让下一篇重要论文，不再错过。

原文摘要

Online advertising platforms balance advertiser value, publisher-assessed ad quality, and platform revenue. However, boost design in this three-stakeholder architecture remains insufficiently studied. In this paper, we develop certified boost mechanisms under a payment rule that deducts boosts from winners' payments. We establish a capped benchmark-competitive condition that guarantees a quality-aware liquid-welfare lower bound parameterized by the certificate strength, boost cap, and bid cap, with an explicit penalty for an insufficient boost cap. We then derive a closed-form canonical boost. For the observed state, it matches the welfare-optimal benchmark allocation whenever its certificate margin seats every benchmark winner, and among all certified boosts that do so it maximizes the platform's retained revenue. On AuctionNet, a public benchmark built on Alibaba's real-world production advertising environment, we evaluate canonical across ten bidding policies on multiple episodes. Under a twofold shock to the strongest rival's bid, canonical achieves $97.6\%$ mean welfare efficiency, compared with $91.1\%$ for the strongest baselines. Controlled stress tests link this protection to the certificate margin, while a ten-policy cap sweep shows diminishing welfare gains as retained revenue falls.