coinsprey

Methodology

Every number on this site is computed by us from public market data. This page says how, and where a number is only an estimate.

coinsprey

Where the data comes from

We read the public market data of 205 exchanges every minute: Binance, YUBIT, Echobit, CEEX, BloFin, Tapbit, KuCoin, OKX, Bybit, Pionex, Gate, Bullish, Azbit, Coinbase, Bitrue, Biconomy, Toobit, Ourbit, Kraken, Hotcoin, XT.COM, DigiFinex, FameEX, HTX, Indoex, MEXC, WhiteBIT, Hibt, Bitget, Binance Alpha, Crypto.com Exchange, Upbit, BitDelta, Batonex, Bithumb, Poloniex, BigONE, Aivora, CoinW, ZKE, Websea, BYDFi, Bitbase, OneEx, PancakeSwap (BNB Chain), GroveX, BingX, Coinstore, BitxEX, Uniswap V3 (Ethereum), Bitfinex, BTSE, Bitstamp, HitBTC, Bitvavo, CZR Exchange, MGBX, BitBaby, Uniswap V3 (Base), P2B, EasiCoin, BitTap, Coinone, Orca, KTX, HashKey Exchange, CoinTR, PointPay, Byte Exchange, OSL Exchange, Meteora, EagleX, Korbit, Binance.US, Raydium, Revolut X, BtcTurk, Coincheck, Gemini, Coins.ph, bitFlyer, Bitkub, Bitso, PancakeSwap (Base), CEX.IO, Vindax, ICRYPEX, bitbank, Bybit EU, NonKYC, BlueBit, CriptoSwaps, Figure Markets, Backpack, Foxbit, XBO.com, KoinBX, MAX Exchange, CoinAmount, Uniswap V3 (BNB Chain), HollaEx, Bittime, Blynex, BitPreço, Indodax, Koinpark, VALR, Luno, EarnBIT, Wallex, Webot, Kanga Exchange, NexDAX, Aerodrome Slipstream, BitTrade, Deepcoin, BTC Markets, Ripio Trade, NEX Exchange, CoinJar, Bitbegin, Kinesis Money, UEX, AltCoinTrader, BIT.TEAM, WazirX, HashKey Global, Quidax, One Trading, Paymium, Blockchain.com, Rabid Rabbit, Richamster, EXMO, FreiExchange, Cexius, FynEx, Qutrade, Trapix Exchange, x.me and 65 more that do not count towards prices at the moment. That is 54,014 markets and 4,639 coins. We do not take prices from other aggregators.

Most exchanges give all their markets in one answer. A few give one market per request; those are asked market by market, a bounded number each minute, so that every market is read at least every second minute. Where an exchange of this kind has more markets than can be read that often without straining it, the markets with the largest volume are the ones kept current and shown, and the others are asked in turn only to learn when one of them has grown into that group.

Price

A coin's price is a trimmed volume-weighted average of the last traded prices of its markets. The volume a market weighs with is its estimated real volume (since 0.31). A market counts when all of this holds:

1. Its data is fresh: the exchange's ticker was fetched less than 3 minutes ago.
2. Its quote converts to US dollars. Markets quoted in USD and USDT are taken at face value; a market quoted in another asset (BTC, EUR, USDC) is converted through that asset's own dollar price.
3. It traded in the last 24 hours, and for at least $100 if the coin has any market that did.
4. Its price is within 10% of the volume-weighted median of the markets that passed 1 to 3.

The median follows the volume, and the volume is each exchange's own word. One market reporting more than all the others together would set the price alone and turn everybody else into outliers. So a price that few exchanges hold gives way to one that many do (since 0.17): when the markets left outside the 10% agree among themselves on at least three exchanges, and on more than twice as many as stand behind the median, they are the price and the others are the outliers. When this was introduced it changed eleven coins, each priced by a single market against three to thirty-one exchanges.

A price that a single exchange holds alone gives way sooner (since 0.36): to two exchanges that agree with each other, when the lone price is 1.5 times or more away from theirs. That far apart it is not a local premium or a thin market: it is another asset trading under the same ticker, or a market that stopped trading and still reports volume. Closer than that, two small exchanges against a large one change nothing and the volume keeps the price. The 1.5 is our own choice and not calibrated. A lone market five times or more away is afterwards set apart as an asset of its own, as described under the pairing of tickers.

A market whose order book does not back its volume steps aside in a price, but its price is not in doubt for that. So where the markets set aside this way agree with each other on a price 1.5 times or more away from what the remaining markets give, on enough exchanges to be a consensus by the two rules above, it is the remaining ones that are off (since 0.37). All of them then stay, weighed by the volume they report, and the price is marked as resting on figures that cannot be stood behind: confidence at most 40, outside the market's totals.

Such a price also gives a coin no market cap, even where its circulating supply is known (since 0.37): it would be ranked by it among coins whose prices are vouched for. And the volume of an exchange that is not included makes no dust of an included exchange's market: rule 3 above compares a small market with the other markets of included exchanges only. Before this, a coin whose one included market fell under $100 a day was priced by an excluded exchange 41,000 times above it and stood 34th by market cap for half an hour.

A market's price is the price of its last trade, however long ago that was, and the volume it reports gives that price its weight. So a market that reports $1M a day or more while its newest public trade is over 30 minutes old steps aside when the coin has markets that trade (since 0.18), and since 0.40 so does a market of any size from $100 a day whose newest public trade is more than a day old: a day without a trade has no day's volume (one reported $60k a day 421 days after its last trade and priced its coin at a sixth of what it traded at elsewhere). Since 0.40, too, a market that has stopped trading steps aside for any market of the coin that trades, even one that does not count. It is read on, and counts again as soon as a reading shows a recent trade. When this was introduced, eighteen such markets weighed in a price; one of them, silent for three months, carried 92% of its coin's. A coin whose only market is such a one keeps that price, capped at 40 in confidence and left out of the market's totals. A market whose trades cannot be read is not judged this way.

A coin none of whose markets traded in the last 24 hours has no price now and no rank. It is not removed: for 30 days after its last price it stays at the end of the list, after every coin that trades, with that last price and the day it is from ("last priced"), and without the figures it does not have (changes, volume, depth, confidence, market cap). It is left out of the market's totals. After 30 days it is found by search only. Since version 0.10.

An asset split off a ticker (see pairing) is named "<ticker> (<exchange>)" until an exchange that trades it gives it a name. While it still carries that name and its price rests on one exchange, it has no rank: it stays in the list after every ranked coin, with its price and figures but without a number, and it is left out of the market's totals, the map, the most traded and the sorted lists. Its figures are that one exchange's word about an asset nobody else names, and some such exchanges only repeat old prices. Since version 0.41.

The price is the average of the counted markets weighted by their 24h volume in dollars. A market that fails a step is still shown on the coin's page under "Excluded markets" with the reason: stale data, quote not convertible, no volume, under $100 of volume, or price too far from the others.

What counts, and how far to trust an exchange. Not every figure an exchange publishes can be taken at its word, and we do not keep a list of the ones we like. We measure, the same way for every exchange, large or small, and show the result instead of hiding the exchange.

  • Each market: does its order book back its volume? We read the order book of every market above $100 a day and show what is in it: the bid/ask spread, the value of the orders resting within 2% of the best bid and of the best ask (with the 5% and 10% bands beside it), and those orders as a share of the volume reported for 24 hours. Each side is measured from its own best price (since 0.16; before it from the middle between them, which gave a market with a spread wider than the band no depth at all, whatever rested in its book: 982 markets on 45 exchanges). How far apart the two sides stand is the spread, a figure of its own. Where the best prices of a published book cross (asks under the best bid), the orders on the wrong side of the market's last traded price are set aside first (since 0.35): they rest in the list but cannot be what the next trade meets. A book with more than a tenth of its price levels on the wrong side is not measured at all. A market is backed when those orders are worth at least 1% of its 24-hour volume. Tens of millions of volume over a few hundred dollars of orders is what invented volume looks like; such a market steps aside when the coin has backed markets. A book that our read cuts short before 2% only gives a lower bound and settles nothing; neither does a book not read yet.
  • Small markets get figures, not conclusions. A market trading under $10,000 a day is shown with the same figures, but nothing follows from them (since 0.16): it does not step aside, its volume is not cut and it carries no mark. On a few hundred dollars a day the ratio of volume to resting orders is noise.
  • Each exchange: a level. The share of its measured volume that trades on backed markets: Backed (90% or more), Mostly backed (50%), Partly backed (10%), Unbacked (less), or Not measured when we cannot read its order books at all. The level is a description, not a gate: a backed market counts even on an exchange where most are not. It is taken over every measured market, small ones included, each weighing as much as it trades.
  • Each exchange: do its prices agree, does its data arrive? Over its markets of coins that another exchange prices too and that traded at least $100 in 24 hours (since 0.11: a market that barely traded shows a last price from some earlier hour), the median distance from our price must be at most 1%, or at most the exchange's own typical bid/ask spread on those markets where that is wider (2% where its tickers publish no spread; since 0.12: on a thin market the last trade falls on the bid or on the ask, and that distance is the market, not a wrong price). Markets that are a different coin under the same ticker are left out. The list of exchanges shows the outcome as Agree, Within its spread, Within 2% or Off the market (with fewer than 5 such markets there is nothing to compare, and this is not held against it). At least 95% of our reads of it must have succeeded over the last 24 hours, with at least 1,000 reads. A decentralised exchange we read from a chain gives nothing while we finish its first reading after our own restart; those minutes are our start, not a failed read, and are not counted (since 0.43). An exchange that fails either, or whose order books we cannot read, is shown but none of its markets count.
  • Each exchange: are its volumes its own? Since 0.30. Two exchanges that match their own customers' orders never report volumes in one fixed proportion, market after market. Over the markets an exchange shares with each other exchange (at least 10, each reporting $10,000 a day or more), the ratio of the two volumes is taken market by market; where the middle half of those ratios lies within 5% of their median and the exchange is the smaller of the two, its figures are computed from the other's, not traded: every market at exactly a tenth of a large exchange's volume, for example. Such an exchange is shown, with the exchange it follows, the proportion and the number of markets, and none of its markets count. No other measurement sees this: the order book copied along with the volumes “backs” them and the trade feed agrees. The limit of 5% is a judgement from the day it was set (independent exchanges stood above 40%, the computed ones under 3%), not a calibration. Like every change of standing it takes six hours in a row.

