CEX methodology
What the cex numbers on this service are, and are not. This page is about how a value is produced; the status page carries the values themselves.
How measurement works
Every value is measured from one of our registered probes at a known location. A browser test such as Cloud Ping starts at your connection instead; those are different starting points and can differ even for the same destination.
- Why methods are never ranked against each other
- Each method times a different operation, so its number answers a different question. A WebSocket request round trip, an HTTP request cycle and a TCP handshake are different amounts of work on the same wire; mixing them into one ranking would look authoritative and mean nothing. Compare rows that share a method, and read the method column beside every number.
- What a failure means
- Latency statistics use successful attempts only, and failed attempts are counted and shown separately. A timeout, a refused connection and a TLS error are three different facts about the network, none of them a timing. A low median can sit beside a high failure count, so read both together with the sample count.
- What freshness means here
- Each row shows when it was last measured and is labelled stale once it is older than the freshness window. A missing value means there is not enough data for that selection, not that the target is fast.
- Where imported reference data appears
- Imported rows come from a published third-party explorer, are labelled as such wherever shown, and are never averaged with our own measurements. They describe someone else's probes at someone else's times.
CEX
This category covers the public market-data and order-entry endpoints of centralised crypto exchanges, listed per venue and product. Each target declares the one method it may be measured with, so a WebSocket endpoint is never silently timed as an HTTP request.
- ws_request — a request the venue answers
- The declared method for most targets here. It sends a request the venue's public endpoint documents and answers, and times send-to-reply with one clock. The agent adapter that runs this method from our probes is not finished (GN-063), so these targets are catalogued, not measured. The public browser test at /cex/ping runs a closely related measurement from your own connection instead.
- fix_logon — a FIX logon round trip
- Two order-entry targets declare fix_logon: a FIX Logon message sent, its reply received. That is a different operation from a market-data request, so the two methods' numbers are never compared or averaged. The adapter is not finished (GN-102), so these targets are catalogued, not measured.
- Why methods are never ranked against each other
- The two methods are kept apart on purpose: a logon round trip to an order-entry gateway and a market-data request to a matching engine describe different paths and different work. Whichever a target declares is the only method it will ever be measured with.