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

Evan: Welcome to the Daily Paper Cast.

Ashley: Today's paper is from the Hugging Face daily paper list of November 11, 2026, and has 23 upvotes.

Evan: The title is 'Don’t Just Trust Your Memory: Evidence-Linked Memory Reduces Memory-Induced Degradation in Large Language Model Agents.'

Ashley: The first two authors are Yukyeong Lee and Yumin Choi from KAIST.

The corresponding author is Jinheon Baek, also from KAIST.

Evan: Alright, Ashley, could you start us off with some background on this paper?

Ashley: Sure thing.

Large language model, or LLM, agents increasingly solve multi-step tasks by interacting with an environment.

External memory allows these agents to improve from experience without updating their parameters.

This means they can distill trajectories of past tasks into concise lessons for new tasks.

Evan: That's interesting.

But, how does this memory affect the agents' performance?

Ashley: Well, memory can actually lead to what's called memory-induced degradation.

This is where an agent fails on tasks it would otherwise solve without memory, often because it follows a lesson that is incorrect, misapplies or misreads a valid one.

This problem arises because lessons are inferred from trajectories labeled only with task success or failure.

They're typically provided without the trajectories they came from, leaving the agent no way to check them.

Evan: But how does this paper propose to address that issue?

Ashley: To tackle this, the authors introduce WikiTrace, an evidence-linked memory system.

It's designed to let agents check each lesson against its original trajectory.

The idea is that a trajectory often holds the evidence needed to catch an erroneous lesson.

Therefore, 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.

Evan: So, with WikiTrace, the agent wouldn't just rely on the lesson but could go back and verify the context in which the lesson was learned.

That seems pretty robust.

Ashley: Exactly.

When starting a new task, the agent begins by considering all insights and whenever an insight lacks the details needed to act, conflicts with another, or originates from a failed task, it can descend through its links to fill in missing details or correct the insight before acting.

Evan: And how effective is this approach?

Did the authors validate it?

Ashley: Yes, they did.

The authors validated WikiTrace on five benchmarks, showing that it consistently outperforms existing memory approaches.

It also reduces memory-induced degradation significantly.

In their experiments, WikiTrace not only demonstrated better performance compared to methods that solely provided insights but also excelled in reducing the rate of regression, where memory turns successes into failures.

Evan: This could be quite impactful for the development of more reliable LLM agents.

Revisiting the source of a lesson before blindly following it seems like a smart approach.

Ashley: Indeed, it addresses critical issues in agent learning by ensuring that reused experiences are verified and applicable to their current tasks.

The study pushes the boundaries of how LLM agents can effectively leverage past experiences while minimizing risks.

Evan: That brings us to the end of the introduction section of this paper.

Evan: So, Ashley, let's break down the methodology.

How did the authors construct and validate WikiTrace?

Ashley: WikiTrace was built to address the issues stemming from memory-induced degradation in LLM agents.

The approach organizes memory into three layers linked together: insights, compilations, and raw trajectories.

Let's delve into each of these components.

Evan: Sure, let's start with the insights layer.

What are they exactly?

Ashley: Insights are the high-level lessons that the agent can use immediately.

They are concise and actionable pieces of information derived from past trajectories, and each insight includes details on what to do and when it applies.

Evan: So, these are the distilled lessons.

But what about the compilations?

Ashley: Compilations are essentially a summary of the trajectory, explaining why a particular run succeeded or failed.

They include key decisions and supporting observations that are critical for understanding the insight's validity and contextual applicability.

Evan: Got it.

And the raw trajectories?

Ashley: Raw trajectories are the full logs of actions, observations, and reasoning from past runs.

These offer a complete picture and allow the agent to verify and cross-reference the insights with what actually happened during the task.

Evan: So, when the agent faces a new task, it starts with all these insights.

What happens if it encounters a problem with an insight?

Ashley: Whenever an insight lacks details needed to act, conflicts with another, or originates from a failed task, the agent descends through the layers with terminal tools like bash or read.

This allows it to recover the missing details or revise the insight before acting.

Evan: Fascinating.

So, each lesson is connected back to its source, permitting real-time validation.

What kind of benchmarks did the authors use to validate their approach?

Ashley: The authors validated WikiTrace on five agentic benchmarks comprising eight suites.

These benchmarks span simulated text worlds like ALFWorld for household tasks, ScienceWorld for science experiments, PlanCraft for Minecraft crafting, as well as tool-use environments like AppWorld and τ3-bench Retail, which involves interaction through application APIs.

Evan: That covers quite a broad spectrum of tasks.

How did WikiTrace perform across these varied benchmarks?

Ashley: WikiTrace achieved the best performance on each suite, consistently outperforming existing memory methods.

On average, it improved performance by 10.7 points over the agent without any memory and by 3.2 points over a baseline given the same insights alone.

