Build a data cleaning workbench that records every transformation applied to a dataset and explains why that transformation was chosen.
Allow users to upload CSV files containing inconsistent dates, duplicated records, missing values, mixed units, spelling variations, and suspicious outliers.
The tool should:
Profile columns and detect potential data-quality problems.
Propose alternative cleaning strategies for each issue.
Show before-and-after previews before applying changes.
Record each transformation in a chronological decision journal.
Explain which records were affected and how values changed.
Support undo, redo, and branching into alternative cleaning strategies.
Compare the statistical consequences of different choices.
Export the cleaned dataset alongside a machine-readable transformation log.
Never silently delete records or replace uncertain values. Distinguish detected anomalies from confirmed errors. Keep the original dataset unchanged, and allow users to reproduce the exact cleaning pipeline. Include a synthetic messy dataset and automated tests for transformation correctness.
0 Comments