/smoke-test — prove a change works at runtime

Take a set of changes, run the real code through the nearest real entrypoint on real infrastructure, and confirm the change did what it should by reading every observable surface — not just the status code. Then fix bugs in a loop until all surfaces agree.

A smoke test is not a unit test and not a build. It answers one question: when a real caller hits this code path right now, does the intended effect actually happen?

Hard rules (refuse these even if tempted)