The article explores the limitations of similarity scoring in matching API queries, emphasizing the need for a more nuanced approach. It begins with a thought experiment where a SQL query is matched against a list of potential queries using similarity scores. While a query with a high similarity score might seem like the best choice, it can lead to errors if the context is overlooked, such as mismatched HTTP methods or already consumed mocks. Tusk Drift, a tool for recording and replaying production traffic, encountered challenges when selecting the correct mock responses during testing. The authors initially considered similarity scoring methods like Levenshtein distance but found them inadequate due to semantic constraints and the need for speed in test execution. Instead, they focused on establishing strict constraints to filter out obviously wrong matches. Examples of these constraints include ensuring that GET requests do not match POST requests and that consumed mocks aren't reused. The article details how they developed a priority cascade for matching, which organizes constraints into two dimensions: scope and criteria. This structure allows for precise matching by prioritizing exact matches over reduced ones and ensuring matches occur within the right context. Interestingly, 89.6% of Tusk Drift requests never reach the similarity scoring phase, as they are resolved through these constraints, highlighting the importance of a well-defined matching system. The authors conclude that effective matching requires identifying these critical constraints first, with similarity scoring serving only as a tiebreaker when necessary.