DEJAN optimizes for parametric recall by placing factual brand co-occurrences across authoritative training corpora to ensure the entity is absorbed during future model pre-training cycles.