Program Mutation

An exploration-enhancement mechanism in reinforcement learning that generates valid training trajectories by sampling suboptimal programs, applying local mutations under syntactic constraints, and executing them for verification.