In a Parquet explainer’s experiment using about ten million New York taxi trips, reading the fare column took 175 ms from CSV and 6.6 ms from Parquet, while the file was about one-sixth the size. Columnar storage and embedded types avoid unnecessary column reads and text conversion; dictionary and run-length encoding plus compression further reduce size. Sorting rows by frequently filtered columns helps minimum and maximum statistics skip row groups, although a few incorrect values can widen those ranges. The original video’s content could not be extracted, so the author’s companion article was checked. Results are specific to this experiment, and row-oriented additions or lookups may favor other formats.