I think all of these studies fail to account for the credit component. i.e. I can see that this man I just met has one cookie, but now he's promising me another cookie (which I may or may not be able to see). And then if I do what he says, he'll give me two cookies. What probability do I assign to the chances he can deliver on his promises? Maybe he's a liar. Maybe before I completed the assigned task, he came across a better deal and allocated all his cookies.
To your point, there are multiple assessments being made, many of which not being accounted for in the original.
Does the listener fully comprehend "the rules" as they're being laid out?
The listener is evaluating the trusthworthiness of the speaker?
The listener may evaluate their own skills in pulling off a deception by taking the marshmallow and lie about it.
Due to "the rules" laid out by the speaker, does the listener consider they may change "the rules" (influenced by their historical experience with adults)?
Does the listener place any value on a 'marshmallow' at all, maybe a toy, or a type of item previously identified as having high value would lead to different results?
Adjusting for variables in the 'fuzzy' sciences can be difficult due to the innate subjectivity.
