shown once at least 30% of the public score weight is measured; under 60% it is labelled "limited data". A test purchase never raises a score; ties are sorted by name. Gambling and case-opening sites are out of scope.

Freshness

Categories

Every signal, its pillar and how it's scored.

Data sources

• The sites themselves — terms, refund and warranty policies, contact and company details, FAQ and help pages, read and linked for every signal.
• Public review platforms — overall rating, number of reviews, reply behaviour and platform warnings, plus our own grouping of recent negative reviews.
• Traffic estimates — third-party estimates of monthly visits, top countries and global traffic rank. Shown rounded; used only to flag very small or collapsing sites.
• Domain records — domain registration date and first web-archive capture.
• Our own tests — test purchases and verified user reviews with proof of purchase.

Changelog

• v1.2 · 3 Oct 2026

  Policy findings are now read from a crawl of each site's own terms, refund, FAQ, help-centre and product pages (including collapsed FAQ answers), and every finding shows the sentence it comes from, checked against the page. A policy counts as missing only after a complete crawl found nothing; otherwise it counts as the average. New unfair-terms penalty (store-credit-only refunds, forfeited balances, cancelling paid orders without a refund, keeping balances after a chargeback). Removed signals that did not separate sites or could not be checked reliably: duo option, verification rules, payout methods and the security pillar (payment and 2FA checks move to test purchases); buyer protection goes to 25%. Risk disclosure is scored only where game rules apply. Review-platform notices about a site's category no longer cost points. Complaint rates need 30+ reviews and use a scale fitted to the observed range (10 at 1% or less, 0 at 20%+). Review ratings with few reviews are pulled toward the average of all sites instead of toward a low neutral value. A site that claims its service is “100% safe” or “risk-free” anywhere scores 0 for risk disclosure. Escrow counts as full protection when funds are held until the buyer confirms or a 3+ day dispute window passes.

• v1.1 · 2 Oct 2026

  Web-first Trust Score (0–100) for every site from public data. Test purchases become a separate Tested pillar. Added review-platform warnings, recent complaint analysis, verified user reviews, track record (time online, established audience and traffic trend — size itself earns nothing). No price pillar: products differ too much between sites to compare fairly. Security counts only what can be confirmed publicly; 2FA is checked during test purchases. Test results are shown next to the Trust Score instead of being added to it, so every site is ranked on the same public data; a failed test still makes a site Avoid. Track record and review-platform warnings became risk penalties instead of a weighted pillar; unchecked items count as the average across all sites; buyer protection cut to 20%. Minimum data coverage before a score is shown.

• v1.0 · 1 Oct 2026

  First methodology: 9 categories, 7 pillars, hard-fail rules and freshness decay.