How would you design a flow that is easy to maintain?

Name every step for what it does. Clean 1, Clean 2 tells the next person nothing.

Filter and remove fields at the input, so everything downstream processes less.

Use wildcard unions and wildcard pivots so new files and new columns are picked up without editing.

Add an exceptions output for rows dropped by a join, so problems surface rather than disappearing.

Use create/replace rather than append on outputs, so a rerun does not duplicate the data.

And keep one flow to one purpose. A single flow doing five unrelated things is the one nobody wants to touch.