Latency here is a real round trip: this page publishes a token to a live channel and stops the clock when that exact token comes back over the socket. Nothing on this page is estimated, and nothing is shown that a browser cannot measure.
Probes answered: --. A probe that is never echoed is reported as an error rather than dropped quietly.
One point per echoed probe. Empty until the first round trip completes.
Counted off the socket in one-second buckets. Run a burst below, or open a second tab, and the bars move.
Publishes this many real messages from this one browser tab and reports how many of them actually came back, and how fast. This is a single browser connection, not a fleet: the numbers are honest about that.
Server CPU and memory, cluster-wide throughput, service uptime and per-region latency cannot be measured from a browser. They used to appear here as generated numbers, so they have been removed rather than guessed.
For account-level throughput and usage figures, see the developer dashboard.