Classification is curated in-house rather than inherited from a provider's tags. These are the rules it follows.
Two conditions, both required: its own base-layer blockchain and its own consensus mechanism. A token issued on someone else's chain fails the first. A fork that borrows another chain's validators fails the second.
The ranking is then organised by that consensus — proof of work split by algorithm, proof of stake by protocol, plus BFT, DAG, proof of space and hybrids — because how a chain agrees with itself says more about it than its market cap does.
Anything whose value is another asset: wrapped and staked coins, fiat-pegged stablecoins, debt-minted synthetics, tokenised treasuries and commodities. They are listed with their peg, backing and host chain — and never added to the asset total, because the thing they represent is already counted.
The listing calls them stablecoins, which is the same idea one level out: a stablecoin is a claim pegged to a reference asset, and a wrapped coin is that structure with a crypto reference instead of a currency.
| List | Ordered by |
|---|---|
| Cryptocurrencies | market cap after claims are removed, with the low-volume tail hidden from anonymous views (see § What market cap misses) |
| Exchanges, DEXs | 24-hour volume, with each contradiction against independent evidence costing the venue position (see § What an exchange can and cannot fake) |
| Buy & spend | P2P first, then on-ramps and gift cards; within each, self-custody before merchant before escrow before custodial, no-KYC before limited before full KYC, and web rank only as a tiebreak |
| Services | decentralised networks first, then gateways, then conventional providers; crypto payment before card-only |
Popularity never leads. It settles ties inside a group, because custody and identity requirements are what a reader is choosing on.
Merchant in the custody column means a shop rather than a wallet: you pay per order and receive a product, and no balance of yours is ever held. Gift-card sellers work this way. It is deliberately not filed as self-custody — there is a gap between paying and receiving the code, so you are trusting the seller to deliver — but it is not custody either, and listing it as such would put a one-off payment in the same bracket as an exchange holding your coins.
Almost everything an exchange publishes about itself, it also controls. Reported volume is a claim about a past that left no artifact — nobody outside the venue can inspect trades that may never have happened. Its order book is better evidence, because a book is a claim about the present that has to be standing there when you look, so this site reads books directly from each exchange's API rather than taking an aggregator's word for the depth. Every 15 minutes.
But a book is still produced by a matching engine the venue owns, and quoting orders you never intend to fill costs nothing on your own market. A new exchange can stand up a deep book and a large volume figure on its first day. So measured depth is allowed to demote a venue whose book contradicts its volume, and never to promote one for looking deep on paper.
The number an exchange cannot type about itself is its web traffic. The Tranco rank shown on every venue counts resolvers in the wild; no amount of self-reporting reaches it. That is why it anchors the ranking here and is never overridden by a flattering book — and why this site did not follow CoinGecko's May 2026 decision to drop web traffic from its Trust Score in favour of order-book depth, which replaced an independent signal with a self-reported one. That decision is discussed in a separate post.
Each CEX row on the ranking carries a 0-100 score built from nine signals — web rank at ×20 and eight more at ×10 each. Weights are stated in the tooltip and again here; the reasoning behind the whole model is written up at length on the blog.
