How the Exchange Data Score Is Computed: Formula, Weights and Sample Thresholds
Exchange display order follows CoinMarketCap rank; this page publishes the separate data score, weights, sample thresholds, missing-data handling and commercial-disclosure rules.
How This Ranking Is Computed: Formula, Weights, and Sample Thresholds, All Public
Bottom line first: This site's ranking isn't our opinion — it puts "objective data + real user ratings" on the table. That's why we publish the formula, the weights, the sample thresholds, how missing data is handled, and the rules for disclosing commercial relationships, all on the page — you can recalculate it yourself, and you can challenge it.
Display order (updated 2026-10-18): the exchange board and DEX directory follow CoinMarketCap rank; the wallet directory follows the existing product source rank because CMC does not publish an official wallet ranking. The data score, weights and sample rules below describe a separate reference score and no longer determine the displayed order.
Data and review scores (reference columns)
| Ranking | Composition | Source |
|---|---|---|
| A: Data ranking | Trading and holding cost · Withdrawal cost and arrival · Liquidity/slippage · Product coverage · API capability · Regional availability and KYC · Stability (last 30 days) | Our own exchange database, each dimension carrying its source and fetch time |
| B: User rating ranking | Deposit experience · Withdrawal speed · Support responsiveness · Fee perception · App/web experience · Stability (1–5 score) | Real user ratings on the site |
| Combined score (reference) | Data score × 60% + rating score × 40% (reference only; it does not determine CMC order) | Same as above |
Four hard rules for the data score
- Sample threshold: a rating sample of fewer than 30 doesn't enter the combined score; it only shows "sample accumulating." We'd rather show a blank than let 3 people rating 5 stars push you to number one.
- Bayesian shrinkage: the rating score is shrunk toward a prior mean (20 / 3.5), so the fewer the samples the closer to neutral — this prevents small samples from storming the ranking.
- Dimensions with no discriminating power get no boost: if every exchange has identical data on a dimension, or only one has data, or all are missing, that dimension is treated as a neutral 50, and it's listed item by item in the ranking notes. Otherwise "everyone is missing data" would become "everyone gets full marks."
- No black-box overall ranking + no selling of placements: positions are never for sale, we take no sponsorship from exchanges, and affiliate relationships are disclosed item by item on every card.
How missing data is handled (this is the easiest place to manipulate)
- Missing dimensions: counted as a neutral 50, never guessed and never faked with a 0.
- Every row outputs a "data completeness" figure, e.g. "Bitget 86% (missing liquidity/slippage)."
- The top of the ranking gives a disclosure: completeness differs, so a rank shouldn't be read as "overall superiority" — because it reflects our collection progress, not a complete verdict on which exchange is better.
The data score and display order can change, but the changes have to be explainable
We store a snapshot on every data refresh. Any rank change has to be able to answer "which piece of data changed," and show the related exchanges' changes at the same time — because a swap in rank is often not one exchange getting better, but a rival getting relatively faster or slower.
→ See the current ranking: current ranking
How you can verify us
- Go to an exchange detail page, click open the source links and fetch times dimension by dimension, and compare them yourself on the official sites
- Use the total cost calculator to recalculate the costing methodology yourself
- Submit your own rating for any exchange (one vote per account; we don't delete negative reviews, only remove rating manipulation, and publish the number removed)
Data source and fetch time: ranking API and per-dimension sources (2026-10-12T14:13:07Z snapshot). Methodology changes are recorded on the page, never edited silently. Snapshot date 2026-10-12.
Disclaimer: The ranking reflects only the dimensions we collected and the ratings submitted, not the whole of the facts; this site is not investment advice.