Research · Researchers and educators
Research previews, technical notes, and testable proposals for researchers, engineers, and evaluators studying Emergent Memory Systems.
Research

Consolidation Leakage in Team Memory
Consolidation leakage: summaries and learned procedures in shared AI memory can expose restricted sources. Why record-level checks miss it, and which defenses to test.
Read post
Benchmarking Non-Deterministic Memory for Language Models and Humans Alike
LoCoMo and LongMemEval measure retrieval accuracy over a largely deterministic, frozen store in single-pass evaluations. For probabilistic memory like ACT-R, we propose an alternative: continuous calibration of dynamic memory as it evolves over time, evaluating trust deficits across three axes—memory writing, ranking, and retrieval.

Jacobian-Causal Memory Control for AI Memory
A research preview on Jacobian-causal memory control: treating persistent AI memory recall as an estimated, observed, and auditable inference-time intervention.
Read post
Activation Is Not Querying
Activation-guided graph retrieval treats memory availability, graph traversal, and evidence reconstruction as distinct from similarity search in long-lived AI systems.
Read post