Some exchanges have no markets of their own: a local branch that trades on its parent's order book, or a venue that shows another exchange's list, sometimes with the volume cut in half. Their figures are the other exchange's. Such an exchange is listed and labelled “passes on” with the name of the exchange behind it (since 0.23), and its markets never count: the same trades would be in a price and in a total twice. It is recognised by comparing the two lists market by market (the same markets, prices equal to the last digit, volumes equal or in a fixed proportion), and entered by hand; it is the one thing about an exchange that is not measured hourly.

Decentralised exchanges (since 0.24) are read from their own public lists of pools: Orca, Raydium and Meteora on Solana, PancakeSwap on BNB Chain and on Base. A pool gives its two tokens by contract, the price of one in the other, the volume of 24 hours and the value locked in it. A ticker on a chain is whatever the token's maker wrote, so a pool's token is matched to a coin only by its contract address, as an exchange that lists the coin gives it (the chain's own coin by the contract pools hold it in). A token that no exchange read here has named by contract is not listed, unless its pools show for a whole day that it is worth listing (see below): a chain holds hundreds of thousands of pools, nearly all empty or made to deceive. Only pools holding at least $50,000 and trading at least $1,000 a day are read, and a pool must have a dollar token or the chain's own coin on one side, in which the other is priced. A pool has no order book. What stands in its place (since 0.26) is what the pool can absorb before its price moves 2%, both ways together, computed from its liquidity. For a constant-product pool that is 1.98% of one side of its reserves, about 1% of the value locked. For concentrated liquidity it is the same share of the reserve the liquidity in range amounts to at the current price, capped at the value locked, because a pool cannot hand out more than it holds. Where only the value locked is known for a concentrated pool, the constant-product figure is a lower bound and settles nothing, like an order book that was cut short. With that depth a pool is weighed exactly like an order-book market: backed when the depth is at least 1% of its 24-hour volume, stepping aside in a price when it is not, and its exchange gets a level and is included in price calculation by the same three checks and the same six hours. One thing differs: a pool's trades are settled on a chain, so its volume is never cut in the estimate of real volume. A pool that turns over many times what it holds is thin against the trading through it, which matters for a price; it is not a sign that the trades did not happen.

A token that trades only in pools (since 0.27) has no exchange to name it. It becomes a listed coin when, taken over all its pools that are read, they hold at least $100,000 and traded at least $50,000 in 24 hours, and that has been so at every hourly check for 24 hours in a row; one hour below either figure and the count starts again. Its ticker must be one no listed coin carries, and its name must neither be a listed coin's name nor a listed coin's ticker: a token under a taken ticker or name is either that coin's copy on another chain or an imitation of it, and a pool alone cannot tell which, so it stays unlisted until an exchange names its contract. Of two unlisted tokens with the same ticker, the one with more in its pools stands. Leveraged products are left out. The coin is then identified by its contract alone, its page says that it was found in pools and that its name is its maker's own word, and each such arrival is recorded. It is priced like any coin: from its pools, when their exchange is included in price calculation, and otherwise with the confidence capped at 40 and outside the totals. Neither limit was calibrated against anything; they are a first setting. This reaches the exchanges that publish a list of their pools (Orca, Raydium, Meteora, PancakeSwap); on Uniswap and Aerodrome pools are looked up by token, so a token unknown here is never met there.

Uniswap V3 (on Ethereum, Base and BNB Chain) and Aerodrome have no list of pools of their own to ask; their pools are contracts on a chain and are read from the chain itself (since 0.25), through a public node. Which pools exist is asked of the exchange's own factory contract, for every token known here by contract against the chain's dollar tokens and its own coin, at each fee the exchange allows. The price is the pool's own state, the value locked is what the pool's address holds of its two tokens, and the volume is the sum of the trades the pool recorded on the chain over the last 24 hours, counted here: nobody reports it. Trades are counted every five minutes for the 150 pools of an exchange that hold the most; a pool below that has too little in it to be listed anyway. A public node hands out the trades of its last few hours only (about five on Base, more than a day on Ethereum), so on a chain where that is less than a day the count runs forward and is kept hour by hour: for the first day after an exchange is added, or after a pause longer than the node reaches back, its 24-hour volume is still filling up and is lower than the true figure.

Uniswap V4 keeps every pool inside one contract, which records each trade with the pool's id, the amounts, and the pool's price and liquidity after it. A pool's id is the hash of what defines it, so the ids of the pools that can be named are computed here: for each token known by contract, against the chain's dollar tokens and its own coin, at the four standard fees and without hooks. Trades of other pools (unusual fees, hooks, tokens not known here) are passed over, so the volume shown for Uniswap V4 is that of the pools named, not of the whole exchange. A V4 pool has no address holding its tokens, so the value locked in it cannot be read and its depth cannot be capped: no depth is claimed for it, its exchange stands as not measured, and its pools are not included in price calculation.

