Percentages close to 100 look alike, but the downtime they allow does not. 99.9% allows 43 minutes and 49 seconds in an average month; 99.999% allows 26 seconds, less than it takes most teams to read an alert.
Higher targets need different engineering, not just more effort: redundancy across machines or regions, deploys without downtime, and detection measured in seconds. A service also can’t be more available than what it depends on: two dependencies at 99.9% each, both needed for every request, leave at most about 99.8%.
The table below uses the same definitions as the SLA and uptime calculator: a year is 365.25 days and a month a year ÷ 12 (30.44 days). Allowed downtime is rounded down to the second.