Exchanges methodology
How every number on these pages is measured, corrected and scored.
Every number on the exchange pages: what it means and what is corrected before it is shown. Two rules apply throughout. We store what the exchange said, verbatim, and apply every correction when the number is read, so the record stays auditable and a rule can change without rewriting history. And a missing measurement is never scored as a bad one: it reads as unavailable, not zero.
Last updated 2026-10-16.
What we track
Each exchange is tracked per market segment. Segments are different instruments with different notional conventions, so they are never merged into one headline without being named.
| Segment | What it is | Volume checked against the book |
|---|---|---|
| Spot | Buying and selling the asset itself, settled in the quote asset. | Yes |
| USD perps | Perpetuals margined and settled in a stablecoin or USD. | Yes, and against open interest |
| COIN perps | Perpetuals margined and settled in the coin itself. | Not yet: reported volume, open interest and depth only |
Before any number is shown
- Units are declared per exchange and market type, never guessed. A contract's face value comes from the venue's own contract size.
- Where a venue counts both sides of a trade or a position, a per-venue correction halves it. The default is no correction without evidence.
- Non-USD quotes convert through rates from our own market data.
- An entity trading on a parent's order book has its own page but stays out of cross-venue totals, so one book is never counted twice.
Reported and verified volume
Reported volume is the venue's own 24-hour claim, after unit and double-count corrections. We publish it because comparison needs it, not because we vouch for it.
Verified volume caps that claim at what the liquidity resting on the book could have turned over in a day. The limit scales with the market's own volatility: a market that moves more genuinely turns its book over more.
It is deliberately liberal. Where we have no book snapshot the reported figure passes through marked unverified, not punished. A big haircut is not a wash-trading accusation: the venue's liquidity may sit where our snapshot did not reach.
Depth and active liquidity
Depth is resting size within 2% of mid, bid and ask kept separate, one snapshot per market per day. Public endpoints cap the levels they return, so published depth understates the real book and errs in the venue's favour.
Active liquidity widens that band to the market's own daily range: the depth a day's trading could actually use. Size parked far from mid counts for little, since nobody can hit it.
Where our own live probes showed a venue's displayed size does not fill, its depth is capped at what the evidence supports. The cap applies when the number is read and only to venues with a finding, so a venue later cleared reverts at once.
Open interest and funding
Open interest is the USD notional of open perp positions at the period close, corrected like volume and published for USD and COIN perps separately. No observed book means unavailable, never zero.
Funding is the venue's current rate normalized to an 8-hour basis, so venues on different intervals compare.
Execution quality
Measured from the tick-level trades and order books we record ourselves: actual fills, not quotes. One basket, one window, one rule for every segment.
- Basket: the 25 collected assets, the top of the vetted list by market cap. A venue is measured on whichever it carries.
- Window: the trailing 30 days of hourly figures. A market's number is the median over its hours.
- Floor: a market needs 24 hours of tick coverage before it counts.
- Segments: spot and perps alike, each compared with the same kind of market elsewhere. Spot books against spot books, perp books against perp books, pair by pair.
- Combining: a segment's assets are weighted by each asset's open interest across all venues (volume for spot), so every venue is priced on the same portfolio. A venue's segments are combined by their verified volume.
| Metric | What it measures | Scored |
|---|---|---|
| Cost to trade | What a $10k market order costs all in: base taker fee plus half the effective spread. The only figure that includes fees; VIP tiers and fee tokens can lower the fee part. | Yes |
| Effective spread | What trades actually paid: twice the distance from mid to the fill, before fees. | Displayed only |
| Quoted spread | The gap the book showed at the touch, time-weighted. A promise, not a fill, so never scored. | Displayed only |
| Slippage at size | What $10k and $100k orders paid beyond the mid in front of them. Each band holds only orders of about that size. | Yes |
| Predicted slippage and delivery | What the visible book said that same order should have cost. The gap to what it paid is the honesty check on the book. | Yes |
| Markouts | Where the price went 1 second to 5 minutes after a fill. Positive means it kept moving against the taker. The one-minute horizon is scored. | Yes |
| Inside the spread | Share of a market's volume printed inside its own displayed spread. 5% or more flags the market. | Yes |
| Feed speed | Average gap between book updates. A fixed publishing interval is a choice, not slowness, so it is never scored. | Displayed only |
| Coverage | Hours of tick data behind the 30-day window. A market counts from 24 hours; below that it shows on the venue's own page only. | Displayed only |
Predicted versus actual is the point. Predicted slippage is deliberately not corrected for fake depth: the gap between what a book promised at a size and what orders of that size paid is what exposes a book that does not deliver.
