CSV cleaning
Clean a CSV without losing track of what changed
CSV files are portable and simple, but real exports often contain inconsistent spacing, category labels, dates, numbers, missing-value tokens, and duplicates. This cleaner turns those issues into a plan you can review before producing a new file.
Profile the file before choosing transformations
The first step is deterministic profiling rather than immediate modification. The product reads the file within the published size, row, and column limits, detects its encoding, and summarizes the structure. You can see column types, missing percentages, distinct counts, duplicate rows, and warnings before requesting AI suggestions.
This separation matters because profiling and cleaning answer different questions. A profile tells you what appears to be wrong; a cleaning plan defines an explicit response. Keeping them separate makes it easier to reject an assumption before it reaches the output.
Use controlled rules for common CSV problems
The real execution path accepts only the operations already supported by the validated rule DSL. These include trimming whitespace, normalizing case and missing tokens, replacing known values, parsing numbers and dates, casting types, renaming columns, removing empty structures or exact duplicates, and flagging outliers.
Advanced actions such as imputing values, winsorizing distributions, or deleting rows based on outlier detection are not quietly approximated. They are kept out of the Worker until their contracts and safeguards are implemented.
- Review affected columns and rule rationale
- Edit only validated parameters
- Preview deterministic impact before running
- Choose CSV or XLSX for the cleaned output
Download a cleaned copy and real audit artifacts
A successful run exposes the selected cleaned-file format, a PDF report, and a JSON audit. The original upload remains separate. Run history and provenance values come from stored run records rather than sample dashboard data.
Reusable recipe exports and data-dictionary downloads are planned capabilities, not current downloads. The public interface labels them as coming soon while the development preview lets the team evaluate those workflows without sending unsupported operations to production services.