The test runs on the server, from the probe's network, with an empty cache, and does not run your site's JavaScript. Files a script adds later (analytics, chat widgets, lazy content) are not included. Use it to compare versions of your own page over time and to find the fixes in the list, not as an exact copy of one visitor's experience.
Public tests never reach private or internal addresses, so nobody can use the server to look into your network. Signed-in users can test them when an administrator has allowed private addresses in the configuration.
Anyone with its link. Results are not listed anywhere public. Your recent tests on the home page are kept only in your browser; you can clear them or switch the history off.
HTTP Archive - a standard file with every request and its timing. Chrome, Edge and Firefox developer tools and many online analysers can open it.
A failure from one probe only makes the monitor suspect; other probes re-check it within seconds. Only when the required number of different probes agree is it declared down. Slow answers are a separate, milder "degraded" state.
Every 30 seconds at the most, once a day at the least.
Yes: an administrator allows private addresses in Settings, or you run a remote probe inside that network.
Java sends real ICMP echo when the system allows it (Windows always; on Linux give the program CAP_NET_RAW or allow unprivileged ping
with net.ipv4.ping_group_range). Otherwise it falls back to a TCP probe on port 7. Use a TCP monitor if in doubt.
They keep running and are logged, but nobody is alerted and the time does not count against the SLA.
Open Alerts - Log: a failed send shows the error (for example a wrong token). If it says "Dry run", alerts are switched to dry-run mode in Settings. E-mail needs MAIL_HOST in the server configuration.
Yes: create an escalation policy per team and choose it on each monitor.
Yes: enter it in the page settings and point a CNAME record at the server. Add the name to your reverse proxy and certificate.
Measured time minus maintenance, then downtime (first failed check to first good one of each confirmed incident) as a share of it.
In your own PostgreSQL database. Nothing is sent anywhere except the alerts you configure and the checks themselves.
Raw check results 30 days, hourly summaries 400 days, speed tests 90 days (all configurable).