RoutePrism: Tracing Construction Order Effects in Agent Memory

2026-09-28

AI Summary

This study addresses the vulnerability of agent memory construction to processing order, which often leads to evidence loss and impedes error localization. To tackle this, we propose RoutePrism, a diagnostic protocol that pioneers order-sensitivity tracing and matching intervention mechanisms based on fixed-variable control. By comparing dual-sequence constructions to isolate the impact of individual steps, the method integrates source-level difference tracing with multi-condition intervention experiments for systematic analysis. Our findings demonstrate that recovering critical records improves accuracy by over 60%, revealing survivor selection bias as the primary driver of failure. These insights provide both theoretical grounding and methodological support for optimizing agent memory strategies.

Abstract

Processing the same records in a different order can discard different evidence, yet endpoint accuracy alone cannot reveal what changed or whether it mattered. We introduce RoutePrism, a diagnostic protocol that builds memory twice from the same source pool in two processing orders, then traces which sources, compiled contexts, and answers differ. Because record content, timestamps, policy, and the answer model all stay fixed, any observed difference is localized to the memory construction step. A matched four-condition intervention tests whether a record displaced by reordering actually carried task-relevant evidence: restoring that single record recovers over 60 percentage points of lost accuracy, while substituting a non-supporting record of equal length does not. We evaluate the protocol on PersonaMem-32K (63 primary queries, 29 users) and 470 LongMemEval-S questions with histories spanning 38 to 62 sessions, replicating the core intervention across five answer models. Survivor selection, defined as the choice of which record a cluster retains, drives most source-level changes, while different memory policies (compaction, bounded recency, MemoChat-style summarization, A-MEM) produce distinct failure signatures at the source, context, and metadata layers.

Problem

• agent memory
• construction order
• evidence loss
• memory processing
• diagnostic evaluation

Innovation

• Diagnostic Protocol
• Memory Construction Order
• Survivor Selection
• Intervention Testing
• Agent Memory