@inproceedings{chen-etal-2026-explicit,
title = "Explicit State Elicitation Is Not Enough: A Controlled Audit of Memory-Policy Classification",
author = "Chen, Yihang and
Qian, Pin and
Wang, Su and
Peng, Chong and
Xu, Huan and
Li, Shuaiting and
Sun, Yiqi",
editor = "Kamalloo, Ehsan and
Gontier, Nicolas and
Rozanov, Nikolai and
Hashemzadeh, Maryam and
Arora, Akhil",
booktitle = "Proceedings of the Second Workshop for {RE}search on Agent Language Models ({REALM} 2026)",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.realm-1.19/",
pages = "271--294",
ISBN = "979-8-89176-485-9",
abstract = "Personalized agents must decide whether retrieved user memory should be used, ignored, updated, or queried before it affects a task. We develop an empirical audit protocol for structured intermediate outputs: auditing dataset shortcuts, isolating bundled prompt changes, testing answer-associated labels and decomposed evidence, and retaining provider-level execution failures. A 480-example synthetic development set initially suggested large gains from a state-structured prompt bundle, but TF-IDF diagnostics showed lexical separability and no positive standalone Ignore cases. We therefore build a frozen 160-example controlled counterfactual set with 40 matched four-way families and rule-derived reference policies. There, exposing four state definitions improves accuracy, but adding an isolated explicit state-output field does not significantly improve policy accuracy for Llama-3.3-70B and gives only a marginal, non-significant gain for GPT-OSS-120B. Supplying benchmark-associated state labels shifts policy predictions; because those labels deterministically map to policies, this is label-conditioning, not evidence of a faithful mechanism. Family-level and seed-stability analyses show that example accuracy overstates counterfactual consistency. A decomposed semantic-evidence follow-up does not improve routing for the cleanly evaluated endpoint, while the GPT-OSS condition fails provider-side request validation. We evaluate policy classification only."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chen-etal-2026-explicit">
<titleInfo>
<title>Explicit State Elicitation Is Not Enough: A Controlled Audit of Memory-Policy Classification</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yihang</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pin</namePart>
<namePart type="family">Qian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Su</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chong</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Huan</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shuaiting</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yiqi</namePart>
<namePart type="family">Sun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Second Workshop for REsearch on Agent Language Models (REALM 2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ehsan</namePart>
<namePart type="family">Kamalloo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nicolas</namePart>
<namePart type="family">Gontier</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nikolai</namePart>
<namePart type="family">Rozanov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Maryam</namePart>
<namePart type="family">Hashemzadeh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Akhil</namePart>
<namePart type="family">Arora</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-485-9</identifier>
</relatedItem>
<abstract>Personalized agents must decide whether retrieved user memory should be used, ignored, updated, or queried before it affects a task. We develop an empirical audit protocol for structured intermediate outputs: auditing dataset shortcuts, isolating bundled prompt changes, testing answer-associated labels and decomposed evidence, and retaining provider-level execution failures. A 480-example synthetic development set initially suggested large gains from a state-structured prompt bundle, but TF-IDF diagnostics showed lexical separability and no positive standalone Ignore cases. We therefore build a frozen 160-example controlled counterfactual set with 40 matched four-way families and rule-derived reference policies. There, exposing four state definitions improves accuracy, but adding an isolated explicit state-output field does not significantly improve policy accuracy for Llama-3.3-70B and gives only a marginal, non-significant gain for GPT-OSS-120B. Supplying benchmark-associated state labels shifts policy predictions; because those labels deterministically map to policies, this is label-conditioning, not evidence of a faithful mechanism. Family-level and seed-stability analyses show that example accuracy overstates counterfactual consistency. A decomposed semantic-evidence follow-up does not improve routing for the cleanly evaluated endpoint, while the GPT-OSS condition fails provider-side request validation. We evaluate policy classification only.</abstract>
<identifier type="citekey">chen-etal-2026-explicit</identifier>
<location>
<url>https://aclanthology.org/2026.realm-1.19/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>271</start>
<end>294</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Explicit State Elicitation Is Not Enough: A Controlled Audit of Memory-Policy Classification
%A Chen, Yihang
%A Qian, Pin
%A Wang, Su
%A Peng, Chong
%A Xu, Huan
%A Li, Shuaiting
%A Sun, Yiqi
%Y Kamalloo, Ehsan
%Y Gontier, Nicolas
%Y Rozanov, Nikolai
%Y Hashemzadeh, Maryam
%Y Arora, Akhil
%S Proceedings of the Second Workshop for REsearch on Agent Language Models (REALM 2026)
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-485-9
%F chen-etal-2026-explicit
%X Personalized agents must decide whether retrieved user memory should be used, ignored, updated, or queried before it affects a task. We develop an empirical audit protocol for structured intermediate outputs: auditing dataset shortcuts, isolating bundled prompt changes, testing answer-associated labels and decomposed evidence, and retaining provider-level execution failures. A 480-example synthetic development set initially suggested large gains from a state-structured prompt bundle, but TF-IDF diagnostics showed lexical separability and no positive standalone Ignore cases. We therefore build a frozen 160-example controlled counterfactual set with 40 matched four-way families and rule-derived reference policies. There, exposing four state definitions improves accuracy, but adding an isolated explicit state-output field does not significantly improve policy accuracy for Llama-3.3-70B and gives only a marginal, non-significant gain for GPT-OSS-120B. Supplying benchmark-associated state labels shifts policy predictions; because those labels deterministically map to policies, this is label-conditioning, not evidence of a faithful mechanism. Family-level and seed-stability analyses show that example accuracy overstates counterfactual consistency. A decomposed semantic-evidence follow-up does not improve routing for the cleanly evaluated endpoint, while the GPT-OSS condition fails provider-side request validation. We evaluate policy classification only.
%U https://aclanthology.org/2026.realm-1.19/
%P 271-294
Markdown (Informal)
[Explicit State Elicitation Is Not Enough: A Controlled Audit of Memory-Policy Classification](https://aclanthology.org/2026.realm-1.19/) (Chen et al., REALM 2026)
ACL