# Reliable Control-Point Selection for Steering Reasoning in Large Language Models

Source: [Reliable Control-Point Selection for Steering Reasoning in Large Language Models](<https://www.sota2.com/research/paper/69cf49bd0636e5bf4afcdbd1>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-04T05:00:11.701Z

Authors: Haomin Zhuang, Hojun Yoo, Xiaonan Luo, Kehan Guo, Xiangliang Zhang

Published: 2026-04-02T14:48:56.000Z

## Abstract

Steering vectors offer a training-free mechanism for controlling reasoning behaviors in large language models, but constructing effective vectors requires identifying genuine behavioral signals in the model's hidden states. For behaviors that can be toggled via prompts, this is straightforward. However, many reasoning behaviors -- such as self-reflection -- emerge spontaneously and resist prompt-level control. Current methods detect these behaviors through keyword matching in chain-of-thought traces, implicitly assuming that every detected boundary encodes a genuine behavioral signal. We show that this assumption is overwhelmingly wrong: across 541 keyword-detected boundaries, 93.3\\% are behaviorally unstable, failing to reproduce the detected behavior under re-generation from the same prefix. We develop a probabilistic model that formalizes intrinsic reasoning behaviors as stochastic events with context-dependent trigger probabilities, and show that unstable boundaries dilute the steering signal. Guided by this analysis, we propose stability filtering, which retains only boundaries where the model consistently reproduces the target behavior. Combined with a content-subspace projection that removes residual question-specific noise, our method achieves 0.784 accuracy on MATH-500 (+5.0 over the strongest baseline). The resulting steering vectors transfer across models in the same architecture family without re-extraction, improving Nemotron-Research-Reasoning-1.5B (+5.0) and DeepScaleR-1.5B-Preview (+6.0). Code is available at https://github.com/zhmzm/stability-steering.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Mathematical Reasoning on MATH-500 (Accuracy, Delta, Token Length, Ratio)](<https://www.sota2.com/research/sota/mathematical-reasoning-on-math-500-accuracy-delta-token-length-ratio.md>): Mathematical Reasoning; MATH 500; Accuracy: 81.6.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.02113>)

[PDF](<https://arxiv.org/pdf/2604.02113.pdf>)