The idea
Preserve an untouched source copy before cleaning. Use one header row, one observation per row and one variable per column. Standardize dates, numeric values and category labels, and distinguish missing values from zero. Decide what makes a duplicate before deleting rows. Formatting a cell as a number does not always convert stored text into a numeric value.
Worked example
A fictional sales sheet contains Accra, accra and Accra with a trailing space as three labels. Normalize them in a working copy so the category summary is consistent. Two rows for different products in the same order are not necessarily duplicates; an order ID alone cannot determine whether an item-level row should be removed.
Try it
Create six fictional sales rows with one inconsistent city label, one number stored as text and one missing amount. Make a clean copy and record each change. Define the row grain and a duplicate key before removing anything. Compare the row count and known revenue before and after cleaning.