LLM-driven evolutionary search evaluation methods were shown to be insufficient due to oversimplification. A new measurement protocol was proposed to evaluate the seeds-by-iterations frontier, providing practical guidance for optimizing LLM-driven search.