| Signal | Weight | Why this weight |
|---|---|---|
| Web rank | ×20 | Tranco. The one signal an exchange cannot type about itself. A domain outside the global top million scores the floor rather than being skipped: that absence is itself a measurement of traffic. |
| Age | ×10 | Years since founding. A decade of surviving cycles is real evidence, and it cannot be typed either. |
| Liquidity size · transparency | ×10 | Dollars resting on the thinner side of the book within 1% of mid — a large order eats one side — summed across the venue's ten largest markets and read from its own books: $50K scores 1.5, $500K 2.5, $5M 3.5, $50M 4.5, $160M the full 5. Where those markets were not read directly, the BTC book's thinner side within 0.1% of mid stands in, scaled ×10, and a sum known only through CoinGecko never scores above 2.5. Capped by how much of the book anyone can read: the whole book or ±1%+ keeps the full signal, the ±0.1% band caps at 4, too shallow at 3, depth known only through CoinGecko at 2.5, no book 0. Transparency is a ceiling, never a bonus. A book read and found empty scores 1; a book we could not read is skipped. |
| Reliable volume | ×10 | The reported 24-hour volume discounted by the reliability ratio below — at 10× the top-ten norm, a tenth of the claim is scored, on the same log scale. A claim with nothing behind it scores 0. |
| Reliability ratio | ×10 | The ratio itself, as a signal: claimed volume against the venue's own traffic, 1× the top-ten norm. At or under 1× the full 5, falling log-linearly to 0 at 25×, where the venue is also flagged. A claim with nothing behind it scores 0. Being small never costs; only claiming more than the traffic supports does, and such a claim loses twice — the volume it inflated and the ratio it broke. |
| Liquidity / volume | ×10 | Liquidity standing within 1% of mid on the venue's ten largest markets as a share of their daily volume — ten dollars traded on one of liquidity is 10%, a hundred on one is 1% — read from its own books, against the top-ten norm. At or above the norm 5; each halving below costs a point; a thirty-second of the norm scores 0. The smaller the share, the less the venue scores. Skipped where the majors' books cannot be read. |
| Proof of reserves | ×10 | Open audit — addresses published and checked on chain by CoinSpectrum > signed third-party attestation = addresses published but not read this build > verified page > unknown > chain holds less than the venue owes = none > a report that a page exists with no page to show, which scores 0. See below. |
| KYC | ×10 | Waived up to a limit scores 3, none at all 2, required 1, on the 0-5 scale, so KYC decides at most 6 of its 10 points. The one property that lands on the reader directly. Required is the industry default and not a finding against the venue; a venue that asks nothing of anyone sits under one that asks above a threshold. |
| CoinGecko trust | ×10 | The aggregator's 0-10 exchange grade, halved. A second opinion on liquidity and traffic, not ours; it also stays a contradiction at 6/10 or lower. Skipped where the feed returns none. |
Each signal scores 0-5 and the weights sum to 100. The one number a venue types itself, volume, never stands alone: it is discounted by the reliability ratio, and the ratio is a signal of its own, so a claim with nothing behind it loses twice. Web rank, liquidity size, volume and age are on continuous log scales rather than fixed bands, so a book of $9M genuinely outscores one of $3M rather than tying at the same band ceiling, and a web rank of #6,000 outscores one of #21,000 rather than sitting in the same tier.
Missing data does not penalize. Where the underlying number is null — a Tranco rank outside the top million, a CoinGecko trust score the endpoint stopped returning — the signal is skipped and the score normalizes to what could actually be measured. Absence of evidence is not evidence. The one exception is written above: a Tranco rank outside the top million is not missing data, it is a reading of zero.
The score is off by default in the ranking column but can be turned on in the profile settings under three display modes (number, letter grade, filled bar). The reasoning behind each row's score is on hover: every signal, its weight, its points.
On top of the composite, each of the following contradictions costs a venue 15 points from its score and drops it beneath every venue that trips one fewer. They accumulate; a venue tripping three sits beneath one tripping a single one. CoinGecko's trust grade no longer enters the composite; it survives here as a contradiction. All but the last three apply only to venues claiming more than $50M a day — a small exchange reporting small honest numbers contradicts nothing and keeps its place. The last three are scale-free and apply to every venue; each has its own section below.
Being unable to read a venue's book is deliberately not counted against it. Several exchanges refuse automated requests outright; that is a limitation on this side, not a fact about them, and it is recorded separately from publishing nothing at all.
A proof-of-reserves page is a claim. Published addresses turn it into something anyone can verify at any time — an open audit — and "anyone" includes this site. Where a venue publishes its reserve addresses in a machine-readable form, every build reads them on chain — Bitcoin, Litecoin, Dash, Ethereum, the ERC-20 dollars and the pegged tokens on BNB Smart Chain — and compares what they hold with what the venue says it owes its users. A coin is judged only when the chains read here carry nearly all of what the venue lists for it; otherwise it is reported as partly readable and decides nothing. A monthly address file such as Binance's or OKX's is read against liabilities counted weeks earlier, so a gap inside a stated tolerance is movement since the snapshot, not a shortfall. If every judged asset covers its liabilities, the venue takes the top of the reserves scale, because nobody had to be trusted for it. A signed third-party attestation sits just under that: someone vouched, but you cannot look for yourself. The venue's own "we hold X" figure is only a snapshot and is allowed to go stale; where it exceeds the chain, the chain figure is shown beside it and used. Only a material shortfall — the chain holding less than the venue says it owes — counts against the venue: the floor of the scale, and a wash wire, since it is the venue's own numbers against public evidence. Privacy coins and assets listed without an address cannot be checked and never decide the outcome either way. A report that a page exists, with no page to show, scores nothing at all — under confirmed none — because an unverifiable claim of reserves is the shape a fabricated one takes. The arithmetic behind each result sits in the row's tooltip, on the venue's page, and on the reserves page, which lists every open audit in full.
