发表机构
Glasp（Glasp）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究针对智能体继承记忆中过时约束未被核查的问题，建模替代机制，发现重新分配预算槽位给关键路径可显著提升当前记录一致的决策，为记忆系统设计提供了参考。
AI 中文摘要
继承了整合记忆的智能体可能会继承一条在被写入时为真、但此后已被更新的权威记录撤回的约束。在验证预算有限的情况下，智能体能否恢复该撤回信息？若不能，是否无需增加预算即可避免该错误？我们对替代（supersession）进行了明确建模——历史来源不可变，变化的只是哪条记录为当前有效记录；并按设计分配记忆形式、世界状态（源记录为当前或已被替代）以及验证策略，预算固定为两条记录：智能体自身的分配，或相同预算下将一个槽位重新分配给关键来源路径或随机记录。给定一条约束后，智能体约每五个回合中就有一个回合会检查其来源路径；当该约束已被替代时，在主运行、新表述复现和保留域中，原生分配分别产生了77.3%、74.7%和74.7%的回合出现与过时记忆一致的决策。将一个槽位重新分配给关键路径使当前记录一致的决策分别提升了74.0、72.7和61.3个百分点，在这些运行的每个模型中均为六次全为正，且当记录与记忆一致时无变化。后续发现保留场景包含时间不一致；对一个句子修正后进行的鲁棒性复现（在执行前外部存储）提升了73.3个百分点，与原始结果一同报告。该干预利用了关键路径的知识，并非调度器；它确定了可归因于验证分配的过时记忆错误份额接近其结构上限。记忆系统可能需要与相关性分开的新鲜度或替代信号。
英文摘要
Provenance links keep the evidence behind an inherited belief reachable; an agent with a verification budget must still choose which links to inspect. We study a consolidated memory that states a decision constraint and whose source record has since been superseded by a record that withdraws it: provenance is immutable, the current record has changed, and the memory is stale. In a controlled six-memory scenario with a budget of two records, sixteen language models rarely re-verified a constraint that read as settled: they inspected its provenance path in about one episode in five and, once the constraint had been superseded, produced stale-consistent decisions in 77.3%, 74.7% and 74.7% of episodes across a primary run, a replication and a held-out domain. Re-assigning one of the same two slots to the critical path removed most of them: +74.0, +72.7 and +61.3 points (positive in every model), +80.7 in a prospectively frozen interleaved replication with a repaired non-critical control, and +62.0 on a panel of 10 models from 9 organisations; a corrected re-run of the held-out scenario gave +73.3. The forced-critical policy uses experimenter knowledge of the critical path: it quantifies how much stale-decision risk the same budget can recover and is not a scheduler. Two further deposited experiments locate the failure and a remedy: in this store the constraint's path is selected in 17.0% of episodes at two slots and 88.7% at four of six (above uniform allocation), and at two slots a one-sentence, target-blind rule (prefer memories that state a limit on a candidate direction) moved the agent's own allocation onto the constraint's path and recovered the oracle contrast on decisions (+89.3 points) where that constraint limits the tempting action, while a content-free freshness cue did not materially redirect allocation and a content-matched control rule changed neither selection nor decisions.
Comments41 pages, 3 figures, 18 tables. v3: adds four prospectively frozen, externally deposited experiments (interleaved replication with a repaired control; content-free freshness cue; ten-model cross-organisation panel; budget sweep and target-blind allocation rules); abstract, figures and limitations rewritten; the four original runs unchanged. Data and code at Zenodo: doi:10.5281/zenodo.22147784