Ramp cost
Ramp cost prices a full fiat round trip per currency and region: deposit, buy, withdraw, and the reverse. Each trade leg is the taker fee plus slippage walked through the venue's depth ladder, floored at the half-spread. A leg is unavailable unless every component is known; an unpublished fee is never treated as zero.
The REAL score
REAL is a weighted mean of four pillars, each 0 to 100: Reserves, Execution, Activity and Liquidity. Size is log-scaled, so a venue a tenth the size of the leader still scores about 70. An input we cannot measure hands its weight to the ones we did and never counts as zero. CEX and DEX are ranked within their own type, and nothing self-reported that we cannot check enters the score.
RReserves
30 pointsWhat the exchange actually holds, from live on-chain wallets, audited filings or attestations. Two thirds is size on a log scale: only the largest balance sheet scores 100, and one a tenth its size still scores about 70. One third is whether the book, volume and open interest it shows are sizes that balance sheet could back. A DEX's reserves are its on-chain protocol collateral, treated as a floor, since margin also sits in vaults and with market makers.
- Magnitude 67%: Clean reserves weighted by evidence: live on-chain wallets count in full, audited filings and third-party attestations a little less, a figure over a year old less again. Log-scaled, so only the largest balance sheet scores 100.
- Consistency 33%: Whether the resting book, daily volume and open interest are sizes this balance sheet could back, plus custody quality: the clean share, and how much is the venue's own token.
When it is missing: A CEX publishing no reserves scores 0 here, since its peers publish. A DEX whose collateral protocol we cannot identify is credited a fixed middle score, set near the measured-DEX median.
EExecution
30 pointsHow orders actually fill, from our own tick data on the 25 collected assets over the last 30 days, spot and perps alike. Each market is compared with the same kind of market on other venues: where the price goes after a fill, what a $10k order costs all in, and slippage at $10k and $100k. Fills worse than the venue's own book promised take up to 12% off, and a market flagged for printing inside its own spread earns no credit.
- Markouts 43%: Where the price goes in the minute after a fill. If it keeps moving against the buyer, takers are trading against faster, better-informed flow.
- Cost to trade 21%: What a $10k market order costs all in: base taker fee plus half the effective spread. On major pairs this is mostly the fee.
- Slippage at size 36%: What $10k and $100k orders actually paid beyond mid, so venues are compared on the same trade.
- Delivery vs quote discount: Did $10k and $100k orders pay what the venue's own book promised at that size? 100 means filled at or better than quoted; a shortfall takes up to 12% off the Execution pillar. Quoting wide and filling as quoted is not penalised here: cost and slippage already charge for the price.
When it is missing: No tick data for the venue, or under 24 hours of it on every basket pair. An unmeasured venue is credited at most the lower of its own score and 50, so missing data never beats being measured.
AActivity
20 pointsVolume we could verify against the order books we snapshot, open interest where we observe the perp book (credited up to 3x clean reserves), and how steadily depth, volume and positions hold over the window. Log-scaled. A spot-only venue is not marked down for having no open interest, and a market flagged for printing inside its own spread loses its share.
When it is missing: None of the reported volume sits in markets we snapshot, so nothing could be verified. A perp venue whose open interest we have not ingested is our gap, not its fault.
LLiquidity
20 pointsResting depth within 2% of mid over 30 days, plus breadth: how many markets hold at least $100k of it. Log-scaled. Depth is credited only up to clean reserves, and discounted for a CEX publishing no reserves or a book that never moves.
When it is missing: A venue whose order book we never observed is not ranked rather than scored low.
Intra-spread volume
Per market: the share of volume that printed inside the spread the venue's own book was showing, at a price no displayed order offered. Those fills did not come from the book on show. A market at 5% or more is flagged, bigger books counting for more. A flagged market earns no execution credit, and each segment's flagged share comes off its credited volume and depth.
Tiers
Fixed bands on the 0 to 100 score, not quantiles, so a new listing does not reshuffle everyone.
- Missing data: an input we have not built or captured is not applicable and its weight redistributes. A venue publishing nothing where its peers publish scores zero. Scores are never imputed from peers.
- CEX and DEX are scored together: one algorithm, the same four pillars, and the same cross-venue reference points, so scores compare across types. Only the rank position is reported within each type.
- Not included: reputation, age, brand, listings count, social following, or anything self-reported that we cannot check.