Why it is a question
A4 §4: “A graph could outperform a summary because it retains more information, because it retrieves information more reliably, or because the model is better at executing operations on explicit relationships. Those are different findings.” QUESTION-038 asks what turns a persisting structure into an anchor. This asks which of three mechanisms is doing the work when one does.
A3’s question B, whether structured addressability reduces unannounced drift, is held here rather than filed apart: it is the drift-measured special case, and CLAIM-182 already states it as a thesis. A3’s counter goes with it: “A structured record may make agents less willing to revise erroneous commitments. Less drift is not necessarily more epistemic progress” (CLAIM-184).
An answer is a design that manipulates retention, access budget and format independently, scored on decisions in an externally specified environment and not on reproducing the record’s own graph (CLAIM-190). CASE-091 is that design. A negative result counts: “if narrative summaries or latent representations preserve the same revision competence more efficiently, that tells us something about the limits of nucleation’s architecture.”