How an LLM Evaluation Benchmark Produces Comparable Numbers: Inside the LFORLA Mechanism