> For the complete documentation index, see [llms.txt](https://docs.llama.fi/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.llama.fi/exchanges/methodology.md). # Exchanges Methodology This page describes every number published on the [Exchanges dashboard](https://defillama.com/exchanges), what it means, and the corrections applied before it is shown. The REAL score on the [rankings page](https://defillama.com/exchanges/rank) is covered at the end. Two principles run through all of it. **We store what the exchange said, verbatim.** Nothing is corrected on the way in. Every adjustment below happens when the number is read, from a declared rule, so the original record stays auditable and a rule can be revised without rebuilding history. **A missing measurement is never scored as a bad one.** Where something has not been measured it reads as unavailable, not as zero. *** ## What We Track Each exchange is tracked per market segment. Segments are never merged into a single headline without being named, because they are different instruments with different notional conventions. | Segment | Description | Normalized volume | | ------------------------ | ---------------------------------------------------------------------- | ------------------------------------------------------ | | Spot | Spot markets, quoted and settled in the quote asset. | Yes, against the order book | | USD-margined perpetuals | Linear perpetual futures, margined and settled in a stablecoin or USD. | Yes, against the order book and open interest | | Coin-margined perpetuals | Inverse perpetual futures, margined and settled in the base asset. | Not yet; reported volume, open interest and depth only | *** ## Corrections Applied to Everything These run before any metric below is computed. ### Units are declared, not guessed Exchanges report volume, open interest and order book size in whatever unit they please: base coin, quote currency, USD, or contract count. The unit is declared once per exchange, market type and contract type, then applied at read time. A contract's face value is read from the venue's own contract size, so inverse contracts, whose face value is quote-denominated, and linear contracts, whose face value is base-denominated, convert correctly rather than merely plausibly. ### Double-count correction Some venues report both sides of a trade as volume, or both sides of a position as open interest, so a $1M trade reads as $2M. Where that convention has been measured, a multiplier per venue and data type corrects it. Measurement is against the venue's own API documentation, per-pair ratios from independent aggregators, and trade-delta tests on low-activity markets. The default is 1: no correction is applied without evidence. ### Volume is priced at VWAP, not close A day's volume is valued at the volume-weighted average price of that day, falling back to the closing price where an hourly series is not available. Pricing a full day at its closing price misstates any day the price moved, and breaks outright across a token redenomination. Open interest continues to use the close, because it is a point-in-time stock and is correctly valued at the boundary. ### Quote currencies resolve to USD Every non-USD quote is converted through a rate derived from our own market data rather than a single external feed. ### Shared books are counted once Some exchanges operate a regional legal entity that trades on a parent's order book rather than its own. Those segments appear on their own venue page, because they are real and a user trading there gets that liquidity, but they are excluded from cross-venue totals where counting them would double the same book. *** ## Price The last traded price from the venue's own candle, published per market and used as the valuation price for open interest. It reads as unavailable when there is no candle inside the freshness window. *** ## Reported Volume Traded notional in USD, as the venue reports it, after unit resolution, double-count correction and VWAP pricing. It is single-counted: where a venue counts both sides of a trade, the figure is halved. On the dashboard and on each exchange page the headline is the **last 24h rolling** figure, refreshed through the day. History is stored per UTC day and finalised after the day closes, so a daily bar and the rolling headline will not match exactly at any given moment. This is the exchange's own claim. It is published because it is what the industry publishes and comparison requires it, not because we vouch for it. The figure that reflects what can be verified is normalized volume, below. *** ## Normalized Volume The question it asks: could this venue's own market plausibly have carried the volume it reports? Two things a venue cannot print are used to answer it: the liquidity resting on its order book, and, on USD-margined perpetuals, the open positions it holds. Wash volume needs neither. It does not consume resting depth and it does not require margin, so it can inflate volume without moving either denominator. Real trading moves both. ``` spot: normalized volume = min( reported volume, active liquidity × β ) USD-margined perpetuals: normalized volume = min( reported volume, active liquidity × β, open interest × k ) ``` Whichever cap is lower is the one that applies. They never compound: a market is capped by its book, or by its open interest, or by neither, and which one bound it is published per market. ### The order book cap: β β is the recycling limit, meaning how many times in a day a real order book can legitimately turn over. It scales with the market's own volatility, because a market that moves more genuinely does recycle its book more. ``` lift = 1 + √(daily move / 4%) β = α × lift α = 18 on USD-margined perpetuals, 16.2 on spot, tripled when both legs of the pair are pegged ``` On a typical 3% day the budget is roughly 34 turns of the active book. Pegged pairs get triple headroom because they are an exchange's conversion and arbitrage rail. The same dollar of depth turns over far more times a day than a BTC book does, with no price risk to slow it down, and being pegged they collect no volatility lift, so they would otherwise be judged at the strictest possible β for doing exactly their job. ### The open interest cap: k k is how many times a day traded volume may turn over the market's single-sided open interest. It uses the same volatility lift as β, because day traders open and close more positions on a volatile day. ``` k = 10 × lift ``` On a typical 3% day the budget is roughly 19 turns of open interest. The large books turn theirs over 1 to 5 times a day, and the 90th percentile across every perpetual venue we track is under 6, so the cap sits well clear of organic intraday flow. Open interest here is the single-sided figure after the double-count correction: one contract is one long and one short, counted once. Why the book alone is not enough: volume divided by depth cannot single out a venue whose book we capture in full while the largest venues' books are truncated by their own public endpoints. The largest venues turn over their captured book 20 to 30 times a day, so tightening β would cut them first. Volume divided by open interest has no such bias, and it separates the population cleanly. ### What is published * **Per market:** reported volume, normalized volume, active liquidity, the band and coverage the liquidity was read at, and `cap`, which names what bound the figure: `depth`, `oi`, or nothing when normalized equals reported. * **Per exchange and segment:** the sum of its markets' reported and normalized volume, the discount between them, and the share of reported volume that was judged against estimated rather than fully captured depth. This figure appears on the exchange page beside the reported volume and in the normalized volume columns of the dashboard. * **History:** a daily series of reported and normalized volume per segment, drawn on the exchange page's Volume chart. Normalized history starts on the first day a depth snapshot exists for the venue, which is later than reported volume for most venues, and the chart marks that day. The gap is always shown as a share of reported volume that did not pass, never as a signed change. Normalized volume is by construction never above reported. ### Where there is no book Volume is only discounted where there is something to check it against. A market with no usable depth snapshot passes its reported volume through untouched, flagged as unverified rather than punished, and the open interest cap does not run on it either. An exchange segment where no market at all has a usable book shows no normalized volume, rather than a "normalized" figure that is just the reported one copied over. A quote-request ladder synthesised from indicative prices is not treated as a book. Coin-margined perpetuals have no normalized volume yet: their notional conventions differ enough that the active liquidity band is not defined for them. {% hint style="info" %} A large discount is not a wash-trading accusation. It means reported volume is large relative to the visible book or to open positions, which can equally mean the exchange's real liquidity sits somewhere our snapshot did not reach, or that its flow is dominated by very short-holding intraday trading. {% endhint %} *** ## Active Liquidity The order book depth a day's trading could actually use, based on observed depth only. Rather than reading depth at a fixed ±2%, the band tracks the market's own daily price movement, using the trailing typical daily range or today's move if larger. On a market that swings 10% in a day the liquidity that matters sits far from mid, and a fixed ±2% would read as artificially thin. The band is capped at ±25%, the widest band stored. A credibility limit applies to the part of the book sitting far from mid, described under [Credibility adjustments](#credibility-adjustments) below. *** ## Order Book Depth Cumulative USD resting within a fixed distance of the mid price, one snapshot per market per day, with bid and ask kept separate. Depth is stored at eleven bands from ±0.1% to ±25%. The headline figures are ±2% and ±0.5%. Depth is one-sided by construction, so no double-count multiplier applies. It is measured, never submitted: