AI 中文总结
该研究针对智能体存在的隐式陈旧依赖问题，提出StateAuditor反向审计方法，在STALE基准上实现显著性能提升，验证了转换机制对修复IPA缺口的有效性。
AI 中文摘要
具备记忆增强功能的智能体可能知晓用户存储的状态已过时，却仍基于旧值进行规划，STALE基准将此称为隐式策略适配（IPA）缺口。我们确定了一个结构性成因：基于草稿的验证机制仅检查响应内容，而在开放式响应中，陈旧依赖通常未被明确表述。因此StateAuditor采用反向审计，从存储状态指向草稿。大型语言模型（LLM）会基于带时间戳的证据提出候选新旧转换；确定性代码将每个引用绑定到单一条目，验证新证据确实更新，且仅允许这些经验证的转换触发修复。验证的是来源和时间顺序，而非语义替代。在STALE的完整协议（400个场景、50个会话历史、每个查询对应一个独立响应）下，严格单查询VTA得分达.736，而在相同评估者下，我们的锁定前体模型得分为.686，配对增益为+5.0个百分点（95%置信区间[+2.9, +7.2]），该增益几乎全部来自IPA和前提抗性（PR）。来自第三方模型家族的基准自身评估者也重现了这一增益（.738对比.680）。在独立的跨家族偏好演化基准HorizonBench上，基于黄金衍生结构化存储的完整草稿-审计-修复流程提升了当前偏好准确率（用户聚类p<.01），不过匹配对照显示，大部分外部增益来自草稿侧审计本身；更困难的创作生命周期集未产生增益，在控制错误无效的同时限定了主张范围。相反，在STALE上，匹配对照（相同证据、适配器和调用预算）仅得.692（较前体+0.6，无统计学显著性），这将STALE的增益归因于转换机制，而非新增上下文或调用。我们不主张通用智能体记忆的相关结论。
英文摘要
Memory-augmented agents can know that a user's stored state is outdated and still plan around the old value. The STALE benchmark calls this the implicit policy adaptation (IPA) gap. We identify one structural contributor: draft-anchored verification checks what a response says, and in an open-ended response the stale dependency is usually unsaid. StateAuditor therefore audits in the opposite direction, from stored state to draft. An LLM proposes candidate old-to-new transitions from timestamped evidence; deterministic code pins each quotation to a single entry, checks that the new evidence really is newer, and lets only these verified transitions trigger repair. What is verified is provenance and chronology - not semantic supersession. On STALE's full protocol (400 scenarios, 50-session histories, one independent response per query), strict single-query VTA scores .736 against .686 for our locked predecessor under the same judge: a +5.0-point paired gain (95% CI [+2.9, +7.2]) coming almost entirely from IPA and premise resistance (PR). The benchmark's own judge, from a third model family, reproduces the gain (.738 vs. .680). On an independent cross-family preference-evolution benchmark (HorizonBench), the full draft-audit-repair pipeline over a gold-derived structured store raises current-preference accuracy (user-clustered p<.01), though a matched control shows most of this external gain is the draft-side audit itself; a harder authored lifecycle set gives no gain, bounding the claim while false invalidation stays controlled. On STALE, by contrast, a matched control (same evidence, adapter, and call budget) scores only .692 (+0.6 over the predecessor, n.s.), attributing the STALE gain to the transition machinery rather than added context or calls. We make no claim about general-purpose agent memory.
Comments9 pages, 4 figures; supplementary material in ancillary files
