Observability for mortals: three metrics that catch most incidents

You do not need a three-pillar observability platform with dashboards nobody reads. You need three numbers with alerts on them. One: error rate per deployment. If a deploy raises errors, roll it back automatically. Two: queue depth on anything async. Growing queues mean downstream pain arriving in slow motion. Three: latency at the ninety-ninth percentile. Averages hide the users having the worst day. Three metrics, three alerts, one Slack channel. We caught ninety percent of our incidents with this setup for two years. The sophisticated platform came later, when scale demanded it. Start with the three numbers. Add complexity when reality asks for it, not when a vendor does.