Could a venue's reported volume have executed against the book it shows? Every build takes each exchange's ten largest spot markets and reads each pair's order book — directly from the venue's own API wherever the book registry can name the pair, and through CoinGecko's keyless tickers only for venues whose books refuse automated reads, marked as such. Each pair's daily turnover is set against the liquidity standing within the widest band of mid its book fully covers, out of ±2%, ±1% and ±0.1%; public books reach different distances, from Coinbase's whole book to sixty levels on the Korean venues, and widening a band only ever makes a venue look more liquid, so a pair that fails at 2% fails harder at 0.1%. A pair whose book does not reach 0.1% is left out rather than guessed at.
An honest market turns over its near-mid book tens to a few hundred times a day. A washed market turns it over thousands of times, because the trades were never matched against standing orders — and the signature is one small pair, not the whole venue. So the thresholds are calibrated on the ten best venues by this score, ten times the median of their pairs with a floor and a ceiling, and the wire's figure is the share of its top-ten volume sitting in pairs beyond the threshold, with the worst pair named. A quarter of the volume beyond the book shows in the liquidity signals; half, read from the venue's own API, is a contradiction like the ones above — a share known only through CoinGecko lowers the signal but never flags. Depth is the venue's own data, so the wire can only demote. The same measurement also feeds the liquidity / volume signal: the liquidity standing within 1% of mid across those markets as a share of their daily volume — ten dollars of volume on one of liquidity is 10%, a hundred on one is 1% — weighted by where the volume trades, so a deep BTC book cannot cover for a washed listing that carries most of it, against the median of the ten best venues, scored continuously: at or above the norm the full five, one point off per halving below it, so the smaller the share the less the venue scores.
The last wire is a ratio — claimed volume against the venue's own traffic, with 1× the norm of the top ten — calibrated rather than fixed, and it applies at every size. Every build takes the ten best venues by this score that trip no wire and measures what they turn over per site visit: claimed monthly volume divided by an independent monthly-visit estimate. That gives a median and a band — the range honest venues occupy — and, across every clean venue with a Tranco rank, a fit of how volume falls off with rank. Each exchange is then measured on both: its own dollars per visit against the median, and its own volume against what its rank predicts, and the smaller of the two ratios is taken, so a single broken traffic estimate cannot convict a venue. The reference figures are published in the data file and the arithmetic sits in each row's tooltip.
Inside the band nothing happens. From five times the median a venue loses ten points; from twenty-five times it is flagged like the contradictions above; from a hundred times it is flagged twice. Below the flag line the ratio does not subtract points on its own: it discounts the volume signal, so a venue at 4× the norm is scored on a quarter of what it claims. A large claim with no traffic estimate and no Tranco rank behind it at all counts as unbounded. The low side never demotes: traffic above what the volume implies is brokerage-style flow or under-reporting, not fraud. Being small is never the problem here. Claiming more than your own traffic supports is.
Every list here that starts with market cap treats it as a starting order, not a verdict. Market cap is circulating supply × spot price — an arithmetic identity. It presumes the price at which the last few coins traded would hold if the rest were sold. For most assets outside the top few dozen, that presumption breaks the moment anyone acts on it: a token with a billion units in supply and one buyer bidding a penny has a $10M cap on paper, and nothing the moment that buyer walks away.
Liquidity is what turns market cap from a number into a claim. Daily volume is the closest observable proxy — coins that trade tens of millions of dollars a day have caps you can act on; coins that trade a few thousand do not. This site treats the two side by side rather than sorted by cap alone.
Assets below $10,000 in daily volume are hidden from the anonymous view. They stay in the data and remain accessible to signed-in visitors who choose to see the tail, but they do not lead any first-time reader's look at the ranking. At that level of trading, the cap column is display, not measurement — a first real seller finds no bids and the paper number vanishes.