Exchanges are weighed every hour, with the figures on each exchange's page, and one changes sides only after the measurements have said so six hours in a row. Markets that do not count never set a conversion rate, and they set a coin's price only when the coin has no market that counts; such a price cannot score above 40 in confidence and the coin is left out of the market's total volume and market cap. Since version 0.7; up to 0.5 a list kept by hand decided, 0.6 judged whole exchanges.

How much of a reported volume is real. An exchange reports its own volume and nobody checks it. We keep that figure and show beside it an estimate of the real one, for every market we can measure. The reported volume is cut only where one of our measurements contradicts it:

  • The order book could not carry it. Where the orders within 2% of the best prices are worth less than 1% of the reported volume, on a market trading $10,000 or more a day, the estimate is at most 18 times those orders. On the markets of regulated exchanges 95% trade less than 18 times their depth. A pair of two dollar-pegged assets (USDT/USD, USDT/USDC) is exempt: it turns over at a single price, the same orders are traded against many times a day, and a small book carries a large volume. Which assets are pegged is measured, not listed: the dollar and USDT by definition, any other asset when at least five exchanges that count price it within 1% of a dollar. The same exemption applies where an order book decides whether a market counts. Since version 0.9. The exemption has a ceiling (since 0.28): a pegged pair that reports more than 1,000 times the orders within 2% of its best prices is treated like any thin market, its estimate is at most 1,000 times those orders, and an order book the exchange publishes only in part settles it too, because the missing part is not a thousand times the part shown. On the day the ceiling was set, the pegged pairs of exchanges with no other contradiction traded under 300 times their book, and one market reported 28,000 times its own. The figure is a judgement from those readings, not a calibrated limit.
  • The public trades do not show it. We read each market's recent trades about every half hour. Once a day has been read at least 24 times over at least 20 hours, and those trades amount to under a fifth of the reported volume, the estimate is what the trades amount to.
  • Nothing has traded. A market that reports a million dollars a day or more and whose newest trade is over 30 minutes old, or any market whose newest trade is over a day old, gets what its last trades amount to over the time they took.
  • The trades are not against the order book. A trade happens at the best bid or the best ask. Where over 80% of the trades of the last two minutes were priced strictly between the two, only the share that was not counts.

Each of these occurs on under 2% of the markets of regulated exchanges with ordinary bank deposits, which is where the limits were set. Three more things are measured and named, but cut nothing, because they are patterns and not proof: buys and sells alternating like coin flips (on a live market they come in runs, as orders sweep the book), trades at regular intervals (people trade in bursts), and trade sizes whose first digits are far from the natural distribution. A market that shows them is marked as showing automated patterns. Many small independent traders can produce the first pattern too.

What this cannot see: a large market on a doubtful exchange can have a deep order book and trades at the best prices, and then nothing contradicts its volume, inflated or not. Read an exchange's estimate together with its typical market and with how many of its markets show automated patterns. Since version 0.8.

The estimate is the weight (since 0.31). In a coin's price each market that counts weighs what it is estimated to trade: its reported volume, or the lower estimate where one of the measurements above contradicts it. A coin's 24-hour volume and the total volume of the market are sums of the same figures, so they are estimates of real volume, never above what was reported; a coin's page shows the reported sum beside it where the two differ. Where nothing was measured (no order book and no trade feed read, a market under $10,000 a day, a pool) the reported volume stands. Up to 0.30 prices and volumes used reported volumes, and a market whose public trades showed a twentieth of its reported volume moved a price with all of it. On the day of the change the total fell by 11%. An exchange's own 24-hour volume in the list of exchanges remains what it reports, with the estimate in the column beside it.

An exchange's 24-hour volume (since 0.32) is the sum of what all its markets report, whether or not they are included in prices. Up to 0.31 a market set aside (for instance one whose order book is under 1% of its volume) was missing from the exchange's figure, which stood below the exchange's own: by a tenth on Binance, by half on HTX. The volume on the markets included in prices is shown beside it on the exchange's page. Markets of tokenized stocks, stock indices and currency pairs are summed apart and are not part of the figure, as they are not part of the market's totals.

A fiat currency's dollar rate comes from the markets that trade it against dollars in either direction (EUR/USDT, USDT/TRY). Some exchanges list a token that merely shares a currency's code; such a market alone sets the rate only when it trades more than $100,000 a day. A currency that no counted exchange trades against dollars at all (the ringgit, the New Zealand dollar, the lari) takes its rate across the Bitcoin, Ether and USDC markets quoted in it (since 0.33): the price of BTC/MYR against the dollar price of Bitcoin, weighted by volume, once those markets trade $10,000 a day. Before that such markets had no dollar value and were missing from their exchange's volume.

Price confidence

Every price carries a confidence score from 0 to 100 and the seven parts it is made of. The score is their weighted average; a part that cannot be measured is left out and the rest re-weighted. 80 and above is high, 55 to 79 medium, below that low.

