The production layer of a pod runs on completion states. Research is done, a draft is done, a lead is qualified, an appointment is booked. Each of those words is a boundary, and the boundary is where the quality of the whole system is set.
A loose completion state produces volume that looks like progress. A lead marked qualified because it met a form threshold has been marked qualified by a definition nobody would defend out loud, and every hour of human attention spent on it afterwards is a cost the system created.
Writing the definition down
Each production seat states what its output must contain before it can be passed on, in terms that can be checked without judgment. What fields are populated and from what source. What has been verified against a second source and what has not. What was inferred and is labelled as inference.
The test of a definition is whether two people reading the same output would agree on whether it is finished. If they would not, the definition is doing no work and the completion state is decoration.
What the definition cannot do
A completion state says the work is done to specification. It does not say the work was worth doing, and nothing in the production layer can say that. This is the distinction the allocation rule exists to enforce. Deciding whether finishing something was the right call is judgment, and judgment is a person's seat.
Systems that blur this reliably fail in the same direction. They get faster, the outputs stay within specification, and the specification drifts away from what the business needed, with nothing in the loop positioned to notice.