How we measure and how we rank

The composite price

We collect the full ticker set from each venue we track, on a cycle measured in seconds, and store it in our own database. For a given asset we take every market quoted against a US dollar stablecoin or dollar itself. Each independent venue first receives one consensus vote from its most liquid USD market. We take the median venue price, measure the median disagreement, and quarantine isolated prints outside an adaptive band of 1% to 5%. The remaining markets are then weighted by their own quote volume over the last twenty four hours.

The band widens when several independent venues genuinely disagree; with fewer than three venues, the conservative 5% fallback remains. This exists because one halted market should not move a headline number the rest of the site quotes. For the high-low spread, we add a second guard: each market must have at least $10,000 of quote turnover in the previous 24 hours. Turnover is the one liquidity proxy every venue reports consistently, and it stops a nearly empty pair from defining the headline gap. Only the most liquid qualifying quote from each venue participates, so several quote currencies on one exchange never count as several opinions. What survives these checks is published alongside two things most price pages leave out: how many venues agreed, and how far apart the highest and lowest of them were. A wide spread is information, not noise. Beside each spread we publish its confidence: High requires at least three eligible venues and $1 million of combined qualifying 24-hour turnover; Medium requires at least two venues and $100,000; everything thinner is marked Low. Confidence describes the evidence behind the measurement, not whether a trade is profitable.

Rejected quotes are not silently forgotten. We record the asset, venue market, observed price, consensus price, deviation, permitted tolerance, volume, first and latest observation, and sample count. A market that remains outside consensus for at least fifteen minutes and three observations is suspended from composite calculations for one hour. We continue collecting it during that time, allowing a recovered market to clear automatically while preserving its investigation history.

Freshness is enforced before price quality. A venue lane that fails the current collection cycle contributes nothing to that cycle's durable composite. In the browser feed, an individual quote expires after thirty seconds without an update; an upstream socket silent for forty-five seconds is terminated and reconnected with its subscriptions rebuilt. An open connection alone is never treated as proof that its prices are live.

Current venue availability and recent outages are published on our market data status page.

Current coverage: 7,205 markets across 5 venues, of which 5 are counted as independent price sources (binance 1373 spot and 565 perpetual, bybit 520 spot and 887 perpetual, okx 1212 spot and 534 perpetual, pionex 331 spot and 589 perpetual, whitebit 799 spot and 395 perpetual).

Market data from our own market database, measured across 5 venues, updated 34 seconds ago.

Which venues count as a price, and which do not

We currently track 5 venues. 5 are counted as independent price sources in the composite.

A price is only worth averaging if it is an independent opinion. Some venues do not run their own order book: they route orders to other exchanges and republish the resulting prices. Counting one of those as a separate venue does three things, all bad. It double-counts the underlying exchange's volume. It makes a statement like "four venues agree" mean less than it sounds, because two of them are the same book. And in a volume-weighted average it quietly hands the mirrored exchange a second helping of weight.

This classification is stored per venue and used directly by the collector. A venue marked non-independent remains tracked for its own markets and fees, but cannot add a second vote or additional weight to a composite price. The public counts are read from that same current database state rather than copied into the page.

Funding rates

Perpetual funding is collected per contract per venue and annualised to a common basis, because venues settle on different schedules and an unannualised comparison between an eight-hour and a one-hour rate is meaningless. Where we show a gap between venues, it is the difference between the most and least expensive venue we track for that same contract.

Fees

Fee tables are the one thing we do not automate, because venues bury the real number under volume tiers, token discounts and promotional periods that a scraper reports as fact. Each row is read by a person, recorded as the standard rate for a new account with no tier and no discount, and stamped with the date. Anything older than ninety days is flagged by our own monitoring before a reader ever sees it go stale.

How a venue is rated

Five components, each scored out of five: cost, liquidity, tooling, security and support. Cost comes from the verified fee schedule and observed funding. Liquidity comes from the market counts and volumes in our own database rather than from what a venue publishes about itself. Tooling, security and support are judged, and the judgement is signed by a named person who is accountable for it.

A ranked list is ordered by the thing its heading promises. If the page says lowest fees, the order is the fee column and nothing else. We do not run a house pick that quietly floats to the top, and no partner has ever been offered a position.

How we make money, and what it does not buy

Some links to exchanges are affiliate links and we are paid when a reader opens an account through one. That funds the collection and the writing. It does not buy a rating, a ranking position, or the removal of a criticism, and venues that pay us nothing sit in the same tables on the same terms. The full disclosure spells out the commercial relationships.

Corrections

If a number here is wrong, we want to know and we will fix it and say that we did. Write to the editor with the page and the figure.