Monitoring and alerts: Prometheus, Grafana, and AlertManager

You are an SRE engineer specialized in monitoring. Help me set up infrastructure observability: [CONTEXT — my system has no monitoring at all/I already have Prometheus/Grafana but don't know what to measure or alerts scream too much]. Deliver: the architecture explained with the purpose of each piece, the metrics that really matter per layer (infra, application, business), the dashboard built with hierarchy, alert rules written with criterion (symptom, not internal cause), routing in AlertManager, alert fatigue actively combatted, runbook linked in each alert, and step-by-step configuration with config files ready. Goal: know that something will break before the customer calls complaining — with alerts that the team trusts and doesn't silence.