an exchange cannot send us a depth figure. ### Why published depth is a floor We snapshot public order book endpoints, which cap how many levels they return. On a dense major market with a fine tick size, even a 1,000-level snapshot can span a fraction of one percent. Bands wider than the deepest level fetched are forward-filled, reported as the same liquidity as the deepest real level observed, and flagged as partial coverage. The effect is that published depth understates the real book, and the error runs in the exchange's favour. Reach varies enormously between venues: one exchange's perpetual book reaches only ±0.03% because its endpoint returns 100 levels no matter what is requested, while another reaches ±28%. Where live probes have measured that a venue's displayed size does not actually fill, the displayed depth is discounted. See [Credibility adjustments](#credibility-adjustments) below. *** ## Open Interest Total notional of open perpetual positions in USD, valued at the period close. It is unit-resolved and double-count corrected the same way volume is, because several venues report open interest as the sum of longs and shorts, which doubles it. A measured two-times convention is corrected; an unmeasured venue is left at one. Published for linear and inverse perpetuals separately. Spot has no open interest, and an absent inverse book reads as unavailable rather than zero. On USD-margined perpetuals, open interest is also the second denominator of normalized volume, described above, and volume that turns positions over implausibly fast is discounted in the REAL score's Activity pillar. *** ## Funding Rate The venue's current perpetual funding rate, normalized to a common 8-hour basis so that exchanges on different funding intervals can be compared. The next funding timestamp, mark price and index price are published alongside it where the exchange exposes them. *** ## Execution Quality Measured from tick-level trades and order book snapshots, which means actual fills rather than quotes, on a shared basket of perpetual markets so that venues are compared on the same instruments. Cost to trade stays on the two deepest assets, because it is the number a reader feels directly and the thinner the market the more a fee schedule rather than the book drives it. ### The short version 1. Take the order book snapshot standing immediately before a market order executes. 2. Walk that book for the order's own size. That is what it **should** have cost. 3. Compare against what the order's fills **actually** paid. The first number is published as predicted slippage, the second as realized slippage, and the difference between them is the honesty check on the book. A venue can display any depth it likes, but only one of these two numbers is a promise. ### A worked example A taker buys $100,000 of a BTC perpetual. The last book snapshot before the order prints shows a best bid of 99,995 and a best ask of 100,005, so the mid is **100,000**. Walking the ask side for the full $100,000 consumes two levels, 0.6 BTC at 100,005 and 0.4 BTC at 100,014, for a volume-weighted average of 100,008.60: ``` predicted = (100,008.60 - 100,000) / 100,000 x 10,000 = 0.86 bps, or $8.60 on $100k ``` The order actually arrives on the tape as three prints within a few milliseconds: 0.5 BTC at 100,010, 0.3 at 100,018, and 0.2 at 100,029.50, a volume-weighted average of 100,016.30. Every print is measured against that same pre-order mid, because the order was sent once, against the book as it stood then: ``` realized = (100,016.30 - 100,000) / 100,000 x 10,000 = 1.63 bps, or $16.30 on $100k ``` The book promised $8.60 and the order paid $16.30. Both figures are filed in the $100k band, and both describe the same order. ### Definitions Everything below uses `mid = (best bid + best ask) / 2` taken from the reference book, and a sign convention where a taker buy is `+1` and a taker sell is `-1`, so a positive result is always bad for the taker. Figures are in basis points, meaning the ratio multiplied by 10,000. | Step | Rule | | ---------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Grouping prints into orders | Consecutive prints on the same side within **10 milliseconds** are treated as one order. A single order sweeping several levels arrives as several prints, and pricing them separately measures each against a mid it no longer traded at. | | Reference book | The last snapshot **at or before** the order's first print. If that snapshot is more than **2 seconds** old, the order is excluded rather than priced against a stale book. | | Realized slippage | Notional-weighted across the order's prints: `sign x (fill price - mid) / mid`. | | Predicted slippage | Walk the same reference book, on the side the taker consumed, for the order's full size, to a volume-weighted average price: `sign x (vwap - mid) / mid`. | | Size bands | $1k, $10k, $50k, $100k, $250k, $500k and $1M. An order counts toward a band only if its total notional falls between the label and **20% above** it, so an order sitting between two bands is published in neither and each column describes that size rather than a