How Uptime Is Measured and Why the Numbers Differ Between Sources

Two sources can report 99.99% and 99.5% uptime for the same service and both be telling the truth, because uptime is not a single fact: it is the output of four measurement choices. Those choices are the time window, the check interval, the definition of "down," and where the checks run from, and no two monitors make them identically.

Once you know the four levers, conflicting uptime numbers stop being confusing and start being readable. Here is what each lever does, followed by StatusBird's own methodology as a worked example of how a transparent monitor should disclose its choices.

Why does the measurement window change the number?

Uptime is downtime divided by total time, so the denominator matters enormously. A service with one bad 9-hour day measures 98.75% over that week, 98.8% over that month, but 99.9% over the year. Vendors naturally quote the window that flatters them, often trailing-12-month figures where one bad incident dissolves into thousands of clean hours. A 90-day window, by contrast, keeps recent incidents visible and relevant. Neither is wrong; they answer different questions. When comparing sources, always check the window first.

How do check intervals affect measured uptime?

A monitor can only see downtime that overlaps its checks. Checks every minute see nearly everything; checks every 10 minutes can miss short incidents completely, recording 100% while customers saw errors. Interval also sets precision at the edges: with 10-minute checks, a measured outage's start and end are each uncertain by up to 10 minutes, which materially distorts short incidents. This is a big enough topic that we covered the detection math separately in why 2-minute check intervals matter.

What counts as "down"? Major versus degraded

The biggest source of disagreement is severity classification. Is the service down when its status feed says "degraded performance"? When one API endpoint errors but the rest respond? When pages load but take 20 seconds? A monitor that counts every degradation reports far lower uptime than one counting only full outages, on identical underlying reality.

There is a genuine trade-off here. Counting everything makes numbers noisy: some infrastructure vendors report minor degradations almost continuously, which would drown real outages in static. Counting only hard failures understates customer pain. The only wrong answer is not disclosing which rule you use.

Why does check location matter?

Modern services run in many regions behind CDNs, so a failure can be total in one geography and invisible in another. A monitor checking from a region the outage did not touch records clean checks while customers elsewhere see errors; monitors averaging across many regions may record a partial outage as a slight dip. Client-side factors like DNS and routing can also make one observer's "down" another's "fine." When two monitors disagree about a specific incident, vantage point is a common culprit.

How does StatusBird measure uptime?

StatusBird publishes its choices so the numbers can be interpreted. As one transparent example of the four levers:

• Window: public grades use a rolling 90-day window, so recent reliability dominates.
• Interval: every service is checked every 2 minutes, around the clock. Over a recent 90-day window that meant more than 4.3 million checks across 84 services.
• Definition of down: only major and critical outages count against public uptime. Minor degradations are tracked but excluded, because some vendors report minor issues near-continuously and including them would pollute comparisons. This means published figures understate total disruption rather than inflate it.
• Source: checks read each vendor's official status feed and severity indicators, normalized to one severity scale across all 84 services, so a "major" means the same thing for a payments provider as for a shipping tool.

Under those rules, as of July 2026, 71 of 84 monitored services earned an A+ over 90 days while measured uptime for the rest ranged down to 97.11%. Every grade is public on the reliability index, and the aggregate picture is in the State of E-commerce Infrastructure 2026 report.

How should a store owner read conflicting uptime claims?

1. Find the window. A 12-month average and a 90-day figure are different questions.
2. Find the severity rule. "Includes degradations" versus "major outages only" can explain a full percentage point.
3. Prefer sources that disclose methodology. A number without measurement rules attached is marketing.
4. Prefer independent measurement over self-reporting for anything tied to money, like SLA claims; vendors' own histories often post late and understate, as covered in status page lag.
5. Translate percentages into hours. 99.9% is roughly 8.8 hours a year; 99.5% is nearly 44. The downtime cost calculator turns those hours into dollars for your store.

The uptime number that matters most is not any vendor's marketing figure; it is the measured, methodology-disclosed number for the specific services your revenue runs through, over a window recent enough to reflect the vendor you are actually dealing with today.