PartWeightWhat it measuresScale
Agreement25Cross-market dispersion: the volume-weighted mean distance of the counted markets from the price.100 at 0%, 0 at 2% or more
Venues20Number of exchanges with a counted market.1 = 20, 2 = 50, 3 = 70, 4 = 85, 5+ = 100
Volume1524h volume of the counted markets, as the exchanges report it.0 at $10k, 100 at $10M, logarithmic
Depth15Value of the bids within 2% of the best bid and of the asks within 2% of the best ask, summed over the counted markets whose order book we read.0 at $5k, 100 at $5M, logarithmic; left out until the books behind half of the price are read
Bid/ask spread10Volume-weighted median of the bid/ask spreads the exchanges publish with their tickers.100 at 0%, 0 at 2% or more; left out when unknown
Concentration10Share of the largest exchange in the counted volume.100 up to 50%, 40 at 100%
Freshness5Age of the oldest counted price.100 up to 90 s, 0 at 3 min

Volume can be inflated by an exchange; resting orders are harder to fake, which is why depth is scored separately. Order books are read market by market, far less often than prices: one round over an exchange takes up to about 45 minutes, and a reading counts for three hours. Markets with less than $5,000 of daily volume are not read.

Exchanges publish a limited number of price levels. On a liquid market with a fine price step those levels end before 2% from the best price; the depth we can see is then a lower bound and is shown with ≥.

"Observed price range" on a coin's page is the highest minus the lowest counted market price. It is not a bid/ask spread; that is shown separately.

The price confidence formulas date from methodology version 0.2. Version 0.1 had no order book depth and judged liquidity by volume alone (weights: agreement 25, volume 25, venues 20, spread 10, freshness 10, concentration 10).

Volume and changes

24h volume is the sum over the markets that count towards the price. The 24h change is the volume-weighted change reported by those markets; the 1h, 7d, 30d and 1y changes are computed from our own price history, which is recorded every five minutes.

Market cap and supply

Every supply figure says how it is known. Measured: read from the token's contract on its blockchain. Reported: published by an exchange or by LiveCoinWatch and not verified by us; these figures can lag. A reported supply is taken only for a coin whose ticker and price both agree with the source, and never replaces a measured one. Where several sources speak, the order is: the chain itself, the coin's issuer, the coin's own project, Gate, Binance, LiveCoinWatch, a figure entered by hand. Gate stands above Binance since 0.17: checked against a third reference, Gate's figure was within 5% for 70% of the coins and Binance's for 27%, Binance's lagging far behind on tokens that are still being unlocked. Prices and volumes are never taken from anyone. Entered: typed in by hand from public figures for a handful of coins, approximate. Derived: computed by us from the others. Estimated: a stand-in, see below.

An exchange's own token is a special case, because the issuer can hold most of it. WhiteBIT publishes for WBT how many units are on the market and how many sit in its own funds, and calls the two together circulating: about 293M, of which its funds hold 245M. Units an issuer holds itself are not in circulation in the sense used for every other coin here, so the circulating supply of WBT is (since 0.19) the units on the market, read from WhiteBIT's own figures and marked reported, and the coin's page shows the funds beside it. Before this WBT stood tenth by market cap. The split is the issuer's statement: the funds' addresses are not published and cannot be checked on-chain.

A dollar token's units are all in circulation, so for USDT, USDC, DAI, USD1, United Stables (U), FDUSD, USDG, RLUSD, the euro token EURC and the yen token JPYC the circulating supply is the sum of the contracts' supplies over the chains the issuer mints on, measured. A coin that shares such a ticker but is priced more than 5% from a dollar is another coin and takes nothing from those contracts. Chains that are not read hold under 2% of USD1 and U; FDUSD is summed over Ethereum, BNB Chain and Solana, about a tenth short of what is issued. EURC and USDG are summed over the contracts their issuers publish (Circle, Paxos) on the chains that are read; USDG's units on Robinhood Chain, about a fifth of it, are not. RLUSD is summed over Ethereum and the XRP Ledger, where a token's supply is what its issuing account owes; its contract on five further chains holds under a hundredth of a percent and is not read. JPYC is summed over the four chains its issuer lists (Ethereum, Polygon, Avalanche, Kaia).

Any other token that has held one unit of a currency for seven days (its highest price within 6% of its lowest, now within 1.5% of a dollar, or of a currency whose code its ticker carries) and that at least two included exchanges trade gets the same measurement where no source reports a circulating supply for it: the sum of its contracts' supplies on the chains where a contract is known. It is marked measured, and it has two limits. Units on a chain that is not read are missing from it, and units the issuer has minted but not put out are in it. For that reason it never replaces a reported figure, and it is not taken where it would exceed 20,000 times the token's daily volume.

Tokenized stocks, funds, metals, stock indices and currency pairs trade next to coins on some exchanges. They follow the price of something else, so they have a list and a ranking of their own and are not part of the market's total market cap and volume. They are recognised by how their issuers name them (xStock, Ondo Tokenized, a company's legal form, a fund house's name), by the tickers those names give away with an issuer's suffix (NVDAX, NVDAON, NVDAG), and, for a bare stock ticker, by carrying the very name such a token has. A name alone is never enough. Metals are those quoted under their own code (XAU, XPT); a token backed by a metal stays a coin. Widened in 0.14: until then about 130 stocks and funds stood in the coin ranking. Since 0.21 a token also counts when it carries the same company's name as a recognised one, once the issuers' own words are set aside (“Micron Technology xStock”, “Micron Technology,”), on a ticker of the same root (MU, MUG, MUX, MUON, MUB), and costs a dollar or more or within 15% of the recognised one; about 75 more were found this way. A memecoin named after a company, at a fraction of a cent, is not one.

