Three terms get used as if they were one, and confusing them produces either an unbuildable promise or a target nobody can check. A service level indicator is a measurement: a ratio you can compute from data the system already emits, such as successful requests divided by total requests, or the fraction of responses under a stated time. It is a fact, and it exists whether or not anybody has set a target. A service level objective is a chosen value for that indicator over a stated window, for example ninety-nine point nine percent over thirty days. It is an internal decision, and it can be changed by the people who set it. A service level agreement is a contract with somebody outside, and its distinguishing feature is that missing it costs money or credits. The practical consequences follow from the order. You cannot set an objective for something you are not measuring, so the indicator comes first. An agreement should always be looser than the internal objective, because you want to know you are in trouble before a customer is owed compensation. And an objective set at a value nobody has ever measured is a guess. After this Concept you can name the indicator behind any reliability promise, say whether a given number is an internal objective or an external agreement, and explain why they should not be equal.
Service Level Indicators (SLIs), Service Level Objectives (SLOs), and Service Level Agreements (SLAs) form a hierarchy of reliability measurement in Site Reliability Engineering: an SLI is a quantita…