Fortunate Recall: Ontology-Driven Memory Lifecycle Management for Persistent Coherence in LLMs

Personal memory stores in large language models often degrade as facts accumulate without distinction between enduring truths and transient states. Fortunate Recall introduces a composable policy layer classifying facts into a 10+1 behavioral ontology, pairing them with deterministic lifecycle mechanisms like differential decay, event-time validity, and slot-key supersession. On LifecycleBench, its implementation reaches a 76.9% pass rate, outperforming baselines like Mem0 and MemoryOS while cutting downstream confabulation by half. Standard retrieval performance remains uncompromised across LongMemEval-S and BEAM benchmarks.