It's up to you (and the query optimizer can reorder), but you have to make sure the column is available when you use it. Broken up:

I had:

    a = employee(name == "Jones")    a has name, salary and age

    w = a[comp = salary / (age - 18)]   w has comp

You want:

    b = employee[name, comp = salary / (age - 18)]   b has name and comp

    w = b(name == "Jones" && comp > 500)   w has name and comp

In one line:

    w = employee[name, comp = salary / (age - 18)](name == "Jones" && comp > 500)

I'm not sure if you misunderstood them or I misunderstood you, but there's no legitimate reason you can't filter on computed columns as is:

  w = employee(name == "Jones")[comp = salary / (age - 18)](comp > 2000)