DATA QUALITY
Keep the meaning.
Improve the shape.
Plan useful data export transformations: select columns, map names, normalize units, preserve identifiers, and validate the result before delivery.
Select only the fields the workflow needs
Start with the next action the recipient will take. Define a field allowlist and choose names that make sense outside your application. Avoid exposing internal notes, infrastructure identifiers, or unrelated personal data simply because they are nearby in the source table.
Column names, units, and null behavior form part of the export contract. A file that contains less data can be more useful when its contents are deliberate.
Normalize without erasing information
Make dates, currency, decimal precision, and enum values consistent. Preserve source identifiers as identifiers; do not turn them into numbers merely because they contain digits. Keep an explicit mapping for transformed categories so the recipient understands the output.
For nested data, choose whether to retain the structure in JSON or produce related flat tables. Flattening every relationship into one row can multiply records and distort totals.
Validate the final file
Check required fields and types, then read the finished artifact with the receiving application or parser. Compare row counts and a stable aggregate where appropriate. Distinguish rejected records from successfully exported records rather than silently discarding mistakes.
Use the data validation guide and CSV best practices to choose checks for your actual consumers.
A good place to start
Make your first export.
Four real sample files. One fictional dataset. A clear next step for your project.