range. | | Orders that cannot be walked | If an order is larger than the visible levels of the reference book it is dropped from **both** columns. Admitting its realized cost against a predicted figure that excluded it is what makes deep books look like they quote what they cannot fill. | | Effective spread | Per print rather than per order, which is the standard convention: `2 x sign x (fill price - mid) / mid`, notional-weighted. | | Markouts | `sign x (fill price - mid at T+h) / mid at T+h`, for h of 1 second, 15 seconds and 1 minute. The book snapshot nearest to T+h is used, and only if it falls within **2 seconds** of it. | | Quoted spread | Time-weighted across the hour, each book state weighted by how long it stood. A state's weight is capped at **5 seconds**, so a silent feed is not averaged in as though the book still existed. | | Aggregation | Notional-weighted within each hour, combined across the trailing window, then taken as an equal-weight median across the markets in the basket. | ### The published metrics | Metric | What it measures | | ---------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | Cost to trade | All-in taker cost: base-tier fee plus half the effective spread. The only fee-inclusive figure, so VIP tiers and fee-token discounts can lower the fee part below the base tier shown. | | Effective spread | What trades actually paid: twice the distance from the mid price to the fill. Excludes fees, and is pooled over whatever sizes traded, so it includes each trade's own impact. | | Quoted spread | What the book displayed at the touch, time-weighted. A promise rather than a fill. | | Slippage by size | Published at $1k, $10k, $50k, $100k, $250k, $500k and $1M. Each band is narrow, counting an order only if it is within 20% above the label, so the number describes that size rather than a range. Both the predicted cost, from walking the pre-order book at that size, and the actual cost are published. | | Markouts | Where the price moved 1 second, 15 seconds and 1 minute after a fill. Positive means the price kept moving against the trader, which indicates trading against faster, better-informed flow. | | Feed speed | Average gap between book updates. Displayed but never scored, because emission policy is not engine latency: a venue publishing on a fixed interval is making a choice, and a quiet venue updates less because nothing changed. | Throughout, a positive number is bad for the taker. ### Predicted versus actual is the point Predicted slippage is deliberately not corrected for fake depth. The gap between what a venue's book promised at a given size and what orders of that size actually paid is the measurement that exposes a book which does not deliver, and normalizing the promise would erase the signal. On one venue, $100k fills were measured paying about 2.6 times the cost its own book implied. ### What this does not model **Speed bumps and latency are not accounted for.** The baseline is the book standing immediately before the order printed, because the moment an order was submitted is not observable from public market data. On a venue that deliberately delays orders before matching, or where a taker's own latency is material, the book the taker saw when sending is not the book the fill is measured against, and the resulting slippage will include whatever the market did during that delay. This is a limit of measuring from public data rather than a judgement about the venue, and it applies to every venue equally. Also outside the measurement: queue position, order types beyond the aggressor side, cancellations and orders that never fill, and any flow that does not reach the public tape, such as RFQ or internalised trades. ### How much data a market needs Every execution figure, whether it comes from fills (effective spread, realized slippage, markouts) or from the book (quoted spread, predicted slippage), is built from the same tick coverage and carries the same minimum. | | | | ------------------ | --------------------------------------------------------------------------------------------------------------------------- | | Minimum per market | **72 hours of tick coverage**, meaning 3 complete UTC days | | Measured over | a trailing 30-day window | | Below the minimum | the market is shown on the exchange's own page, but does not count toward the exchange's published median or its rank score | | No coverage at all | the metric reads as unavailable for that exchange | A newly tracked market therefore becomes rankable on its third full day. Because a 3-day market sits alongside a 30-day one in the median, an exchange's figures can move as a young market matures. *** ## Cost to Trade and Ramp Cost What it actually costs to move money through an exchange, end to end. **Cost to trade** prices a BTC market order against the venue's own book: the taker fee plus slippage walked through the real depth ladder, at $1k and $10k. **Ramp cost** prices the full round trip, per fiat currency and per region: * On-ramp: fiat deposit, buy BTC paying the taker fee and book slippage, then withdraw BTC paying the venue's withdrawal fee. * Off-ramp: deposit BTC