Sleep–Wake Orchestration in Hierarchical LLM Cohorts

<p>This paper proposes a formal architecture for <strong>self-optimizing large language model ensembles</strong> that alternate between <em>wake</em> (active inference) and <em>sleep</em> (targeted fine-tuning) phases. Grounded in the <strong>Adjoint Projections on Computational Hierarchies</strong> framework, the system treats cohorts of models as dynamically rotating through resource-bounded cycles: roughly one-third of compute devoted to retraining on <em>informational gaps</em>—regions of the task space poorly covered by peers yet solvable by higher-level models—while two-thirds remain active in production. The paper defines the underlying mathematics of <strong>projection–collapse adjunctions</strong>, introduces a <strong>gap metric</strong> for selecting training data, and presents an implementation plan with canary rollouts, envelope-based resource management, and safety guarantees. The approach connects biological sleep consolidation with continual meta-learning in artificial systems, offering a path toward <strong>autonomous, energy-efficient LLM ecosystems</strong>.</p>