When Stored Evidence Stops Being Usable: Scale-Conditioned Evaluation of Agent Memory

TL;DR

This paper introduces a scale-conditioned evaluation protocol for agent memory that assesses how evidence usability degrades as irrelevant sessions accumulate, providing nuanced diagnostics across different memory interfaces.

Contribution

It presents a novel evaluation framework that measures memory reliability under evidence-preserving growth, revealing complex failure regimes and supporting scalable-memory claims.

Findings

Reliability loss varies with agent and interface.

HippoRAG maintains call budget but loses reliability as irrelevant sessions increase.

Memory failures depend on agent size and interface, not a single phenomenon.

Abstract

Memory-agent evaluations report fixed-snapshot accuracy or retrieval quality, but these scores do not show whether evidence remains usable as irrelevant sessions (sessions not annotated as task-relevant evidence for the query) accumulate. We present a scale-conditioned evaluation protocol for agent memory under evidence-preserving growth: for each query, task evidence is held fixed while irrelevant sessions are added. The protocol logs agent--memory trajectories and reports four diagnostics: budget-compliant reliability, tail memory-call burden, failure-regime decomposition, and the usable-scale boundary where reliability falls below the target. Applied to LongMemEval and LoCoMo across flat, planar, and hierarchical memory interfaces, the protocol shows reliability loss is not a single phenomenon. On LongMemEval, HippoRAG stays within the two-call budget but loses 16--20 percentage points in budget-compliant reliability as irrelevant sessions are added; LiCoMemory's observed failures depend strongly on the agent, with Qwen3-8B exceeding the budget while Qwen3-32B and Qwen3-235B remain reliable in the tested range. The result supports a framework for making scalable-memory claims conditional on agent, interface, scale range, and interaction budget.