Amazon's cloud platform, which a large share of the web runs on.
This is what our probe sees when it requests AWS's public homepage, every five minutes, from our US West region. It is not AWS's own status page, we have no relationship with them, and we can only see what any visitor could.
So a service can look fine here while a specific feature, region or account is broken — we're checking the front door, not every room. It cuts the other way too: every check is made once, from that one region, with nothing in a second region re-checking it, so a bad network path on our side — or a service that turns datacenter traffic away — can read as an outage here while real visitors are fine.
We only call something down after 3 failed checks in a row — about 10 minutes of continuous failure — which is why we'd rather be a few minutes late than publish a wrong outage about someone.
Same product, pointed at your site instead. Want to hear next time, before your customers do? We'll watch it free — 25 monitors, no card — or run a one-off check now with the free tools.