# Diagnosing Memorization in Chain-of-Thought Reasoning, One Token at a Time

arXiv: https://arxiv.org/abs/2508.02037 · HTML: https://mlsanity.com/paper/2508.02037/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 12 of its 29 references that are in the map.
- First posted on arXiv 4 Aug 2025.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

## Builds on (most cited first, 12 of 12)

- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Improve Mathematical Reasoning in Language Models by Automated Process Supervision](https://mlsanity.com/paper/2406.06592/) (2024)
- [2 OLMo 2 Furious](https://mlsanity.com/paper/2501.00656/) (2024)
- [On Memorization of Large Language Models in Logical Reasoning](https://mlsanity.com/paper/2410.23123/) (2024)
- [Generalization v.s. Memorization: Tracing Language Models' Capabilities Back to Pretraining Data](https://mlsanity.com/paper/2407.14985/) (2024)
- [Disentangling Memory and Reasoning Ability in Large Language Models](https://mlsanity.com/paper/2411.13504/) (2024)
- [AI as Humanity's Salieri: Quantifying Linguistic Creativity of Language Models via Systematic Attribution of Machine Text against Web Text](https://mlsanity.com/paper/2410.04265/) (2024)
- [VersaPRM: Multi-Domain Process Reward Model via Synthetic Reasoning Data](https://mlsanity.com/paper/2502.06737/) (2025)
- [The Reasoning-Memorization Interplay in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2503.23084/) (2025)
- [None of the Others: a General Technique to Distinguish Reasoning from Memorization in Multiple-Choice LLM Evaluation Benchmarks](https://mlsanity.com/paper/2502.12896/) (2025)
- [Quantifying In-Context Reasoning Effects and Memorization Effects in LLMs](https://mlsanity.com/paper/2405.11880/) (2024)
- [Judging a Book By its Cover](https://mlsanity.com/paper/1610.09204/) (2016)

## Built on by (most cited first, 4 of 4)

- [Reasoning or Retrieval? A Study of Answer Attribution on Large Reasoning Models](https://mlsanity.com/paper/2509.24156/) (2025)
- [Quantifying the Effect of Test Set Contamination on Generative Evaluations](https://mlsanity.com/paper/2601.04301/) (2026)
- [RIDE: Difficulty Evolving Perturbation with Item Response Theory for Mathematical Reasoning](https://mlsanity.com/paper/2511.04120/) (2025)
- [What Really Controls Temporal Reasoning in Large Language Models: Tokenisation or Representation of Time?](https://mlsanity.com/paper/2603.19017/) (2026)

## Abstract (arXiv)

Large Language Models (LLMs) perform well on reasoning benchmarks but often fail when inputs alter slightly, raising concerns about the extent to which their success relies on memorization. This issue is especially acute in Chain-of-Thought (CoT) reasoning, where spurious memorized patterns can trigger intermediate errors that cascade into incorrect final answers. We introduce STIM, a novel framework for Source-aware Token-level Identification of Memorization, which attributes each token in a reasoning chain to one of multiple memorization sources - local, mid-range, or long-range - based on their statistical co-occurrence with the token in the pretraining corpus. Our token-level analysis across tasks and distributional settings reveals that models rely more on memorization in complex or long-tail cases, and that local memorization is often the dominant driver of errors, leading to up to 67% of wrong tokens. We also show that memorization scores from STIM can be effective in predicting the wrong tokens in the wrong reasoning step. STIM offers a powerful tool for diagnosing and improving model reasoning and can generalize to other structured step-wise generation tasks.