Reliability is not "the servers are up". It is whether users get what they asked for, quickly enough. So measure it from their side. A service level indicator (SLI) is a measurement of that experience, written as a ratio: good events divided by all valid events.
Pick a concrete definition, for example: a request is good if it succeeds and finishes in under 300 ms. Suppose the checkout service handles 10,000 requests in an hour and 9,950 of them are good. The SLI is 9,950 / 10,000 = 99.5%.
The same shape fits other promises: the share of page loads that finish quickly, or of jobs that complete within a minute. A good SLI describes something a user would recognise as "it works".
💡Good to know
SLI = good events / valid events, measured where users are.