LLM Evaluation: How a Benchmark Turns Raw Answers Into Comparable Numbers