Additionally, it significantly reduced memory-induced degradation, with regression dropping from 6.5% to 4.5% on average.

Evan: That’s quite impressive.

What metrics were used to measure this performance?

Ashley: The study measured accuracy, gain, and regression rates separately.

Gain refers to the rate at which memory turns failures into successes, highlighting improvement in tasks the agent previously failed.

Regression measures the rate at which memory turns successes into failures, indicating reliability of reused experiences.

Evan: That distinction between gain and regression is critical.

How did WikiTrace handle tasks of varying difficulty?

Ashley: WikiTrace's advantage was most pronounced on harder tasks.

The agent descended more frequently on these tasks, in 44% of runs on hard tasks compared to 22% on easy ones.

The results show that source runs help significantly on harder tasks, allowing the agent to pull more relevant information when the challenge is greater.

Evan: Interesting.

And how did the agent decide when to descend through the layers?

Ashley: Each descent was triggered by specific issues, such as a blocked attempt, an observation that contradicted the agent's plan, or doubt about an insight.

Routine checks of past runs were the most common trigger, covering 38% of descents, but the largest advantage came when a blocked attempt sent the agent looking for a specific fact.

Evan: Can you share an example of how descending through source runs helped the agent?

Ashley: Sure, one notable example is from PlanCraft.

An insight from a failed run incorrectly ruled out a valid procedure.

The agent revisited the successful run linked to the insight, identified that the procedure actually worked, revised the insight, and succeeded.

This case highlights how raw trajectories enable agents to reassess interpretations and apply lessons accurately.

Evan: That sounds like a robust validation method.

It shows the practical benefits of having access to original trajectories as evidence.

Ashley: WikiTrace allows agents to verify and adjust insights dynamically, preventing the pitfalls of misapplied lessons and ensuring that experiences are reliably reused.

This leads to more effective and trustworthy LLM agents.

Evan: Great, that covers the method section of this paper.

Evan: Let's dive into the experiment results.

How was WikiTrace validated and what were the key findings?

Ashley: The authors validated WikiTrace on eight suites across five benchmarks, covering simulated text worlds like ALFWorld, ScienceWorld, and PlanCraft, as well as tool-use environments like AppWorld and τ3-bench Retail.

The first step in their experimental setup was to collect a fixed corpus of past runs without memory.

This dataset was then used to construct the evidence-linked memory structure.

Evan: And how did WikiTrace perform on these benchmarks?

Ashley: WikiTrace consistently achieved the highest performance on every suite tested.

On average, it improved performance by 10.7 points over the agent without any memory and outperformed a baseline given the same insights alone by 3.2 points.

This shows that the added access to source compilations and raw trajectories significantly enhances the agent's task-solving capabilities.

Evan: That's impressive.

So, how did WikiTrace affect the rate of memory-induced degradation?

Ashley: The results showed a marked reduction in memory-induced degradation.

Specifically, the regression rate dropped from 6.5% with insights alone to 4.5% with WikiTrace.

This means fewer instances where the agent failed tasks it would have solved without memory.

In a striking contrast, the rerun baseline, which measures outcome changes without memory, exhibited a regression rate of 9.7%.

Evan: That's quite a drop.

How about the gain rate, which measures the improvement on previously failed tasks?

Ashley: WikiTrace also excelled here.

On average, its gain rate was 35%, which is notably higher compared to 27% for the baseline insights alone and 7 to 14% for existing memory methods.

This substantial improvement can be attributed to the agent's ability to validate and adjust insights dynamically by revisiting their source evidence.

Evan: That's quite a benefit.

Did the authors analyze how WikiTrace performed on tasks of varying difficulty?

Ashley: Yes, they did.

WikiTrace showed its biggest advantage on harder tasks.

The agent descended more often on these tasks, at a rate of 44% compared to 22% for easy tasks.

This indicates that the linked evidence was especially beneficial when navigating more complex challenges, ensuring that even under difficult circumstances, the agent could pull relevant information.

Evan: Interesting.

So, the system's real-time validation and adjustment capability must have been crucial during those harder tasks.

Ashley: Exactly.

Each descent or validation touches on critical issues, such as blocked attempts, contradictory observations, or doubts about insights.

The largest improvements came when a blocked attempt led the agent to search for specific facts in past runs – a scenario where having access to detailed, linked memory proved invaluable.

Evan: Was there any particular benchmark where WikiTrace's benefit was most pronounced?

Ashley: WikiTrace performed exceptionally well across all benchmarks, but the gains were most notable in environments requiring nuanced reasoning and complex interaction, like PlanCraft and ScienceWorld.

In PlanCraft, for example, revisiting the source of insights allowed the agent to correct errors and apply lessons accurately, leading to better task completion rates.

Evan: This really underscores the system’s robustness.

By ensuring that every reused experience is verified, WikiTrace significantly boosts the reliability of LLM agents.

Ashley: Indeed.

