Ask what is wrong with NYC's open data, or how it breaks down. Answers come from the silver layer's quality views, ranked by how many people each problem affects, and every figure ends with the SQL that produced it.
It reads facts about datasets, not the records — Grades, freshness, null rates and column documentation for all ~2,400 datasets. To ask about the records themselves — noise complaints, collisions — use the agent.
Unknown is not clean — A dataset with no score has not been assessed yet, and a few large datasets (311 among them) publish no column statistics at all. The analyst says so rather than reporting them as problem-free.