Start with the source, not the cleanup button
Save an untouched copy and identify where the data came from. A downloaded report, a form export and a manually maintained sheet have different failure modes. Record the expected header names, number of columns, and whether the first line is a header. Compare the number of records before and after cleaning so accidental row loss is visible.
Understand quotation and delimiter rules
A CSV field may legally contain commas, line breaks and double quotation marks. A robust parser uses quote delimiters rather than splitting every line on a comma. For example, a description reading ‘Bolts, assorted’ must stay in one field when surrounded by quotes. Data from some regional spreadsheet settings uses semicolons or tab characters instead of commas; detect the separator before exporting.
Choose the right duplicate rule
An identical row may be an accidental repeated export, but repeated customer names or order totals do not prove that two records represent the same transaction. Keep a stable record ID when possible. If your workflow removes exact duplicates, audit a few removed records and retain the original file for rollback.
Protect identifiers and validate downstream
Postal codes, SKU values and account-like identifiers often begin with zero. Spreadsheet programs can convert them into numbers and drop those zeros. Preserve them as text, inspect date formats, and check non-ASCII names. After exporting a cleaned CSV, reopen it and test it with the intended receiving program or import pipeline.