Zero volume is not zero information. A chain with a live block time and a working ledger but no exchange listings has functioning consensus and a broken market. The listing marks that state — a dash in the volume column rather than a zero — rather than ranking on a cap the market is not defending.
Each chain is asked for its own figure first — its explorer or node API — with aggregators used to fill gaps and, where both exist, to check the first against the second. That check earns its keep: one explorer publishes a hashrate field a thousand times lower than its own difficulty implies, and one daemon counts hash attempts where the network is measured in solutions. Both were caught by comparison, not by trust.
Multi-algorithm chains have no single hashrate in the strict sense. DigiByte mines on five algorithms at once and Verge on five; adding SHA256 hashes to Scrypt hashes sums numbers whose units cost different work. Every tracker sums them anyway, so we follow the same convention — per-algo difficulty × 232 ÷ the per-algo target time — to stay comparable with what a reader sees elsewhere.
Hashrate does not compare across algorithms. A hash is not a fixed unit of work: SHA-256 is deliberately cheap, so an ASIC performs some 1014 of them a second, while Scrypt is memory-hard and each hash costs far more silicon and energy. Dogecoin at ~4 PH/s is not 80,000 times weaker than a SHA-256 chain at ~330 EH/s — it is merge-mined with Litecoin, so it inherits the Scrypt world, where the entire global fleet is a few PH/s. Read against its own parent it carries more hashrate than Litecoin does. Comparing security across algorithms needs cost — energy burned, or money spent per unit time — not hash counts.
The column reports a chain's own work, not every hash recorded against it. Merge-mined hashpower is borrowed: a miner securing that chain is paid to secure another one, and stops the moment the parent's economics change. Borrowed security is real security — it is simply not the chain's own, so it is shown, marked, and never ranked on.
How much is borrowed differs, so it was measured rather than assumed. Dogecoin: 60 of 60 consecutive blocks carried an AuxPoW header, so its own hashrate is nil and the figure beside it — around 3 PH/s — belongs to Scrypt parents, some of them other than Litecoin. Fractal Bitcoin does mine for itself, interleaving its own blocks with ones carrying Bitcoin's work, but publishes no split; inferring one from a version bit whose meaning cannot be confirmed would be inventing a number, so its own share reads as unknown.
Marked chains sort last among the mined ones, because a column of numbers cannot show that the miner behind a figure is being paid by someone else. Verge carries the mark for a different reason: no source we can reach publishes its Scrypt hashrate, so its figure covers four of its five algorithms and is a floor rather than a total. It is dominated by Blake2s regardless.
Exchange volume is self-reported and gameable; web traffic is harder to fake. Where a venue claims large volume while drawing no measurable traffic in the Tranco list, it is flagged and ranked last. A small venue with no rank is simply below the global top million — that is not held against it, and it is never flagged on volume it never claimed.
Wash trading is the reason to distrust volume in the first place. On the exchange side, some venues generate volume against themselves — or credit accounts with volume never traded — to climb aggregator rankings. On the asset side, low-float tokens are pumped through pairs on obscure venues to make a coin look active. Both patterns are common enough that a raw 24-hour figure is a starting point, not evidence.
Detected wash behavior drops the venue to the bottom. The Tranco cross-check above is the current signal — no web traffic while claiming large volume is the cheapest thing to spot. Signals being added: volume claimed against orderbook depth (a book that fills a $10K market order with 3% slippage cannot honestly be moving $100M a day), and the ratio of a listed pair's volume to the parent asset's volume on venues that don't fake theirs. None is proof by itself; together they push venues whose numbers can't survive daylight to the end of the list.
The direction is real liquidity, not reported volume. Orderbook depth at fixed trade sizes is what tells a reader whether a market can absorb a $10K or $100K order without moving the price — and what a market-cap figure would actually cost to realise. This site is moving toward showing that depth alongside volume, so a reader can distinguish a market that clears real orders from one whose reported volume is theatre.
Where a regulator has taken a position, it is reported as theirs: the SEC and CFTC's 2026 joint interpretation naming sixteen tokens as digital commodities, and the assets alleged to be securities in earlier complaints. Assets nobody has ruled on are left unclassified rather than sorted by our own guess.
The portfolio and watchlist live in your browser. Signing in is optional; it syncs them with a key derived on your device from your seed phrase and never sent, so the server holds ciphertext it cannot read. An off-site copy is opt-in per account and off by default.