Since 19:03 UTC every monitor has a Speed figure: the share of its checks that finished within a target, and the p75, the time three in four checks stayed under. It counts the total time of each check, redirects included, on the scheduled checks of the voting probes that came back up. Since 19:12 UTC only the probes in the monitor's audience regions count, so a site whose visitors are in Europe is not counted slow because Sydney is far away; a monitor with no audience counts every voting probe, and By city greys out the cities it leaves out. The target is the new Fast within (ms) setting; left empty it is 1,000 ms for HTTP and TLS and 200 ms for the other types. Speed is only a figure, with no alerts, and it is separate from the Slow above threshold.
The monitors list shows the last 24 hours as «96% fast», and the monitor page has a Speed section for 24 hours, 7 days and 30 days. Under it, By city shows each voting probe's median check split into network, DNS and TLS, server, and redirects and download, and Server time ≈ gives the part your server takes, which is the same from every city; the rest is distance. The last 14 days were counted at once from the stored results, so the 24-hour and 7-day figures are complete now and the 30-day one is complete in about two weeks. Changing the target or the audience regions recounts those 14 days straight away.
Status pages on the new design have a new option, Speed (30 days), off by default and overridable per entry, which adds the figure to an opened entry and to status.json. In the API, monitors carry speed_target_ms, speed_target and, in the detail, speed for each window. More in Managing monitors.