Coins are ranked by market cap in two groups. First those with a known circulating supply. Then, by estimated market cap, those without one; they are marked with ~. Coins with neither come last, by volume. An estimate never outranks a known figure, and the market's total market cap and Bitcoin's share of it are summed over known circulating supplies only (since 0.14; before it estimates added about a tenth to the total).

A wrapped, staked or bridged token (Wrapped Bitcoin, Lido Staked Ether) stands for another coin held somewhere else. It has a price and a rank of its own, but its market cap is that other coin's counted a second time, so the market's total and Bitcoin's share of it leave it out (since 0.17; about $42B at the time). Such tokens are recognised by their name.

Market cap is price times circulating supply. Where no circulating supply is known we use the measured total supply instead. That overstates the value, because it includes locked and unissued tokens, so the figure is marked with ~ and labelled estimated. An estimate more than 20,000 times the daily volume is dropped as meaningless. Fully diluted value uses the maximum supply where there is one and the total supply otherwise, and says which.

Dollar stablecoins issued on several blockchains are summed over the chains they are issued on, and that measured sum is their circulating supply. The native coin of a blockchain never takes its supply from a wrapped copy on another chain, and a wrapped token is never merged with the coin it wraps.

For the largest blockchains the supply of the chain's own coin is measured on that chain twice a day: Ethereum from the balances on its execution and beacon chains, Solana from its own supply figure, Bitcoin, Litecoin and Bitcoin Cash from the coins issued so far, Cardano from its epoch totals (Koios), Stellar from the foundation's account of lumens, TRON from Tronscan's account of issue and burns. A reading more than 30% away from the figure we hold is rejected. Other native coins keep an entered figure.

A reported circulating supply comes from LiveCoinWatch and is taken only for the same coin: the same ticker and the same name with prices that agree, or, where the two sides name a coin differently, a price within 0.5% for a coin that trades at least $1M a day and has no second candidate. A coin the reference does not value gets none. Where its figure is known to be wrong and no other source gives one, the coin is left without a circulating supply and a market cap rather than ranked by it: KuCoin Token (KCS), 75.6M against about 139.7M from KuCoin's own burn notices (since 0.43).

Many small mined coins have one source only: the project's own block explorer or an address of its own that answers with the supply. It is read for the coin an exchange lists under that ticker, at the address that exchange's asset list names (NonKYC's) or one found by hand on the project's site, and shown as reported by its project. An address that belongs to another price tracker is not read, a coin whose price rests only on excluded exchanges gets no figure, and none is taken that would put the coin's market cap above 20,000 times its daily volume. Since 0.39 it stands above the exchanges' figures and replaces them: Gate's is a market cap over Gate's price, and for tokens that are still being unlocked it lagged the project's own count far behind (peaq at 133M against the 2.58B its explorer reports, Walrus at half, Aptos a third above the Aptos Foundation's figure). Where both spoke for the same coins otherwise, they stood within 3% of each other.

Some projects publish no figure but name the addresses that hold what does not circulate: treasury, team and investor vesting, foundation, ecosystem funds. For those the circulating supply is measured: the token's total supply on the chains that issue it, less what those addresses hold now, read from the chains at every pass, so that it follows each unlock. The coin is found by its token contract, not by its ticker; a bridged copy whose units are locked on another chain is not counted twice. Where the project names the addresses (its documentation, or the list it gave a price tracker's supply disclosure) is kept beside each entry. It is shown, like the project's own figure, as reported by its project, and stands where that figure stands. Since 0.42.

Binance publishes a circulating supply for the coins it lists. Gate publishes a market cap; divided by Gate's own price it says how many coins that cap is over, and that number is taken as Gate's reported circulating supply (since 0.15). It is not taken where the cap is simply price times total supply, where the result would exceed the total, where it would put the market cap above 20,000 times the coin's daily volume, or where the coin's price stands more than 15% from Gate's own (since 0.16: a coin priced on another exchange far from Gate's price is not the asset Gate's figure is about). Both are taken only for coins that trade on that exchange, by its own listing. Gate also files what it lists under classes; a coin it files under stocks, under the name Gate gives it, is treated as a tokenized stock.

Exchange quality

An exchange is rated by what its own market data shows and, since 0.29, by how visited its website is. This is not a rating of custody, licences or solvency: we have no data on those, and we do not guess. A part that cannot be measured is left out and the rest re-weighted; without liquidity and integrity there is no score.

