Understood, thanks.

But is that really different from the point of view of the developer? As long as the column is not used in the output it's just a variable.

I elaborate because I checked the examples since my first comment:

There are two derive sections in introduction.prql.

  derive [  # This adds columns / variables.
    gross_salary = salary + payroll_tax,
    gross_cost = gross_salary + benefits_cost # Variables can use other variables.
  ]

Note the comments about "columns / variables" and "Variables can use other variables", so the distinction is probably not very clear even now.

From my point of view that would be

  let gross_salary = salary + payroll_tax
  let gross_cost = gross_salary + benefits_cost

Then I'd get gross_salary and gross_cost in the result or not, according to the select I'm doing at the end. [1]

By the way, in the examples I saw

  select [name, salary, average_country_salary]

which has some unnecessary brackets. Why not just omitting them?

  select name, salary, average_country_salary

Same for the parentheses in

  aggregate average_country_salary = (average salary)

Maybe omitting [ ] and ( ) would make the parser too complex but I'm all in to offload work to computers instead of to developers :-)

And why not only one type of brackets? Example: the join clause

  join average_salaries [==country]

which gets compiled to

  JOIN average_salaries ON newest_employees.country = average_salaries.country

Why square brackets there instead of round ones?

That's a condition and we're conditioned to put conditions inside (), as in if () {} else {}

And omit == by default, so

  join average_salaries (country)

[1] Edit: maybe you can add a negative select, to remove columns from the output. Maybe:

  remove col1, col2

That would be handy where there are 100 columns and one wants to remove only 5 of them. No need to enumerate the other 95.