@inproceedings{zhang-etal-2026-reasoning-confidence,
title = "Reasoning Confidence Is Not Policy-Invariant: Calibrating Prefix Success under Model and Budget Shift",
author = "Zhang, Xin and
Wei, Qiyu and
Yuan, Chenhan and
Ananiadou, Sophia",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.387/",
pages = "8875--8891",
ISBN = "979-8-89176-459-0",
abstract = "Inference-time scaling of LLM reasoning relies on a confidence score over each partial trace, yet existing estimators treat prefix quality as intrinsic to the prefix. A given prefix can succeed at very different rates and rankings under different continuation policies. Reasoning confidence is therefore not policy-invariant, and a squared-loss lower bound shows that policy-blind predictors retain residual error proportional to cross-policy disagreement. Calibration must condition on the continuation model and inference budget. To bridge this gap, we propose the Comparable Policy-Conditioned Calibrator (CPCC), built from a shared frozen-LLM prefix encoder, comparable scalar policy features under per-policy z-score normalization, and a Beta{--}Binomial head trained on bootstrap success counts. Across six continuation policies and three reasoning benchmarks (GSM8K, SVAMP, MATH-500), all evaluated under majority-vote aggregation, the CPCC family attains the lowest Brier on every evaluation cell, and CPCC is strongest on the held-out model families. It also avoids the held-out-family collapse of raw-hidden conditioning, lowers ECE by 13 to 26 percent on out-of-distribution math, and reduces downstream selective risk by 10 to 30 percent."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2026-reasoning-confidence">
<titleInfo>
<title>Reasoning Confidence Is Not Policy-Invariant: Calibrating Prefix Success under Model and Budget Shift</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xin</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qiyu</namePart>
<namePart type="family">Wei</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chenhan</namePart>
<namePart type="family">Yuan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sophia</namePart>
<namePart type="family">Ananiadou</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 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</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">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Inference-time scaling of LLM reasoning relies on a confidence score over each partial trace, yet existing estimators treat prefix quality as intrinsic to the prefix. A given prefix can succeed at very different rates and rankings under different continuation policies. Reasoning confidence is therefore not policy-invariant, and a squared-loss lower bound shows that policy-blind predictors retain residual error proportional to cross-policy disagreement. Calibration must condition on the continuation model and inference budget. To bridge this gap, we propose the Comparable Policy-Conditioned Calibrator (CPCC), built from a shared frozen-LLM prefix encoder, comparable scalar policy features under per-policy z-score normalization, and a Beta–Binomial head trained on bootstrap success counts. Across six continuation policies and three reasoning benchmarks (GSM8K, SVAMP, MATH-500), all evaluated under majority-vote aggregation, the CPCC family attains the lowest Brier on every evaluation cell, and CPCC is strongest on the held-out model families. It also avoids the held-out-family collapse of raw-hidden conditioning, lowers ECE by 13 to 26 percent on out-of-distribution math, and reduces downstream selective risk by 10 to 30 percent.</abstract>
<identifier type="citekey">zhang-etal-2026-reasoning-confidence</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.387/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>8875</start>
<end>8891</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Reasoning Confidence Is Not Policy-Invariant: Calibrating Prefix Success under Model and Budget Shift
%A Zhang, Xin
%A Wei, Qiyu
%A Yuan, Chenhan
%A Ananiadou, Sophia
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F zhang-etal-2026-reasoning-confidence
%X Inference-time scaling of LLM reasoning relies on a confidence score over each partial trace, yet existing estimators treat prefix quality as intrinsic to the prefix. A given prefix can succeed at very different rates and rankings under different continuation policies. Reasoning confidence is therefore not policy-invariant, and a squared-loss lower bound shows that policy-blind predictors retain residual error proportional to cross-policy disagreement. Calibration must condition on the continuation model and inference budget. To bridge this gap, we propose the Comparable Policy-Conditioned Calibrator (CPCC), built from a shared frozen-LLM prefix encoder, comparable scalar policy features under per-policy z-score normalization, and a Beta–Binomial head trained on bootstrap success counts. Across six continuation policies and three reasoning benchmarks (GSM8K, SVAMP, MATH-500), all evaluated under majority-vote aggregation, the CPCC family attains the lowest Brier on every evaluation cell, and CPCC is strongest on the held-out model families. It also avoids the held-out-family collapse of raw-hidden conditioning, lowers ECE by 13 to 26 percent on out-of-distribution math, and reduces downstream selective risk by 10 to 30 percent.
%U https://aclanthology.org/2026.emnlp-main.387/
%P 8875-8891
Markdown (Informal)
[Reasoning Confidence Is Not Policy-Invariant: Calibrating Prefix Success under Model and Budget Shift](https://aclanthology.org/2026.emnlp-main.387/) (Zhang et al., EMNLP 2026)
ACL