PartWeightWhat it measuresScale
Liquidity40Three figures, weighted 4 : 3 : 3. The typical spread: the plain median bid/ask spread over its counted markets (where its tickers carry no best prices, since 0.30 the spread seen in each market's order book when it was last read; before that such an exchange was scored on depth alone). The spread where the money trades: the same median weighted by volume. Depth to volume: its order book depth within 2% of the best prices as a share of the same markets' 24h volume.Typical spread: 100 at 0.05%, 0 at 1%. Largest markets: 100 at 0.002%, 0 at 0.1%. Depth to volume: 0 at 1%, 100 at 50%. All logarithmic
Integrity30Mean of three: the share of its traded markets priced too far from the others, the volume-weighted distance of its counted prices from ours, and the share of its listings with no trade in 24 hours.Outliers: 100 at 0%, 0 at 5%. Distance: 100 at 0%, 0 at 0.15%. Dead listings: 100 at 0%, 0 at 20%
Reliability20Share of our reads of its tickers that succeeded in the last 24 hours.0 at 95%, 100 at 100%; left out with under an hour of data
Breadth10Number of markets that count towards a price.0 at 1, 100 at 1,000, logarithmic
Website visits30The band of the Chrome UX Report's ranking of the million most visited websites in which the exchange's domain stands (the best of the hosts under it). Not applied to decentralised exchanges or to exchanges that pass on another's markets. Applied to exchanges of every size since 0.30 (0.29 left out those under $10M a day, and the smallest unvisited exchanges topped the score).100 among the top 10,000; 90 top 50,000; 80 top 100,000; 65 top 500,000; 40 top 1,000,000; 0 when not listed

Depth to volume is the most telling figure. Volume is whatever the exchange reports; resting orders have to be there. A large volume on a thin book is what inflated volume looks like.

Two cautions. Where an exchange's public order book is short, we see only part of its depth and its depth to volume is understated. And an exchange with very high turnover has, by this measure, less depth per dollar traded than a quiet one with the same book.

Website visits (since 0.29) are the one part that is not the exchange's own data. Spread, depth, prices and trade feed are all published by the exchange, and an exchange that copies a large one's order book scores well on every one of them: before 0.29 several exchanges nobody is seen visiting stood above Coinbase and Kraken. Visitors cannot be written into an API. The ranking is Google's Chrome UX Report (CC BY-SA 4.0), read from its monthly public list of the million most visited origins; it gives bands, not exact places. It has known blind spots: it counts Chrome only, not mobile apps or API clients, so an institutional venue or one serving a country where Chrome is rare scores lower than its trading deserves. The weight and the points per band are a judgement, not a calibration. Website visits enter no price and do not decide which exchanges count.

Scales were tightened in version 0.3: under 0.2 every large exchange scored above 90 and the score told them apart only from a very small one.

Red flags

The pages of exchanges and coins show red flags. A flag states what the measurements above say against an exchange or a coin, with the figures; it changes no price and no score, and none is entered by hand. Flags are evaluated once an hour. One is raised after its condition has held six hours in a row and lowered after six hours without it. Since version 0.9.

  • Exchanges with at least $1M of judged volume, one flag for each contradiction of reported volume (order book could not carry it; public trades show less than a fifth; no trade in 30 minutes, or in a day; most trades inside the spread) when the markets that carry it hold 25% or more of the exchange's judged volume. Also: prices off the market (median distance from Coinsprey prices above 1%), and data often unavailable (under 95% of reads succeeded in 24 hours). Since 0.29 also volume without visitors: $100M or more reported a day by an exchange whose website is not among the million most visited (see website visits); the one flag that rests on a figure not measured by Coinsprey. Since 0.30 also volumes that follow another exchange's in a fixed proportion (see are its volumes its own). Since 0.34 also two different volumes published: where an exchange publishes its markets in a second public list beside its trading API, both are read once an hour and compared market by market (at least 10 markets of $10,000 a day or more); a typical market reporting 1.5 times or more in one list than in the other is flagged, with the proportion. The figure shown remains the trading API's. Only exchanges whose second list is known are compared.
  • Coins trading at least $1M a day: priced on one exchange; priced only by exchanges not included in prices; price confidence under 40; estimated real volume under 25% of reported, over judged markets that carry at least half of the coin's volume.

A flag is a finding to weigh, not a verdict: a measurement can be wrong, and the absence of a flag proves nothing.

Fraud Signals are separate and are not computed: the records are entered by an editor, never automatically, and a record is not accepted without a link to its evidence. Each is marked alleged, confirmed or resolved. Readers can report cases through a form; a report is read and checked and publishes nothing by itself. Neither flags nor records enter prices or scores.

Mining estimate

The mining calculator is the one place that uses figures not measured by Coinsprey: block reward, block time and network hashrate are reported by WhatToMine and read once an hour. A network is tied to a coin only when the ticker is the same and the reference's price is within 15% of the Coinsprey price; otherwise it is listed without a price.

coins per day = hashrate ÷ network hashrate × block reward × 86 400 ÷ block time; revenue = coins per day × Coinsprey price; electricity = power draw (kW) × 24 × price per kWh. The estimate assumes that difficulty and price stay as they are; pool fees and hardware are left out.

History

Prices and their confidence are recorded every five minutes. The markets behind every price, with their weights and the reasons for exclusions, are recorded once an hour and kept for 90 days, so a past price can be traced to what it was made of (/api/v1/coins/{id}/sources?at=…).

