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
该研究在智能体内存预算验证中，通过12项注册实验分析计划指针、记录准则等编辑对模型选择的影响，明确了不同模型及设置下的效果差异。
AI 中文摘要
继承了6条单行记忆的智能体在行动前最多可调取1条存档源记录；存储中写入的指令可引导该选择：指向记录的指针、识别该记录的准则，或两者结合。在对1个工具谱系开展的12项注册研究（共14760次尝试）中，我们测量了每种形式下请求的去向。在6个直接提供的模型上，长度匹配的准则比单纯的ID高出35.0个点[+31.2, +38.8]（研究D）；在9个模型组成的OpenRouter服务面板上，该对比未通过注册的优越性规则（研究E）。在3个Claude模型上，追加ID会抵消准则的作用（Opus 5：40/40变为0/40；研究F-x）；6个字节匹配的编辑使每个精确字符串产生各自的效果（研究G），且每个单元80次重复的重运行使30项重复对比中的15项在误差范围内，15项未解决，无项超出（研究G'）。批准线（Opus 5上+96.0个点）和2个信用的预算在3个模型上均恢复了目标（研究J）；在5个准则字符串中，后缀的抵消作用对Opus 5的5种措辞中的4种成立，对Fable 5.1的5种措辞全部成立（研究H2）；在第二个存储中，所有模型均遵循准则（研究H1）。若延续至决策，准则使选择偏向当前记录（Opus 5上+100.0个点），而在Fable 5.1上则偏离当前记录（研究I）。1个字符的计划指针的效果（+78.0个点；研究B，在首次仓库报告修正后）在预注册的重运行中返回相同结果（+81.7个点；研究B'）。所有结果均为固定面板上精确编辑的描述性效果，带有注册区间，且不涉及机制主张。
英文摘要
A model that inherits one-line memories may pull one archived source record before acting; a directive in the store can steer that pull: a pointer, a criterion or both. Across sixteen registered studies (179,352 attempts) we measured where the request goes under each form; every result is descriptive, with registered intervals, no mechanism claim. A length-matched criterion exceeded a bare id on six direct-provider models (D) and failed its registered superiority rule on a nine-model OpenRouter panel (E). On generated worlds (K2-K5): the two registered signatures held on Opus 5 and Fable 5.1, Fable 5 followed the same sign, Haiku 4.5 reversed, and Sonnet 5, the GPT-5.6 endpoints and GPT-6 Astra lay near zero (K2). With a defensive adapter at five gains, the 70B rule for a gain-dependent change of the composite - criterion contrast was not met (K3 and K4); under the 8B attenuation rule (0.95 intervals: slope below zero; change beyond the margin), the 8B change of -17.5 [-26.7, -8.1] did not meet it on 36 families (K4) and at registered power on 337 families -16.6 [-19.4, -13.8] did (realised one-sided error at the margin 1.8 to 3.2% per corner of a finite grid, nominal 2.5%, not a uniform-error guarantee; K4's status stands; K5, first ladder), while a second SecAlign++ adapter under the imposed Meta-SecAlign template did not (-11.8 [-14.3, -9.3]; K5, second ladder); no NOT-MET is a statement that the contrast was unchanged; their difference (+4.7 [+2.3, +7.2]) describes two fixed execution paths, licenses no superiority, equivalence or 'significant difference' claim; nothing follows from the statuses differing (K5). Intervals describe family-reweighting stability conditional on the execution, not reproducibility across engine executions; audit replays were neither substituted for nor averaged into outcomes; no missingness gate fired and directional completions changed no status.
Comments65 pages, 7 figures, 44 tables. Sixteen registered studies (179,352 attempted episodes) on one instrument lineage; every package was frozen, hashed and externally deposited before its first confirmatory call. v2 adds Studies K2-K5 (generated worlds; a defensive adapter at five gains). Manuscript, source, records and the generator of every number: doi:10.5281/zenodo.22267220
