How we know something is down

Three independent signals, fused into one confidence score — and we always show you the evidence.

1. People report it

Reports are counted per minute and compared with what's normal for that service at that hour of the week, learned over the last 8 weeks. A spike only counts when it's statistically surprising (Poisson tail p < 0.0001) and large in absolute terms (at least 10 reports and 3× the usual rate), so three people with flaky Wi-Fi never take a service "down". Each network can report a service once per 30 minutes; cloud and VPN networks count a quarter.

2. We check it ourselves

Every 60 seconds we test DNS, TLS and HTTP from 9 Cloudflare regions plus independent servers outside Cloudflare (so we can still see problems when Cloudflare is the one having a bad day). A region counts as failing after two consecutive failures. Bot-protection pages (403/429) count as up.

3. We read the official status page

Where a service publishes a machine-readable status page we poll it every 2 minutes.

The verdict

We say	When
No problems detected	No meaningful signal.
Users report problems with X	Crowd spike only — nothing else confirms it yet.
X is having issues	Crowd spike confirmed by probes or the official status page.
X is down	High confidence for 2 minutes and most regions failing, or the vendor declares a major incident.
Looks like an ISP issue	Over 70 % of extra reports come from one network or city — likely your provider, not the service.

We raise fast and clear slowly: a service returns to normal only after 10 calm minutes with every probe green, one step at a time. That's why you won't see flapping.