Structured Belief State and the First Precision-Aware Benchmark for LLM Memory Retrieval

Jeffrey Flynt

Current LLM memory benchmarks evaluate answer quality rather than retrieval accuracy. Consequently, a system that dumps its entire belief store can achieve perfect recall and mask…