Answers are grounded in this paper and cite their sources.
Long-running agentic work commonly appears continuous only while one conversation, process, provider, or context window remains available. When that surface ends, summaries and prompts attempt to reconstruct what mattered. They may preserve narrative direction while losing exact artifacts, unresolved alternatives, authority boundaries, custody, external state, intermediate evidence, and the reasons a transition was admitted or refused. This paper proposes Cold Boot Continuity as a deployment primitive: the capacity for a newly instantiated runtime, beginning without the predecessor's conversational memory, to recover enough verified semantic, artifact, task, custody, authority, environment, and evidence state to resume the same governed work under an explicit claim ceiling. Cold boot does not mean bit-identical internal cognition, deterministic continuation, or automatic authority transfer. A replacement model may reason differently, lack required capabilities, discover that the archive is stale, or legitimately refuse to continue. Continuity is instead preservation of identified obligations, exact heads, admissibility decisions, open alternatives, work custody, authority constraints, evidence debts, and revision lineage through a verified resumption protocol. The paper distinguishes bootability, orientability, resumability, semantic continuity, authority validity, and evidenced continuity. It specifies a context capsule as a bounded projection from the Governed Meaning Base (GMB), not as the whole corpus or an executable grant. It also defines failure modes, adversarial tests, the universal realization funnel, and integration with project lineage, receipted reasoning, context governance, and work custody. This is a conceptual architecture and verification program. The source names a JN reactor and a historical Constitutional Crystallization Framework (CCF); this revision applies the author-confirmed terminology lineage CCF -> CMB -> GMB while preserving JN reactor identity as an unresolved system candidate. No named implementation is certified by this paper.
No takes yet. Share an insight, caveat, or question.
Adam Ableman (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: