Nuro: Non-Stationary Optimization of Epistemic Artifacts Under Provenance-Preserving Compositional Constraints in Agentic Runtime Environments

                   
  ▎ We present Nuro, a formal framework addressing the   
  open problem of epistemic artifact lifecycle management
   within non-stationary agentic runtime environments.   
  The framework relaxes distributional stationarity    
  assumptions through a compositional architecture      
  enforcing provenance-preserving constraints on all 
  state transitions within a hierarchically stratified  
  artifact space. We formalize the selection problem as
  posterior sampling over heterogeneous, rank-invariant
  scoring functionals and introduce an evidence-gated
  refinement protocol satisfying monotonicity invariants.
   The system has been validated in continuous production
   deployment since March 2026.