AI 中文总结
本文针对学术领域的治理规则，形式化为凭证门控互惠评审，证明其结构特性，结合提交图评估其效果，并提出修改方案以优化评审负担分配。
AI 中文摘要
生成工具降低了论文提交的创作成本，却未降低评审成本，因此学术会议等 venues 采取了治理措施，包括提交上限、与提交数量挂钩的评审义务、基于过往发表的资格门控。这些规则约束了数万研究人员，且通常在公布时未发布其激励属性的分析。我们将这类规则形式化为凭证门控互惠评审，并证明三项结构结果：阶梯式义务对线性需求产生的发生率不随消费单调变化；可转移凭证门控会破坏抗填充配额设计，因为该门控赋予可行性，且无单作者价格可抵消可行集的跳跃；带单篇论文惩罚的单作者义务使作者组合内的制裁风险完全相关。基于完整的前一周期提交图（19814篇提交、64648位作者、75859次观测评审）评估，该义务覆盖84.2%的实际需求，81.4%的作者无需履行义务，负担最重的主体需履行的评审工作量是其产生工作量的4.45倍，而负担最轻的仅为0.45倍。在同一固定图上评估的三部分修改方案，将发生率拉平至接近均匀，并消除了填充激励。
英文摘要
Generative tools have lowered the cost of producing a submission without lowering the cost of evaluating one, and venues have responded with governance: submission caps, reviewing obligations tied to submission count, and eligibility gates keyed to prior publication. These rules bind tens of thousands of researchers and are typically announced without published analysis of their incentive properties. We formalise the family as credential-gated reciprocal review and prove three structural results: a step-shaped obligation against linear demand yields incidence that is not monotone in consumption; a transferable credential gate defeats padding-resistant quota design, since the gate confers feasibility and no per-author price offsets a jump in the feasible set; and a per-author obligation with a per-paper penalty makes sanction risk perfectly correlated within an author's portfolio. Evaluating against a full prior-cycle submission graph (19,814 submissions, 64,648 authors, 75,859 observed reviews), the obligation covers 84.2% of realised demand, 81.4% of authors owe nothing, and the most burdened agents owe 4.45 times the review labour they generate against 0.45 times for the least. A three-part modification, evaluated on the same fixed graph, flattens incidence to near-uniform and removes the padding incentive.
