Supervisory Quantum Neural Networks: An Integrated Architecture with Four Distinct Constraints

This paper specifies a supervisory quantum neural network architecture that separates four distinct constraints: informational and operational isolation through no-write-back (W=0), support-restricted scrambling, quantum-nondemolition-compatible monitoring where quantum measurement of the protected core applies, and Markovian interface dissipation. The architecture treats the training loop itself as a potential write path and distinguishes protected-core update authority from scrambling support, measurement backaction, and open-system dissipation. Explicit assumptions and falsifiable failure conditions are provided for each constraint, together with candidate formal proof obligations. Existing work in quantum neural network training, protected quantum subsystems, open-system dynamics, quantum monitoring, agentic-system governance, security monitoring, and quantum formal verification supports individual components and verification methods but does not establish their integration. The work presents an engineering specification and does not claim a completed machine-checkable verification or integrated physical implementation.