The relaxed-based heuristics typically use unary operators to get the job done. Unfortunately, the simplification process isn't exactly sound for non-deterministic domains. This has only been observed in the probabilistically interesting toy domain of climber, but it may appear elsewhere. The simplification process has been disabled for now, but there should be a way to re-introduce this in the future. For example, we could only remove an operator if the dominating one is deterministic.