●builderYou must account for memory-induced reasoning distortions when implementing long-term memory for your LLM applications.
●researcherThis benchmark shifts focus from simple retrieval accuracy to the qualitative impact of memory on model reasoning.