A system does not obey a plan. Content engineering stays inside the loop. It does not publish a standard and walk away.

Watch before you change. After publishing, read at least one real sample before you decide the structure of the next piece. Adding rules without watching only records anxiety. Annotation results, rejection reasons, and citation delay are the watching. They are not decoration.

Put responsibility on the structure, and keep human judgment. If a page is restated wrongly, ask first which part of the fact base, the comparison, or the gate made the error easy. Then ask about the editor. Blaming only the person leaves the structure free to produce the next page. Human judgment stays on standards such as whether a fact may be said in public.

Stretch the time span. A single sprint hides delay. Use weeks to see whether the same subject is still restated steadily, and months to see whether the fact base has split into contradictory versions. Use the short cycle only for rejections and stale numbers.

The cadence can be concrete. Each day, clear rejections. Each week, read samples and annotations. Each month, check which values in the fact base have been rewritten differently across pages. That cadence is part of the content system.