# Transforms Between the source and the destination, a flow can reshape what it read. Every stage is optional, and every stage is shown to you before it runs. ## Typed columns, proposed for you SnoutData reads a sample and proposes the columns and their types, with a reason for each. You review the proposal and change anything you disagree with: a column's name, its type, whether it is included at all. Where a record contains a **repeating group** (a list of line items inside an order, for example) and the parent has a key, that group can become a **child table** with a foreign key back to the parent, instead of being flattened or stringified. ## Map, cast, derive Rename a column, change its type, split one column into several or combine several into one, and derive new columns from the ones you already have. ## Filter Keep only the rows that matter. Rows that do not match are not written. ## Dedupe Drop duplicate rows, on the columns you choose rather than on the whole record, so "the same customer twice" works even when a timestamp differs. ## Redact Personal data redaction is a stage of its own, which means the sensitive columns are removed **before** the data reaches the destination, not after it arrives. This matters most when the destination is somewhere you would rather not put personal data at all, such as a file you are going to share or a [fine-tuning set](https://docs.snoutdata.com/flows/destinations).