with the user paying the network fee, sell BTC paying fee and slippage, then withdraw fiat. Slippage comes from the stored depth ladder, walked band by band. Liquidity is assumed uniform inside each band and the order consumes the part nearest mid first, so an order filling a fraction of a band pays proportionally rather than being charged the band's midpoint. The result is floored at the half-spread, because a first fill can never be better than top of book. A leg reads as unavailable unless every component is known. An unpublished fee is never treated as zero. Because these walk the same normalized depth ladder as everything else, a venue with a measured fake-depth finding is priced against its discounted book rather than its displayed one. *** ## The REAL Score One composite from 0 to 100 over four pillars, summarised as a tier. The pillars give the score its name: | | Pillar | Weight | The question it answers | | ----- | --------- | ------ | -------------------------------------------------------------------------------------------- | | **R** | Reserves | 30 | Is there a balance sheet behind this venue, and does the rest of its business fit inside it? | | **E** | Execution | 30 | What does a trade actually cost here, and how do fills behave afterwards? | | **A** | Activity | 20 | How much of the reported business survives verification, and does it hold steady? | | **L** | Liquidity | 20 | How much resting depth is really there, across how many markets? | The composite is a weighted mean, and weights renormalize over whichever pillars apply, so a missing input is treated as not applicable and never as a zero. ### R: Reserves (30) The one number a venue cannot print, which is why it carries the most weight alongside execution. **Magnitude, 20 of the 30.** Clean reserves, meaning holdings in the venue's own token are excluded, credited by evidence tier: live on-chain wallets receive full credit, regulator-filed audited financials slightly less, third-party attestations or company disclosures less again, and any figure more than a year old drops a tier. Log-scaled from $100M upward, so the largest balance sheet sets the top and a venue a tenth its size still scores well. For a decentralized exchange, collateral is read on-chain from the protocol's tracked value and credited at a multiple of it, because that value is a floor rather than the whole balance sheet: margin also sits in vaults, on other chains, and with market makers. **Consistency, 10 of the 30.** Four reconciliations, each scoring full when the venue's figure is a size its balance sheet could support and ramping toward zero past the range established venues occupy: resting depth against reserves, daily volume as a turnover of the custodied base, open interest against reserves, and custody quality, meaning the clean share and how concentrated it is in the venue's own token. A venue that publishes nothing where all of its peers publish scores zero here rather than being excused. ### E: Execution (30) Measured from the tick-level data described above, on a fixed basket of perpetual pairs identical for every venue, with at least 72 hours of coverage per pair. Scoring is **peer-relative on the same pair**: a venue's BTC reading is scored only against other venues' BTC readings, placed linearly between the best peer and a robust worst anchor. Pair scores are then combined into one pillar score, weighted by each asset's open interest across all venues, so the basket that is scored and the basket that is displayed mean the same thing. | Sub-metric | Weight | What it catches | | ---------------- | ------ | ------------------------------------------------------------------------------------------------------ | | Markouts | 30 | Adverse selection: fills that systematically happen at bad moments | | Cost to trade | 20 | The all-in price of trading. Fee-dominated on majors, which is why it is deliberately not the heaviest | | Slippage at size | 20 | Size-controlled fill quality, the same trade compared on every venue | **Delivery discount.** Fills are also compared against what the venue's own book promised at that size, hour by hour. A venue whose fills come in materially worse than its displayed book loses up to 12% of the pillar, with no penalty below a small threshold and the full discount some way above it. Quoting wide and filling wide is not penalised here, because cost and slippage already charge for the price. Only quoting tight and not delivering is. **Displayed but not scored:** quoted spread, because it is a promise rather than a fill; sub-minute markouts, which are mostly noise; and feed speed, because a venue's publishing cadence is not its matching-engine latency. ### A: Activity (20) **Verified volume, 45%.** Only normalized volume is credited, meaning volume that survived the order book cap and, on USD-margined perpetuals, the open interest cap. Volume reported in markets with no visible book earns nothing, and is discounted further where a venue's non-major markets report volume but show no book. **Open interest, 35%,** on perpetual scopes. Credited where we observe the perp book, capped at three times clean reserves, log-scaled. Volume turning positions over implausibly fast is discounted, unless the venue's reserves make that turnover affordable, because high turnover on modest open interest is ordinary market-maker churn for a venue custodying a very large book. **Stability, 20%.