If I provide two model sticks to a physics simulation, it can work out how to put them back together with no more effort than it took to model the break?
Like I can model shooting a cannon ball out of a cannon and it will tell me where it lands. Or I can model a cannon ball sitting on the ground and it will tell me where the cannon was?
A hypothetical perfect simulation should be able to do it both ways indeed!
However, the current consensus take of quantum uncertainty means _if_ such a simulator exists, it cannot be of our universe. Or, more likely, such a perfect simulator does not exist.
Of course all of this is about a hypothetical of a hypothetical at this point...
(This is the same problem as entropy (our current understanding of it). We know it is increasing one-way w.r.t. time, and we can imagine what it means to "reverse entropy" and that there's nothing really theoretically preventing that, but we can't build an actual machine to do that.)
Like I can model shooting a cannon ball out of a cannon and it will tell me where it lands. Or I can model a cannon ball sitting on the ground and it will tell me where the cannon was?