Every uptime percentage calculator assumes continuous 24/7 coverage. Real monitoring dashboards don't work that way: paused monitors and unmonitored gaps get excluded, not counted, which is why the two numbers rarely match.
A broken IPv6 path does not take a dual-stack site down. It adds real, measurable delay for a share of visitors while an IPv4-only uptime check keeps reporting "up" the entire time.
A check run from one location can fail for reasons that have nothing to do with your site, including, in one real 2025 incident, a problem with the monitoring vendor's own infrastructure in that region. Here is what actually causes this, and what confirming from elsewhere buys you.
For programmatic incident consumption, webhooks deliver in about a second. Polling adds an average of half your poll interval on top, pure waiting, after the incident already exists. Here is the real tradeoff, and the failure mode most guides skip.
A monitor that says your site is down is not always right. Cloudflare can block your monitoring bot specifically, or have its own outage, and neither one means your server is broken.
Cloudflare’s 521 through 526 errors all mean the same broad thing: Cloudflare reached the internet fine, but something went wrong between Cloudflare and your own server. Here is what each code means, and what actually catches each one.
The famous downtime cost numbers (Gartner's $5,600/min, IDC's $1M/hour) come from enterprise IT surveys that rarely apply to most sites. Here is where each one comes from, why it overstates the cost for most businesses, and a calculation framework that does apply.
Picking an uptime check interval is not a question of "as fast as your plan allows." It is a question of which failure mode you are catching and how fast the world needs to know it is broken. Here is how to pick by impact, plus the alert latency math vendors rarely show.