A platform that cannot explain itself is not ready for production. We treat observability as part of the build, the same way we treat authentication and backups.
Before the first cut, every service emits structured logs, a small set of golden metrics, and traces across the request path. Dashboards and alerts are written against those signals, with an owner named for each page.
What we refuse to skip
Request latency, error rate, saturation, and the business event that actually matters — a payment captured, a record saved, a job finished. Vanity charts do not count.
Alerts fire on symptoms a human can act on. A page that nobody can silence in five minutes is a page we rewrite before launch.
What the on-call team receives
A runbook next to each alert: what the signal means, the first query to run, and how to roll the deploy back. The person on call should not have to invent the diagnosis at 2 a.m.
That is the bar for a Y-Tech handoff. The system stays operable after the engagement ends, because the evidence of its health was built in from the start.