The study's results confirm that evidence-linked memory systems like WikiTrace can revolutionize how agents learn from past experiences.

By mitigating the risks associated with memory-induced degradation, WikiTrace paves the way for more effective and trustworthy LLM agents.

Evan: That brings us to the end of the Experiment section of this paper.

Evan: Ashley, let's explore the related work that sets the stage for WikiTrace.

What foundation did the authors build on?

Ashley: Sure, Evan.

The paper highlights several key areas of related research, starting with experience-driven procedural memory for language agents.

Language agents are increasingly converting past execution experiences into procedural memory for guiding future decisions.

This involves retaining verbal feedback, successful trajectories, accumulating executable skills, and distilling experiences into reusable guidelines, manuals, and workflows.

Evan: So, these procedural memories sort of form the do's and don'ts for future tasks?

Ashley: Exactly.

Various methods focus on distilling, selecting, and updating experiences for reuse.

For instance, prior approaches update reasoning memories with new successful and failed experiences, developing mechanisms for efficiently distilling and selecting relevant experiences.

Evan: That sounds like a complex system.

How does it all play into reliable reuse of unverified memory?

Ashley: That's a crucial point.

Reusing past experiences doesn’t always improve an agent’s performance; in fact, agents might ignore or misinterpret condensed experiences.

There are issues such as retrieval competition and negative transfer during continual learning, which can amplify unsafe behavior when reusing execution-oriented experiences from benign tasks.

Evan: So, without proper validation and context, these experiences could mislead the agent?

Ashley: Exactly.

This has led researchers to evaluate failures introduced by memory alongside tasks it helps solve.

The paper by Choi et al., 2026, for instance, validates synthetic practice trajectories before storage to ensure only useful experiences are retained.

Evan: How does WikiTrace navigate this complex terrain of verifying and updating memories?

Ashley: WikiTrace complements these approaches by reassessing stored lessons during task execution, linking each lesson to its source evidence.

This allows the agent to examine the basis and applicability of the lessons in real-time, ensuring that they are justified for the current task.

Evan: That real-time validation seems essential.

How about the structural aspect of memory management?

Ashley: Memory management involves organizing past experiences and providing efficient access during task execution.

Hierarchical approaches play a key role here, organizing experiences at multiple levels of detail for compression, planning, and retrieval.

Approaches like HORMA connect filesystem summaries to raw interaction history and learn to select minimal sufficient context.

Evan: Beyond organization, how do agents decide when and what to retrieve from memory?

Ashley: Active-access methods are pivotal.

They determine when and what to retrieve during reasoning.

For example, Direct Corpus Interaction offers a general approach to corpus access through terminal tools.

These mechanisms allow agents to access and reassess procedural lessons dynamically.

Evan: So, bringing together these organizational and active-access approaches, how does WikiTrace stand out?

Ashley: WikiTrace integrates these concerns by linking lessons directly to their source execution.

Its structure preserves the connection between a lesson and its original experience, while its access mechanism enables real-time reassessment of guidance during task execution.

This holistic approach leverages both procedural memory and dynamic validation to enhance decision-making reliability.

Evan: It's impressive how WikiTrace synthesizes earlier methodologies into a comprehensive system.

Any last thoughts on related work?

Ashley: The integration of reliable reuse, hierarchical memory structuring, and dynamic access in WikiTrace highlights the evolution of LLM agent memory mechanisms.

By addressing the gaps identified in related work, WikiTrace presents a robust framework for improving performance and reliability in LLM agents.

Evan: Alright, that brings us to the end of the Related Work section.

Evan: Ashley, let's summarize the key contributions and takeaways from this paper.

Ashley: The key contribution of this paper is the introduction of WikiTrace, an evidence-linked memory system designed to reduce memory-induced degradation in LLM agents.

By linking each lesson to its source trajectory, WikiTrace allows the agent to verify the context and accuracy of the lessons in real-time before applying them to new tasks.

Evan: That's a game-changer in terms of reliability and performance.

Ashley: Precisely.

The authors demonstrated that WikiTrace outperforms existing memory approaches across various benchmarks.

The approach significantly improves task performance and reduces the rate of regression, which is critical for maintaining reliability.

Evan: By providing a mechanism for real-time validation and adjustment of insights, WikiTrace enhances the robustness of LLM agents, ensuring that reused experiences are both relevant and correctly applied.

Ashley: Exactly.

This paper advances the field by integrating procedural memory with dynamic evidence-based validation.

The approach leverages hierarchical memory structuring and active-access methods to enhance decision-making reliability in complex tasks.

Evan: That’s a wrap for today's episode.

Thanks for tuning into the Daily Paper Cast.

Ashley: We hope you found our discussion on WikiTrace enlightening and useful.

Don't forget to join us tomorrow for another deep dive into cutting-edge research.

Stay curious, everyone!

Evan: Take care, and see you next time!