Unconfounded Assignment Mechanism
An assignment mechanism is unconfounded if treatment assignment is independent of potential outcomes:
$$P(T \mid Y(0), Y(1)) = P(T).$$Under unconfounded assignment, differences between treatment groups can be interpreted more cleanly because treatment is not systematically tied to the missing potential outcomes.
Randomization is the canonical way to obtain unconfounded assignment.