Bottom-line recommendation of the CSV-parsing lightning talk: when CSV performance matters, reach for Polars Ruby (≈22× faster than stdlib CSV, ~100MB memory on a 30MB file with 1M rows). DuckDB is the best on memory (~90MB) and still 15× faster. Smarter CSV is only ~3×. Polars Ruby is battle-tested by the speaker in production for 300MB CSVs within a single web request with no caching.