How to clean CSV data on Windows without losing the source
A practical CSV review workflow covering encodings, dates, duplicates, row counts, cleaning recipes and export checks on Windows.
Preserve the input and define the result
Keep the original file unchanged. Work on a copy and write down the intended columns, row grain and destination format. A customer list, an event log and an invoice table have different definitions of a duplicate. Decide what one row represents before removing anything.
Inspect text, identifiers and dates first
Check the delimiter and encoding using names with accents or non-Latin text. Keep identifiers such as postcodes or account codes as text when leading zeros matter. Inspect ambiguous dates explicitly: 03/04/2026 can mean different days. The Tidyset Windows candidate uses day-first interpretation for ambiguous dates, so check a known example against its source.
Separate missing data from valid zeroes
A blank, a zero and the text N/A may carry different meanings. Count missing values before replacing them. Review transformations on a few representative rows, including an empty value and an unusual character. A suggested cleanup is not a business rule until you have checked that it fits the data.
Use an explicit duplicate key
Identify the fields that should uniquely identify a row. Two rows with the same name may be different people; two events with the same amount may be legitimate repeats. Inspect candidate groups and keep an audit of removed records. Compare counts and totals before and after each major transformation.
Save a recipe and export to a fresh file
A saved recipe makes the steps repeatable, but a new source may have a different schema. Test the recipe on a copy before applying it broadly. Export to a new filename, reopen it in the destination program, and check column names, row counts, leading zeros, dates and a representative total. Keep the input and recipe with the result.
Know the candidate limits before choosing Tidyset
Tidyset for Windows includes profiling, cleaning suggestions, formulas, clustering, undo and CSV/JSON/Excel export. Inputs are limited to 128 MB; Excel import reads the first worksheet and JSON import expects flat objects. Saved recipes support up to 1,000 steps. Its product page states the release status and remaining native validation.