One draft round, traced token by token: a small model guesses ahead, a big model checks the whole guess in one sweep, and decides — token by token — what to keep, what to throw out, and what to mint fresh.