The Myth of the Five-Nines Uptime

Five nines sounds precise, which is exactly why it gets thrown around in pitch decks and SLAs without much scrutiny. It works out to about five minutes of downtime a year — a target that's less a statement of engineering fact and more a statement of how much money and organizational discipline a team is willing to spend chasing it.

Most outages people actually remember have nothing to do with raw infrastructure reliability. They come from a deploy that went out everywhere at once, a config change with no rollback plan, or a dependency nobody realized was load-bearing. Five nines of hardware uptime doesn't protect against any of that.

The teams that actually hit high reliability numbers tend to talk less about uptime targets and more about blast radius — how much of the system a single bad change can take down, and how fast they can tell. That's a much less quotable metric, which might be exactly why it doesn't show up on the slide.