ReaPER+

In 1-qubit small-rotation compilation, ReaPER+ achieved the highest success probability at every tested fidelity tolerance. ReaPER+ reduced the episodes needed for 2-qubit ZZ(pi) approximation compared with fixed ReaPER, PER, HER, and PPO. In 1-qubit HRC compilation, ReaPER+ matched the best success and fidelity while using fewer gates than fixed ReaPER and PER. ReaPER+ anneals replay priority from TD-error prioritization toward reliability-aware replay during training. The annealing schedule is motivated by unreliable early reliability estimates and more informative later estimates. ReaPER+ is not claimed to dominate universally across all environments and schedules.