Nine chapters built the telemetry; this one decides what it is for. The checkout SLO — 99.9% availability and 95% of requests under 500 ms, over a rolling 28 days — gets set in a meeting, turned into an error budget, wired into burn-rate alerts that replace the guessed thresholds of Chapter 9, and backed by the one-page policy that lets reliability win a roadmap argument with data instead of anecdotes.
5 topics
The Chapter 9 pages fire on thresholds somebody guessed in a code review — error rate above 2%, latency above a number that felt right. This chapter replaces guessing with a contract: Mara, the bookings team, and product agree that checkout succeeds 99.9% of the time over 28 days and answers 95% of requests under 500 ms, built from the Chapter 5 histogram and the Chapter 3 load-balancer metrics. Everything else follows from that one agreement.
Five topics carry it end to end: the SLI/SLO/SLA distinction that keeps engineering targets out of legal contracts; the two ratio SLIs computed from telemetry that already exists; the error budget — 40.3 minutes per window — that turns reliability into a spendable resource; the multi-window burn-rate ladder that pages only when that budget is at risk; and the cultural machinery — report, decision rule, one-SLO discipline — that makes the number change decisions.
From measurement to decision — the chapter in one line