Opening the book…
Code merged, tests green, and a successful deploy tell you the change reached production, not that it works there. The only proof is seeing the intended behavior in real telemetry: the metric moving, the errors staying flat, the feature actually used. Until you've watched it behave under real traffic, the work is in progress, not done, no matter what the pipeline says.
Define what success looks like in production before you ship, in terms you can observe: a metric, a log line, a dashboard. After the deploy, go look. Confirm the new path is exercised, error rates and latency hold, and the number you meant to move actually moved. Only then call it done, close the ticket, and move on. If you can't observe it, you can't claim it.
Some effects are slow or rare and won't show immediately; a weekly batch job or a seasonal path may need a defined observation window instead of instant confirmation. Purely internal changes with no runtime signal are done when merged and verified another way.