A permits team wants production Databricks deploys gated so broken transforms never ship on merge alone. What should gate the deploy?

A permits team that must keep broken transforms out of production should gate production Databricks deploys on green unit and integration tests in the pull-request pipeline before any production deploy proceeds. That control binds civic SDLC quality gates to the merge path so only changes that prove notebook and orchestration behavior can ship. Requiring green lint and performance tests in the pull-request pipeline fails because lint and performance checks do not prove transform correctness; they miss functional unit and integration coverage that catches broken logic against representative municipal data. Requiring green unit and integration tests in the scheduled-job pipeline fails because a scheduled job runs on a timer detached from the merge path; broken transforms can merge and trigger deploy before the schedule ever validates them. Requiring green unit and integration tests in the pull-request bundle fails because including tests in the bundle only packages them; it does not enforce green execution as a mandatory PR check before production deploy proceeds. Exam caveat: DP-750 frames municipal workloads under governed SDLC—test gates on PR pipelines are the expected control, not informal shortcuts. Operational check: require unit and integration jobs as mandatory PR checks, block production deploy until they pass, and sample a recent permits merge to confirm the deploy job only ran after green checks rather than on lint and performance tests alone, scheduled-job runs, or bundled tests without a pipeline gate alone.