PathCal: State-Aware Reflection-Marker Calibration for Efficient Reasoning

2026-05-21

AI Summary

This work addresses a critical limitation in existing test-time inference control methods, which conflate reflection tokens of distinct functionalities and overlook their heterogeneous roles and timing-sensitive effects, thereby failing to guide large language model reasoning effectively. The study is the first to reveal the functional and temporal heterogeneity of reflection tokens and introduces a training-free decoding controller that dynamically estimates the distribution of such tokens to assess local competition intensity between the current trajectory and alternative branches. By differentiating token types under uncertainty and softly adjusting their logits, the method enables state-aware, lightweight path calibration. Evaluated across six reasoning benchmarks, it achieves a superior efficiency–accuracy trade-off—reducing generation length without additional sampling or external verifiers while maintaining or even improving accuracy.

Abstract

The emergence of Large Reasoning Language Models (LRMs) has paved the way for tackling complex reasoning tasks through test-time scaling by generating long-form Chain-of-Thought (CoT) trajectories during inference. Meanwhile, these trajectories often contain explicit reflection markers such as ``wait'', ``but'', and ``alternatively'', signaling hesitation, revision, and the consideration of alternative explorations, respectively. Recent studies on test-time control leverage such markers as lightweight handles for steering reasoning, typically treating them as a single coarse-grained category rather than distinguishing their distinct functional roles. In this paper, we conduct type-wise suppression and fixed-prefix intervention, revealing that reflection markers differ not only in their functional roles but also in when they exert the greatest influence. Specifically, different marker classes affect accuracy and generation length in distinct ways, and marker choices are most consequential before the model settles into a stable reasoning trajectory. Motivated by these findings, we introduce PathCal, a novel training-free decoding controller that calibrates reasoning paths by distinguishing marker types and intervening only at locally uncertain states. At each decoding step, PathCal utilizes the distribution over reflection-markers to estimate local competition between maintaining the current reasoning trajectory and initiating a competing branch, and softly rebalances marker logits when competing-branch evidence becomes excessive. Experiments across six reasoning benchmarks demonstrate that PathCal achieves a better efficiency--performance trade-off, improving or preserving accuracy while reducing generation length, without relying on external verifiers or additional sampling.

Problem

• reflection markers
• reasoning calibration
• Chain-of-Thought
• test-time control
• Large Reasoning Language Models

Innovation

• reflection markers
• state-aware calibration
• test-time control
• reasoning efficiency
• Chain-of-Thought