This study models cryptographic auditing of off-chain data as a Constrained MDP (CMDP) under partial observability: the hidden type and corruption state of the storage node render the problem a POMDP, while a miss-rate ceiling \rho imposes an explicit security constraint. We propose DRQN-CMDP, a Deep Recurrent Q-Network whose GRU layer maintains a belief over the latent node type, paired with Lagrangian dual ascent that automatically adapts the miss-rate penalty \lambda. A pairing-free homomorphic-MAC primitive ensures O(1) on-chain verification cost. Among the 13 methods compared—four DQN variants, PPO, A2C, PPO-Lagrangian, a stateful Bayesian heuristic, three fixed-rule baselines, and an oracle-informed heuristic—DRQN-CMDP achieves a favorable balance: 83% lower gas costs than fixed high-frequency auditing, a single-digit miss rate (7.5%), and moderate detection latency—a combination unmatched by any other method across all three objectives.
Blogger's Review: This paper significantly enhances the efficiency of off-chain data auditing by integrating deep learning and reinforcement learning techniques, achieving a commendable balance between cost and accuracy. The innovative dynamic penalty adjustment mechanism of DRQN-CMDP showcases the potential of modern machine learning in security domains. Future research could explore applications in more complex environments.