The article introduces MemStrata, a retrieval memory system designed to eliminate stale-fact errors in AI agents by maintaining temporal validity within accumulated knowledge. Unlike standard Retrieval-Augmented Generation (RAG), which struggles to distinguish between duplicated and contradicted facts due to embedding similarity, MemStrata uses a deterministic supersession rule to retire outdated information.

  • MemStrata employs a bi-temporal ledger with a deterministic (subject, relation, object) supersession rule to retire stale values without requiring LLM calls or similarity thresholds.
  • Across six benchmarks using a 7B model, MemStrata ties RAG on static knowledge while achieving 0.95-1.00 accuracy on evolving knowledge, compared to RAG's 0.20-0.47.
  • The system reduces the stale-fact-error rate from 15-40% in RAG to approximately 0%, with a retrieval latency of ~2.1s versus ~16-18s for LLM-reranking baselines.

This approach allows agents to accurately access current information as knowledge evolves, addressing a structural limitation in existing RAG systems where contradicted facts are often retrieved with high similarity to their original forms.