In plain English
A team records a realistic search question and identifies which documents or passages should be relevant to it. The query becomes part of a reference test set, often called a golden query set. Running the same set before and after a retrieval change helps the team compare how well the system finds and ranks useful information.
Example
A startup adds "How do I export my invoices?" to its golden query set and marks its invoice-export help article as relevant. After changing its search system, the team checks whether that article still appears among the top results.