# Calibration-Free Early Stopping for Chain-of-Thought Reasoning via E-processes

DevFeed: [Calibration-Free Early Stopping for Chain-of-Thought Reasoning via E-processes](<https://devfeed.tech/articles/calibration-free-early-stopping-for-chain-of-thought-reasoning-via-e-processes-93444.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.25212>)

Author: Dong-Jae Lee, Gyojin Han, Junmo Kim

Published: 2026-10-21T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [sequential-testing](<https://devfeed.tech/topics/sequential-testing.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [chain-of-thought](<https://devfeed.tech/tags/chain-of-thought.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [framework](<https://devfeed.tech/tags/framework.md>), [inequality](<https://devfeed.tech/tags/inequality.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [signal](<https://devfeed.tech/tags/signal.md>), [stop](<https://devfeed.tech/tags/stop.md>), [tokens](<https://devfeed.tech/tags/tokens.md>)

## AI overview

The framework stops chain-of-thought generation by accumulating sequential evidence, avoiding held-out threshold calibration. It runs two e-processes on a shared per-chunk signal: a commit stream tracks the signal's position relative to its slow trend, while an abstain stream tracks its signed velocity; they are anti-correlated by construction. Under signal-level nulls on per-chunk increments, Ville's inequality gives an anytime-valid signal-level false-alarm bound at user-chosen significance levels. The abstract reports that across various model-dataset combinations, the method matches or exceeds calibration-based stopping on the token-accuracy Pareto front with zero calibration data, and transfers the same significance level to settings where calibration is infeasible.

## Source excerpt

arXiv:2610.25212v1 Announce Type: cross Abstract: Reasoning large language models (LLMs) benefit from test-time scaling, motivating adaptive policies that spend tokens when they help and stop early otherwise. Existing rules halt when a confidence or stability signal crosses a threshold, but such thresholds rarely transfer across models and datasets, and validation-calibrated rules lose their guarantee under shifts in test-time horizon or signal dynamics. We reframe adaptive stopping as sequential evidence accumulation along the reasoning trajectory rather than threshold calibration on held-out data. Our framework runs two e-processes on a shared per-chunk signal: a commit stream tracks the signal's position relative to its slow trend, and an abstain stream tracks its signed velocity, making the two anti-correlated by construction. Under signal-level nulls on the per-chunk increments, Ville's inequality yields an anytime-valid signal-level false-alarm bound at user-specified significance levels, with no held-out calibration. Across various model--dataset combinations, the framework matches or exceeds calibration-based stopping on the token--accuracy Pareto front using zero calibration data, and the same significance level transfers unchanged across cells where calibration becomes infeasible.