Part of engineering knowledge cannot be taken from books alone. It appears while building and repairing. Content engineering is the same. A handbook supplies the first standard. Knowing that a certain kind of page gets skipped comes from annotation and sampling.

Split what you learn. Knowledge of the subject: which boundary of this product is usually misstated, which figure must carry a date. Knowledge of the method: a comparison after the definition is restated more often than one at the end, and steps beyond seven get collapsed and dropped. Write both back into the fact base or the scorecard. Otherwise they stay in one editor's memory.

Repair creates as much knowledge as building. One rejection that is fixed teaches more about a vague standard than ten new pages. Failure records have to be searchable, so the next person can see whether the last gap was a source or a dimension.

Knowledge goes stale. After a model or corpus change, an old rule that "this shape gets cited" may fail. Date the method notes. If a note is not rechecked within a measurement cycle, demote it to a hypothesis. Do not keep it as a rule.

Each week, reserve time for one task: turn the gaps that repeated in annotation and rejection into one new acceptance sentence. That is the research program of the content system. It does not need a separate report.