While the analysis shown in the study group uses the normal form of a game, it is often practically possible to turn the extensive form into normal form. You can see this outside reference for more on representing games in different forms: Link . As you point out, the model has to contain all possible moves to lead to correct conclusions. In either extensive form or normal form, if the person doing the modeling hasn't included a relevant possibility, then the conclusions won't be right. As long as you can assign a utility value to strategies that might be considered cheating in the colloquial sense of the word, they can be included as just another strategy available as a choice to agents.
