When cross-model tracking is enabled, each content draft is scored concurrently by observer models to record how competing LLMs would rank the same variant without influencing the reward, claim posteriors, or win conditions of the primary optimizer run.