These days, I lean more toward the idea of being able to "decorate" tuple types with element names. In this approach, as in the ancient proposal in this issue description, record types would remain as a typechecker-level fiction, that is, they would not be reified types. You could write...