The observability starter we install on day two

We do not wait for scale to add traces. A thin OpenTelemetry layer on day two saves weeks of guesswork when the first real incident arrives at 2am. Guessing is not an operations strategy.

The starter is deliberately small: traces around login and checkout, structured logs with a request id, and one dashboard that a human can read without a training course.

One question per dashboard

If the board cannot answer whether a customer is stuck, it is decoration. We start with traces on the money path and the session path. Everything else is optional until those two are boringly green.

Alerts page a human only when a user-visible path is failing. CPU graphs stay in a folder nobody is paged for. Noise trains people to ignore the one alert that matters.

Install it like a dependency

The starter is a package, not a wiki page. New services import it. If they cannot, they are not ready for production traffic.

Day two is cheap. Day two hundred, without traces, is an archaeology project you invoice in sleeplessness.