A deployment pipeline connects three workspaces, one per stage: Development where people build, Test where a few people check the content against test data, and Production that everyone uses. You don't publish to Production from Power BI Desktop; you Deploy from one stage to the next.
A deployment copies the content: the reports and the semantic models, at the version in the source stage. It does not copy what belongs to each stage: workspace access, gateway mapping and credentials, refresh schedules and row-level security membership stay as they are. After the first deployment to a stage, set those up there.
Each stage should read its own data. A deployment rule on the target stage changes a semantic model parameter, such as the server, every time content arrives: Test reads kl-sql-test and Production reads kl-sql-prod. Create the rule before you deploy, because it applies during the deployment.
Deploy only what was tested: Development to Test, check, then Test to Production.
The same version moves through the stages; rules switch the server at each step.