Prepare the Data
Know which layer does the work
Every question in this domain asks you to pick the right tool at the right layer. Get Data chooses the connector and the storage mode (Import, DirectQuery, Dual); Power Query Editor performs all shaping; the data model stores the result. Learn the common transforms - Remove Columns, Keep Rows, Split Column, Group By, Pivot/Unpivot, Merge Queries, Append Queries - and know when a custom column written in M is the honest answer instead of a chain of clicks.
Profile before you clean
Column quality, column distribution and column profile expose errors, blanks and outliers. By default they evaluate only the top 1,000 rows, which is a classic trap: a query can look perfectly clean while hiding bad values deeper in the table. Switch to the entire data set only when you need certainty, because it costs performance.
Append versus merge
Append stacks rows from queries that share the same columns (a union). Merge joins columns by matching keys (a join). Twelve monthly files with identical columns get appended; a product category brought onto each sales row gets merged. Questions deliberately blur these two.
Folding and refresh performance
Steps that translate into the source's native query language fold; once folding breaks, every later step runs locally. Right-click a step and check whether View Native Query is greyed out. Filtering rows and removing unused columns early shrinks the model, and disabling load on staging queries keeps them out of the semantic model entirely.
Worked example
A refresh takes 40 minutes. You filter the fact table to the last three years, drop six unused columns, and confirm the final step still folds. Rule of thumb: push work upstream - filter rows and remove columns as early as possible, never after a step that breaks folding.