Introduction

You cannot manage what you cannot measure, and you cannot promise a reliability number you never observe. This final module of the path is about both halves: how Cogitave measures a running system, and how it turns that measurement into an engineered reliability discipline instead of a hope.

The two halves are owned by two standards, and they are deliberately kept separate. The observability standard is the measurement layer - the telemetry spine (traces, metrics, logs) and the SLI/SLO definitions. The reliability standard sits on top of it and adds the engineering: how to set an SLO, how to alert on it, and the error-budget policy that decides when the team stops shipping. The process decision behind both is recorded in ADR-0020.

Day-0 honesty

Read every rule here as the spec each service conforms to as it is built, not a report on a running fleet. At Day 0 no service is in production, no SLO is being burned, and no pager has fired. "The on-call engineer pages" describes the spec'd behavior once a service is live and instrumented - not a claim that a rotation runs today. This matches how the rest of the estate reports itself honestly: Yuva still marks live inference as MOCK and learning as DORMANT.

What you will get from this module

A working command of the telemetry model - what a trace, a metric, and a log are in Cogitave and why they are bound to one queryable Core model - plus the vocabulary of reliability engineering: SLI, SLO, error budget, burn rate, and the policy that gives an SLO teeth. By the end you should be able to instrument a service and reason about whether it is reliable enough to keep shipping.