Parquet vs CSV: when analysts still reach for flat files

Parquet compresses and types data efficiently for Spark, DuckDB, and cloud warehouses. CSV stays the lowest-common-denominator for email attachments, regulatory submissions, and quick human review.

Split the workflow

• Store canonical tables in Parquet/Iceberg inside the lake.
• Emit bounded CSV slices for stakeholders who will not query SQL.
• Use a viewer for those slices instead of re-importing to Sheets.