Self-Verification Dilemma: Experience-Driven Suppression of Overused Checking in LLM Reasoning

TL;DR

This work analyzes self-verification in large reasoning models, revealing that most rechecks are confirmatory and rarely improve outcomes, leading to inefficiency. It introduces Experience-Driven Self-Verification (EDS), a test-time controller that uses an offline experience pool to estimate whether a current recheck is likely to be beneficial, suppressing redundant checks via a lightweight signal. Across multiple models and math benchmarks, EDS reduces reasoning length by up to 20.3% while maintaining or slightly improving accuracy, without altering model parameters. The approach offers a practical, scalable path to more efficient and reliable LLM reasoning by leveraging past verification outcomes to guide future checks.

Abstract

Large Reasoning Models (LRMs) achieve strong performance by generating long reasoning traces with reflection. Through a large-scale empirical analysis, we find that a substantial fraction of reflective steps consist of self-verification (recheck) that repeatedly confirm intermediate results. These rechecks occur frequently across models and benchmarks, yet the vast majority are confirmatory rather than corrective, rarely identifying errors and altering reasoning outcomes. This reveals a mismatch between how often self-verification is activated and how often it is actually useful. Motivated by this, we propose a novel, experience-driven test-time framework that reduces the overused verification. Our method detects the activation of recheck behavior, consults an offline experience pool of past verification outcomes, and estimates whether a recheck is likely unnecessary via efficient retrieval. When historical experience suggests unnecessary, a suppression signal redirects the model to proceed. Across multiple model and benchmarks, our approach reduces token usage up to 20.3% while maintaining the accuracy, and in some datasets even yields accuracy improvements.

Figures (8)

• Figure 1: Reflective behaviors commonly observed in step-by-step mathematical reasoning. We illustrate three categories: rethink, where the model revises its strategy and explores an alternative line of reasoning; and recheck, where the model verifies already-derived intermediate results through re-computation, arithmetic checking, or optimality validation. Recheck behaviors can be further divided into confirmatory cases, which only confirm previous results, and corrective cases, which identify errors and alter the reasoning outcome. Our empirical analysis shows that recheck behaviors overwhelmingly fall into the confirmatory category.
• Figure 2: Percentage of steps classified as reflections.
• Figure 3: Proportion of reflection types for Qwen3-8B annotated by GPT-5. More results are presented in Appendix \ref{['sec:appendixB']}. Human evaluation results are presented in Appendix \ref{['sec:appendixC']}.
• Figure 4: Overview of the experience-driven verification control framework. Recheck episodes are collected and annotated offline to form an experience pool, which captures how often rechecks succeed in correcting errors for specific mathematical manipulations or formula-level derivations, e.g. taking derivatives. The experience pool will be consulted at the test-time to suppress redundant self-verification during LLM reasoning.
• Figure 5: Accuracy–efficiency trade-off.