Transformation puts the extracted data into the shape the target expects: cleaning, type changes, renaming, column calculations, aggregations and joins. It follows explicit business rules, applied deterministically and testable separately from the rest of the pipeline.