跳转至

受控持久化内存：面向长周期智能体的源绑定状态语义与故障关闭释放机制

文章背景与核心概要

本文介绍了“受控持久化内存”（Governed Persistent Memory, GPM），这是一种旨在解决长周期智能体（Long-Horizon Agents）长期记忆管理难题的创新方案。传统的智能体记忆系统多依赖于简单的“选择-存储-检索”机制，这种方式无法有效验证检索到的记录是否存在矛盾、已被取代、撤回、删除或过期等问题，从而导致智能体在决策时可能引用错误信息。

GPM 通过引入可审计的双时态状态转换模型解决了上述痛点。该模型包含源绑定准入、派生生命周期状态、当前公共屏障以及故障关闭结构化释放等核心机制。通过在 GPM-ReleaseBench 基准测试、封闭式端到端服务评估及有限模型验证中的严苛测试，GPM 展现了极高的准确性，不仅消除了基准模型中的所有失败案例，还实现了确定性且高度可靠的输出，且未产生任何回归问题。

摘要

Long-term agent memory is usually treated as select--store--retrieve, but retrieval does not decide whether contradictory, superseded, retracted, deleted, or stale records may support an outgoing claim. We introduce Governed Persistent Memory (GPM), an auditable bitemporal state-transition model with source-bound admission, derived lifecycle state, current public barriers, and fail-closed structured release. Five executable clauses cover ledger integrity, source binding, conflict isolation, non-revival after retraction or deletion, and exact claim closure over a fresh view at one verified head.

长期智能体记忆通常被视为“选择-存储-检索”的过程，但检索机制本身无法判断矛盾、已被取代、撤回、删除或过期的记录是否支持当前的结论。我们引入了“受控持久化内存”（GPM），这是一个可审计的双时态状态转换模型，具备源绑定准入、派生生命周期状态、当前公共屏障以及故障关闭结构化释放功能。五个可执行条款涵盖了账本完整性、源绑定、冲突隔离、撤回或删除后的不可复活性，以及在单一验证头（verified head）下对最新视图的精确结论闭合。

On a prespecified hash-frozen 3,600-case GPM-ReleaseBench, GPM matches all complete outcomes; the strongest of three intentionally simple complete policies matches 1,800/3,600 and makes unmatched releases on 50% of violation cases. A separate sealed end-to-end service evaluation exercises real ingestion and release across eight query families. In its publicly disclosed V3 arm, the governed lane is correct on 2,400/2,400 clusters versus 600/2,400 for ungoverned local Qwen2.5-7B; it repairs all 1,800 baseline failures with no regression (one-sided 95% lower bounds 99.875% and 99.834%). A later V5 reseal over Chinese- and English-command arms, with generation-date pinning and no post-freeze reducer amendment, again obtains 2,400/2,400 per arm. A production-code-independent finite model explores 331,776 semantic and 1,990,656 query states without a full-contract counterexample, and a 100,000-trace three-engine differential yields zero mismatches.

在预先指定且哈希锁定的 3,600 个案例的 GPM-ReleaseBench 测试中，GPM 匹配了所有完整结果；三种故意简化的完整策略中最强的一种仅匹配了 1,800/3,600，并在 50% 的违规案例中做出了错误的释放。一项独立的封闭式端到端服务评估在八个查询族中进行了实际的摄入和释放测试。在其公开披露的 V3 版本中，受控路径在 2,400/2,400 个集群上表现正确，而未受控的本地 Qwen2.5-7B 仅为 600/2,400；它修复了所有 1,800 个基准失败案例且无回归（单侧 95% 下界分别为 99.875% 和 99.834%）。随后的 V5 版本在中文和英文指令集上进行了重新密封测试，通过生成日期锁定且未进行冻结后的归约修正，再次在每个分支上获得了 2,400/2,400 的成绩。一个独立于生产代码的有限模型探索了 331,776 个语义状态和 1,990,656 个查询状态，未发现全契约反例，且 100,000 次追踪的三引擎差异测试产生了零不匹配。

These are bounded contract and implementation results, not open-world model accuracy or evidence of world truth. Governed answers in the sealed service evaluation are deterministic service outputs; the 7B result is the ungoverned comparison, not a claim that a language model itself became perfectly accurate.

这些是受限的契约和实现结果，而非开放世界模型的准确性或世界真理的证据。封闭服务评估中的受控答案是确定性的服务输出；7B 的结果仅作为未受控的对比，并不代表语言模型本身变得完全准确。

访问与资源