You open the site. It spins. Or it times out. Or your monitoring Slack channel lights up red. The question that matters is not "is something wrong" — something is. The question is where.
Three common answers that are not "the whole internet"
Just you (or just your office). Corporate Wi‑Fi, a bad VPN, a captive portal, DNS on your laptop pointed somewhere cursed. Before you wake the team, load the URL on your phone off cellular. If that works, stop. Fix your network, not production.
Just one path on the internet. A peering blip between one cloud region and your host, a CDN POP having a bad hour, a firewall rule that only some routes hit. From one vantage point the site is "down." From another it is fine. Customers in some cities complain; others never notice.
Actually everyone. Origin is unreachable, the certificate expired, the deploy broke the app for all traffic. Multiple independent places fail the same way at the same time.
Most false pages live in the middle bucket. Most tools that trust a single probe treat the middle bucket like the last one.
How to tell without a war room
- Open the site from a second network — phone cellular is the fastest truth serum.
- Look at more than one monitor region. If Frankfurt fails and Ashburn and Singapore are green, you do not have a global outage. You have a path problem, a regional dependency, or a flaky probe.
- Check the usual third parties — your DNS host, CDN, and cloud provider status pages. A surprising number of "our outage" nights are someone else's.
- Only then page on-call if the evidence still says customers are broken.
If your monitor opens an incident only when at least two regions agree, steps 2 and 4 collapse into one: the tool has already asked the internet for a second opinion.
What partial disagreement should make you do
Do not ignore it. A single region failing hard can still be customers in that geography. Treat it as a regional incident: check CDN routing, anycast, geo DNS, and whether that region's dependency (a payment provider POP, a cache layer) is in the path.
Do not escalate it like a full outage either. Page the person who can look at traffic by region, not the entire company Slack with "WE'RE DOWN."
The short version
"Is it down for everyone?" is answered by independent vantage points, not by refreshing harder from the same desk. Build that into how you monitor, and you will spend fewer nights debugging your ISP while the pager screams about a site that was never offline.
