Notes from dependable operations.
Short, actionable reference material for people maintaining services.
Explore the reference notes
Reading latency without guessing the cause
A field note on percentile charts, sampling and the limits of a single response-time number.
Use error budgets without hiding failures
A short explanation of service objectives, allowed failures and why the observation window matters.
Map dependencies before the incident
A practical note on service dependency maps, ownership boundaries and graceful degradation.
Designing alerts without teaching people to ignore them
Route meaningful symptoms, specify actionability and review recurring noise.
Further concepts are explained in the signal model, runbook methods and the interactive budget lab.
Each note begins with a specific operational question and stays within the behaviour it can explain. The material is a public reading reference, not a live service dashboard. For a diagnosis, retain timestamps, distinguish request failures from monitoring gaps and label assumptions that have not been tested. The related runbook material describes how to connect an observation to a reversible first action.
A practical way to use this page is to write down the behaviour described by “field-notes”, then compare it with one observable example. Keep the actual result distinct from the assumption that motivated the test. When something differs, inspect the smallest relevant part before changing other settings.