A cleaned file arrives with no explanation of what changed. Its columns are tidy, but the next person cannot tell whether missing rows were excluded, units were converted or labels were combined.
Describe each operation
For every meaningful change, record the input field, rule, reason and resulting field or file. A note such as “Converted centimetres to millimetres by multiplying by ten” is more useful than “Fixed length.”
When filtering, record the condition and the number of records affected. Keep a clear reference to the source version used for the work.
Separate correction from interpretation
Fixing a documented typing error differs from grouping categories for an analysis. Both may be reasonable, but they should not be presented as the same kind of operation.
Mark decisions that require judgement and explain their limits. Do not label an uncertain replacement as verified simply because it removes an awkward blank.
Test whether the log is sufficient
Ask whether a reader could understand the sequence without asking what you meant by clean. For automated work, connect the note to the script or exact procedure used.
Our guide to dataset context covers the starting information. A transformation log continues that context through the work, preserving the difference between the collected records and the version used for analysis.