Query Relevance Scores are calculated by querying an LLM across multiple sampled query trials with a strict yes/no recommendation prompt and dividing the positive responses by the total attempts.