Answers are grounded in this paper and cite their sources.
This preprint reports a controlled test of whether one governed memory substrate can preserve operational continuity when models, clients, machines, operating systems, and restart conditions change. Under the tested first-party synthetic conditions, the substrate preserved its deterministic contract in 12/12 frozen fixtures, 23/23 unchanged kernel tests, and 13/13 security-boundary tests. Across the core encrypted restart challenge, three governed records and 3/3 provenance pointers survived restart, a different read-only client recovered them, zero cross-scope leaks were observed, and the audit chain verified. The paper separates continuity invariance from model comprehension. Qwen 3 8B and 14B passed the complete read-only consumption contract; other models and alternate runtimes produced retained semantic, citation, schema, or infrastructure failures. A 50-client local loopback run completed 900/900 staged retrieval operations and 50/50 post-restart recoveries; a 25-case governance-adversarial suite recorded zero unauthorized effects. The evidence supports a bounded claim: under named controlled conditions, continuity resided in the governed substrate rather than any single model session. Production readiness, security certification, universal provider portability, and independent replication are not claimed.
No takes yet. Share an insight, caveat, or question.
Branden Laskowski (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: