An objective of ninety-nine point nine percent over thirty days is not a promise of perfection. It is a statement that roughly forty-three minutes of failure in that window is acceptable, and that quantity behaves like a budget: it is allocated at the start of the window, spent by every incident, planned maintenance and risky release, and refilled when the window rolls over. Reading it this way changes what the number is for. An objective set at a hundred percent forbids every change, because every change carries risk, and it also costs unboundedly more for each additional decimal place. A budget with room left in it is explicit permission to ship, to run an experiment, to move a database. A budget already spent is a signal to stop changing things and repair reliability instead, and it makes that decision arithmetic rather than an argument between two teams. The number therefore has to be set at a value the organisation genuinely wants, not the highest one anybody can imagine, because you will be living inside the consequences either way. After this Concept you can convert a stated objective into an allowance of failure time or failed requests over its window, and say what the remaining allowance permits you to do this month.
An error budget is a quantified allowance for acceptable unreliability, derived as the complement of a Service Level Objective (SLO), that gives teams a measurable threshold governing the tradeoff be…