** How steadily depth, volume share and open interest hold across the window. Note that this does not detrend, so fast growth reads as slightly less stable. ### L: Liquidity (20) **Depth, 70%.** The 30-day median resting depth within ±2% of mid, capped at clean reserves, log-scaled. A book that never moves across 30 days is discounted as decoration, and a venue publishing no reserves takes a haircut here because nothing reconciles its book. **Breadth, 30%.** How many markets hold at least $100k of depth, scaled by how much of the venue's total book its reserves could support, log-scaled. Snapshots are taken from public endpoints that under-capture the deepest books, so depth enters this pillar as a conservative floor that errs in the venue's favour. ### Tiers **S at 85 and above, A at 70, B at 55, C at 40, D below 40.** Fixed bands rather than quantiles, so a new listing does not reshuffle everyone. The S tier additionally requires a measured execution pillar, because the top grade cannot be earned from displayed data alone. A venue with no observed book at all is shown as insufficient data rather than scored low. ### How missing data is handled | Cause | Meaning | Handling | | ------ | ------------------------------------------------------ | -------------------------------------------- | | Ours | We have not built the integration or captured the data | Not applicable, and the weight redistributes | | Theirs | The venue publishes nothing where its peers publish | Scored zero | | Parked | The metric is too incomplete for anyone yet | Not applicable for every venue | Scores are never imputed from peers or from a tier average. That was tested and it both credited and punished venues for numbers nobody had measured. ### Checks that run across the pillars Every credibility check lives inside the pillar whose data it questions and lowers that pillar. There is no separate penalty layer, and no check can disqualify a venue outright. * Open interest is credited at most three times clean reserves, because positions margined by the venue are bounded by what it custodies. * Depth is capped at clean reserves, and a centralized exchange publishing none takes a haircut, since nothing reconciles its book. * A book that never moves is treated as decoration: near-zero variation in depth across 30 days discounts the liquidity pillar by up to 30%. * Where probes established that a book is substantially decorative across the whole ladder, depth credit is withheld entirely rather than capped, because whatever survives a cap is the cap rather than measured liquidity. * Execution cannot out-credit measurement. A venue with no measured execution pillar is capped on it, so missing data never beats being measured. * Perpetual-only inputs scale with the perpetual share of the business, so a spot-first exchange is judged on its spot business. Rank is a position within an exchange's own type group. Centralized and decentralized exchanges are scored separately, because the composite renormalizes away reserves and regulatory inputs for decentralized venues, which makes raw scores non-comparable across types. ### What the score does not include Reputation, age, brand, number of listings, social following, or anything self-reported that cannot be checked. A venue's score moves when its measured behaviour moves. ### Fairness rules and limits * **Our gaps are not your score.** A missing integration does not mean missing liquidity, and it is handled as not applicable rather than as a low reading. * **Unverified volume earns nothing**, and never inherits the clean reading of the part that was verified. * **Execution is only ever compared across venues on the same pair**, never across different assets. * **Regional entities are scored on their own evidence.** Grouping under a brand affects sort order only. * **Full-credit bands are ranges, not targets.** Options venues run deep books at low turnover and institutional venues run high turnover on modest open interest; both take full credit. * **Execution coverage spans a minority of venues today.** An uncovered venue has the pillar absent rather than low, and cannot reach S until it is measured. Spot-only venues are not covered yet. * **A venue that authors both its trade feed and its book could in principle produce a self-consistent clean tape.** A suspiciously clean reading is treated as something to investigate, not as a crown. * **Reserves are point-in-time at best.