I must have not been clear and concise. I apologize for this. I was discussing the scenario where we introduce a set of cards that allow for an arbitrary number of triggers without human involvement. We clearly need to implement a way of handling nested loops, branching, and recursive calls. So the set of cards facilitating this need to either need to do this completely by themselves or construct the behavior given the state of the playing field. I did not assume we can do what is described in the paper (the site) because I have no reason to think the new cards must function in a way that is compatible with their setup. This is why I am interested in the possibility of implicit and bounded recursion.
The fuel comment was just relevant to implementing an infinite loop in a real game. If our supposed 100% autonomous system (with the new cards as mentioned above) relied on an additional resource (such as burning through cards each "tick"), then this wouldn't work in real life but would work if we generalized the size of the deck or available cards. This is not supposed to be a characteristic of Turing Machines in general (or at all). It was just a precaution for attempting to translate an instance of the generalized case to real life.