PHI // DRIFT introduces persistent memory architecture for language models with five core components including Decision Memory Unit and behavioral continuity metric

Researchers present PHI // DRIFT, a cognitive middleware architecture adding persistent internal state across user interactions. The system includes a Decision Memory Unit (DMU) that weights retrieved memories using time-decay, reinforcement, and contextual salience rather than vector similarity alone.

The architecture introduces the Persistence-Embodiment-Drift Index (PEDI), a five-component falsifiable metric for behavioral continuity across context windows, plus a homeostatic regulation layer modeling seven internal state variables with setpoints and drift rates, a security defense layer guarding against four attack classes, and a logic chain reasoning trace system preventing repeated failed approaches through query fingerprinting.

Ablation testing showed DMU re-ranking injects 14.8% more context into prompts than cosine-only retrieval. Live stress testing at 50-thread concurrency produced 100% success rate across 50 requests. All development occurred on consumer hardware including a Dell Inspiron 5543 desktop with no dedicated GPU.