# Don't Just Trust Your Memory: Evidence-Linked Memory Reduces Memory-Induced Degradation in LLM Agents

**Authors:** Yukyeong Lee, Yumin Choi, Jinheon Baek, Sung Ju Hwang  
**Lab:** KAIST · **Published:** 10 Nov 2026 · **Upvotes:** 65 · **Level:** Intermediate  
**Links:** [arXiv](https://arxiv.org/abs/2611.14320) · [PDF](https://arxiv.org/pdf/2611.14320) · [Code](https://github.com/yukyeongleee/wikitrace) · [This page](https://omen-bit.github.io/ai-paper-daily/paper/2611.14320.html)

## Curator's notes

**TL;DR:** WikiTrace links each memory insight to its original trajectory, letting LLM agents verify and correct lessons, reducing performance drops from faulty memory.

- **What's new:** Introduces evidence-linked memory storing raw trajectories with insights, enabling agents to check and amend lessons on the fly, unlike prior memory that lacks traceability.
- **Key result:** WikiTrace consistently outperforms existing memory approaches on five benchmarks while reducing memory-induced degradation.
- **Why it matters:** Builders can create more reliable LLM agents that learn from experience without harming future task performance.

## Abstract

External memory enables LLM agents to improve from experience by distilling trajectories of past tasks into concise lessons for new tasks. However, memory can also lead to memory-induced degradation, where an agent fails on tasks it would solve without memory, often because it follows a lesson that is incorrect, or misapplies or misreads a valid one. This is because lessons are inferred from trajectories labeled only with task success or failure, yet are typically provided without the trajectories they came from, leaving the agent no way to check them. To address this, we introduce WikiTrace, an evidence-linked memory system designed to let agents check each lesson against its original trajectory. Specifically, motivated by the finding that a trajectory often still holds the evidence needed to catch an erroneous lesson, WikiTrace stores each lesson as a concise insight linked to a summary of why its trajectory succeeded or failed and to the raw trajectory itself. Then, on a new task, the agent starts from all insights and, whenever one lacks details needed to act, conflicts with another, or comes from a failed task, descends through its links with terminal tools to fill in missing details or correct the insight before acting. We validate WikiTrace on five agentic benchmarks, showing that it consistently outperforms existing memory approaches and a baseline given the same insights alone, while also reducing memory-induced degradation. Our code is available at https://github.com/yukyeongleee/wikitrace

## What you could build

1. Implement a tutoring bot that revises its advice by consulting original solution steps.
2. Create a game AI that checks past strategies against recorded play logs before reusing them.
3. Develop a code assistant that validates suggested snippets against the execution traces that generated them.

## Quiz yourself

How does WikiTrace allow an agent to handle a lesson that lacks needed details on a new task?

**Answer:** The agent follows the lesson's links to the summary and raw trajectory, using terminal tools to retrieve missing information or correct the insight before acting.

---

Notes written by AI from the paper, numbers copied from the abstract. Check the paper before you rely on them. From AI Papers Daily.