What each phase does
Latency comes first. After one warm-up request that is thrown away, the page makes ten tiny requests to the server's ping endpoint, which answers with an empty 204 response, and times each one. These are HTTP requests over the already-open connection, not ICMP pings, so they include a little protocol overhead. The lowest value is reported as ping and the mean change between neighbouring values as jitter.
Download then opens four streams at once, each asking the server for 25 MB of random data, and keeps them busy for eight seconds, counting bytes as they arrive. Upload sends random blocks in four parallel requests for eight seconds, starting at 1 MB and doubling to 16 MB so a slow link still completes a request and a fast one does not need hundreds. Throughput is bytes times eight divided by elapsed time.
Reading the four tiles
Download and upload are in megabits per second, the unit your plan is sold in. Divide by eight for megabytes per second, the unit file managers show, so a 100 Mbps result moves about 12.5 MB each second. Most home plans are asymmetric, and an upload figure a tenth of the download is normal on cable and DSL. Fibre is often closer to symmetric.
Ping matters for anything interactive: page loads, games, remote desktops. Jitter matters for voice and video, where packets that arrive at irregular intervals have to be buffered or dropped. A connection with a high ping but low jitter is predictable and usable. Low ping with high jitter is the one that makes calls break up unpredictably.
Why the figure is lower than your plan
The plan speed is the rate at the modem under ideal conditions. Between it and this page sit your Wi-Fi link, which loses most on distance and walls, any other device streaming or backing up, a VPN that re-encrypts every byte, and the browser itself on a phone, where a single CPU core can cap a gigabit line well below its rating.
Then there is the path. The test measures the route from you to this site's server, so it includes every network in between and the server's own capacity. A result that is low here but normal elsewhere points at the route, not your line. Run traceroute to see where the hops slow down, and ping a few hosts to compare latency.
Rate limits, data use and interrupted tests
A full run moves roughly 50 to 300 MB depending on your speed, which matters on a metered mobile plan. Each request is capped at 25 MB, and the speed-test endpoints are limited to 120 requests a minute and 1,500 a day per address, about 30 full runs. Repeated runs from a busy office behind one address can hit the per-minute limit.
If the limit is reached before any data moves, the test stops with a message asking you to wait a minute. If it is reached in the middle of a phase, that phase simply ends early and the figure is calculated from what was transferred, so a very fast connection that ran out of requests may show a slightly lower download than a second run would.
What this test cannot tell you
It is one server in one location, with no choice of test server. It cannot say what your ISP delivers to its own backbone, only what reaches this site, and it has no history, so note the numbers if you want to compare days. Results are not shared or stored. The page holds them only until you reload.
A low number is a symptom, not a diagnosis. Compare a wired and a wireless run to separate the Wi-Fi from the line, test at a quiet hour to separate congestion from capacity, and check what is my ISP to confirm the traffic is leaving through the provider and route you expect rather than a VPN or a captive portal.