In This Memory Test, Relevance Wasn't Authority

I ran the next set of tests on my AI memory judgment, and the result changed the shape of the problem. The earlier result was: Retrieval accuracy and action correctness can diverge. The new result is sharper: Some memory failures are not representation failures. They are authority-arbitration failures. In plain language: The system did not only need to find the memory most related to the query. It needed to find the memory that was allowed to govern the action. Those are not the same objective.