Personally, I’m skeptical of synthetic DIDs unless we have good reason to believe that there’s something wrong with the donor pool of controls, as the pre-trends are unrealistically precise.
But I would be more convinced by both approaches than one in isolation.
