I would go with the latter (assuming you mean with replacement or "true" randomization). You can add an unconditional skip to each of the first twenty (randomized) items such that no matter which comes first, the participant is skipped past the other 19.