Uptime
Availability
Availability is the share of a period during which a service did what users needed it to do. It is the ratio behind an SLA, such as 99.9 percent over a month.
How it is measured
Compute good time over total time, or good probes over all probes in the window. Say the window (a calendar month is typical), the check interval, and what good means, for example a 200 under 2 seconds from at least two regions.
Time-based and request-based versions differ. A 30-day month is 43,200 minutes, and 99.9 percent allows 43.2 bad ones. A 1-minute probe sees one-minute granularity. A 5-minute probe can miss a four-minute outage entirely.
Worked example
A paid newsletter archive on Astro behind Cloudflare is probed every minute from three regions. In April there are 43,200 minutes per region. The monitor logs 41 failed minutes: 26 from a bad deploy on the 9th, 15 from a certificate renewal glitch on the 22nd. 43,159 good minutes out of 43,200 gives 99.905 percent.
The status page rounds to 99.9. The owner notes that the 26-minute deploy outage alone used about 60 percent of the allowed 43.2 minutes.
How it differs
Availability measures whether the service worked, including partial failures if you define them. Uptime is the narrower reading: did the host answer at all. A site can show 100 percent uptime and poor availability if every login returns a 500.
Common errors
Not stating the window. Counting maintenance as good or bad without saying which. Averaging per-region numbers with no global rule. Probing only the homepage. Rounding 99.85 up to 99.9.
In practice
Write one sentence defining good for your site: URL, status, time limit, regions. Compute last month's figure with it and reuse that sentence next month. If the number goes into a contract, check the SLA text agrees.