** A venue moving funds between snapshots is not caught here. Scores are recomputed after new listings or material data changes, reviewed, and then published. There is no unreviewed automatic re-ranking, and every venue ships its pillar decomposition so the arithmetic behind a tier can be inspected. The pillars and their weights are subject to change as coverage improves. {% hint style="warning" %} The score is not an assessment of conduct, honesty or solvency. A low pillar means weak measurable evidence, which can be a genuine weakness, an unusual but legitimate market structure, or a limit of our own measurement. {% endhint %} *** ## Credibility Adjustments Three adjustments can reduce a published figure where evidence says the raw one overstates what is really there. All of them are applied when a number is read, never when it is stored, so an exchange that is re-measured and cleared reverts immediately with no rebuild. ### Far-band liquidity limit Beyond ±2% of mid, a market's active liquidity may not be credited above five times its own ±2% depth. The near book is the part that gets arbitraged and the part every venue is compared on, so it is the part exchanges keep honest. Size parked far from mid is cheap to display and hard for anyone to hit. The multiple is set from what clean books actually do, high enough that it costs an honest venue essentially nothing, and very small readings are never capped because a ratio means little when the underlying number is tiny. ### Depth that does not fill Where live probes have measured that a venue's displayed size is not actually consumed by trading, its displayed depth is discounted to what the evidence supports. Either an absolute cap near mid, or a limit relative to that market's own trading volume, can apply. The bar is deliberately high: it takes a reproducible capture of the full book alongside the trade tape, compared against a clean-book venue recorded in the same window. A suspicious ratio alone is never enough, because honest books on quiet markets produce suspicious ratios too. These adjustments are only ever applied to a venue carrying a measured finding, never across the population. ### Volume the market cannot support Reported volume is capped against the liquidity resting on the book and, on USD-margined perpetuals, against single-sided open interest, which is the normalized volume rule described above. It only applies where there is a depth snapshot to check against. *** ## Exchange Information Alongside market data, each exchange has a profile covering the details that differ by country: whether the exchange is available, which legal entity and licence covers users there, which products they can access, KYC levels and what each unlocks, trading fees, and the cost of moving money in and out in both fiat and crypto. This information is authored per region rather than as one global answer, published with a public source link, and dated to the month it was verified. *** ## Frequently Asked **Why is my exchange not listed?** Either it has not been submitted yet, or its public API does not expose enough for us to compute volume independently. See [How to list an exchange](/exchanges/list-your-exchange.md). **Why does your volume differ from ours?** Usually the daily boundary, since we use UTC days, one-sided versus two-sided reporting, or valuing the day at VWAP rather than the close. Send us the figure you expect for a specific market and day and we will reconcile it. **Why is our normalized volume below our reported volume?** Either the reported figure was large relative to the depth observed on your book, or, on USD-margined perpetuals, large relative to your open interest. Each market on your exchange page says which of the two applied. For the book, the most common cause is that our snapshot did not reach the part of your book where your real liquidity sits; depth reach varies enormously between exchanges, and it is worth checking how many levels your public endpoint actually returns. For open interest, most venues turn theirs over 1 to 5 times a day; a market running far above that on a calm day is capped. **Why does my exchange show no normalized volume at all?** No market on that segment had a usable order book snapshot, so there was nothing to check the volume against. Reported volume is still shown. A quote-request price ladder does not count as a book. Coin-margined perpetuals have no normalized volume for any exchange yet. **Why are our execution metrics unavailable?** We do not have a tick-level source for your exchange yet. These metrics need per-market daily files of trades and order book snapshots, with the last 7 days always available and up to a day's delay. We read files rather than recording live feeds ourselves. See [How to list an exchange](/exchanges/list-your-exchange.md#3-execution-quality-data-optional-but-needed-for-execution-metrics). **Can we submit our own volume, depth or reserve figures?** No. Every published figure is computed from data we collect or from public evidence, which is what makes exchanges comparable. Wallet addresses and licence registers are welcome, because those are evidence we verify rather than numbers we accept. **Why does my exchange have no open interest or depth history?** Both are point-in-time quantities. Where an exchange publishes history for them we backfill it; where it does not, the series begins on the day tracking started. **How often is data updated?** Market data refreshes throughout the day, with each UTC day finalised after it closes.