How LLM Evaluation Produces Comparable Numbers: Inside the LFORLA RL Game Agents Benchmark