What a coin is

A ticker is not an identity: exchanges reuse tickers for different coins and list one coin under different tickers. Every coin here has an asset id that never changes, and for every exchange we record which asset its symbol means. A symbol keeps meaning that asset whatever happens to the ticker elsewhere.

Two tickers with the same token contract and the same price are merged into one coin. A ticker that an exchange uses for a different coin is given its own asset, so the two are priced separately. This is recognised by price: when every market of that ticker on an exchange that traded in the last 24 hours stays at least five times above or below a price that two or more other exchanges form, for six hours in a row, they are not markets of that coin. One market far off beside others of the same exchange that sit on the coin's price is a broken market, not another coin, and moves nothing (since 0.13; for a few hours before it one such market could take the whole ticker along, and those moves were undone). The symbol then joins another coin of the same ticker whose price is within 15% of its own, or becomes a coin of its own, named after the exchange until an exchange that trades it names it. A smaller distance is treated as a price gap of the same coin, with two exceptions. Several exchanges together (since 0.37): when the markets of a ticker on three or more exchanges stand twice or more away from the coin's price, within 10% of one another, while three or more other exchanges form that price, they trade another asset under the ticker and leave together after the same six hours. One exchange two or three times off can be a thin or a stale market; several that agree on that price cannot. Both limits are our own choice and not calibrated. Since 0.38 such a group becomes one asset whatever each exchange calls it (EDGE, EDGE1, EDGEBASE), made under the ticker most of them use; and a coin a split left without a name of its own joins a coin whose ticker contains its own or is contained in it (at least three letters) when the two stay within 1% of each other for six hours, the one fewer exchanges trade joining the other. Since 0.38, too, a market that reports under $100 in 24 hours proves nothing either way: it is set aside as dust in the price and is never made a coin of its own. A stale market drifts anywhere (an exchange's XMR market stood at $73 against $550), and before this each such drift became a “coin”; those made that way were put back. And (since 0.20): what an exchange itself calls the asset. Seventeen exchanges publish a name for each ticker they list. Where an exchange's name for its asset is none of the names that the exchanges behind the coin's price give it (at least two of them must name it; a different spelling, a shortened name or a name one of them also uses is not another name), a distance of 20% is enough: LBank's WALLET is “Robinhood Wallet” at $0.019, the coin others price at $0.012 is “Ambire Wallet”. Where the data cannot tell two coins apart, this is decided by hand. Since version 0.11; before it, such a market counted as a price far off and was held against the exchange. Addresses are readable (btc-bitcoin) and follow the name, so they can change; old ones redirect.

A coin's address is its ticker and its name together, so two coins sharing a ticker never share an address (qtc-qitcoin, qtc-quantus). For that to mean something, the name has to sit on the right coin. A name comes from an exchange that trades the coin and counts in its price. Since 0.17 it belongs to that asset alone: another coin of the same ticker that carried it (taken earlier from an exchange's list by the ticker) goes back to its bare ticker until something that knows it names it, and an exchange that lists the ticker without counting in the coin's price no longer names the coin. Where two coins of a ticker are traded under one name, it stays with the one more exchanges trade.

Names come first from the exchanges whose own API gives them (Binance, Gate, KuCoin, Coinbase, MEXC, Cexius). Since 0.20 the asset lists of eleven more exchanges are read as well; those are kept less carefully, so they name only a coin that has no name yet and never replace one. A coin nobody trading it names may still take the name another exchange's list has for its ticker, which is a guess by ticker alone. Since 0.20 that guess is not made with a company's or a fund's name (a stock's ticker is very often some coin's ticker too: a token trading at $0.70 had stood among the tokenized stocks as “RTX Corporation”), nor against what an exchange listing the coin calls it.

Exchanges' lists keep old names for years, so a coin's name does not simply follow the first list that has it. Since 0.22 it follows the exchanges behind its price: when at least three of them use one name, and at least twice as many as use the name the coin carries, the coin takes that name (“Crypto.com Coin” became “Cronos”, as seven exchanges against two call it). A bare majority renames nothing: five lists against four is as often a misspelling or a name already dropped. The two dozen coins named by hand are left alone.

One asset under two tickers is made one coin (since 0.22). Exchanges list the same coin under tickers of their own (TRUMPSOL beside TRUMP, MASK2 beside MASK). Two coins are one when they carry the same name, bracketed remarks aside and nothing looser than that, and their prices stay within 1% of each other for six hours in a row; the one fewer exchanges trade joins the other and its address redirects. Tokenized stocks are left out (one company, several issuers' tokens), and so are wrapped and staked tokens: a copy is never merged into its original. Each merge is recorded.

News

News are stories from crypto outlets. A language model running on our own hardware decides whether a story is relevant and writes a one or two sentence summary; the headline is the outlet's and links to the original. A summary can contain mistakes. The same event reported by several outlets is shown once.

What we do not do

Advertising is confined to four marked places and never touches the data. No affiliate "buy" buttons, no paid listings or placements. A coin appears here because it trades on an exchange we read, and for no other reason.