Whole‑page ranking jointly optimizes ads and organic content to balance business outcomes and user engagement under complex layout and business constraints. Existing non‑autoregressive (NAR) approaches enable fast page generation but usually score candidates independently of previously selected items, which limits their ability to capture order‑dependent interactions. Autoregressive (AR) methods address this limitation through sequential conditioning, yet they incur substantial inference overhead that hinders large‑scale deployment.\n\nWe propose Multi‑Item Scoring (MIS), a prefix‑aware NAR‑style framework that incorporates sequential context while preserving efficiency. MIS integrates global candidate interactions, localized prefix conditioning, and position‑aware scoring within a shared representation architecture. The system computes reusable representations once per request and performs lightweight prefix‑conditioned scoring during constrained beam search, capturing important sequential dependencies without repeated full‑model inference.\n\nExperiments on KuaiRand‑1K and RL4RS show that MIS approaches the generation quality of an autoregressive CGA baseline while achieving 4.07× and 3.54× faster CPU decoding, respectively. In a randomized evaluation on a large‑scale social‑video platform, MIS improves the realization metric by 0.104% and watch time by 0.058%, while reducing ranking‑service CPU consumption by 9.73%. These results demonstrate that prefix‑aware multi‑item scoring can narrow the quality gap between NAR and AR generation while retaining the efficiency required for industrial scalability.\n\nReview