In a repeated prisoner's dilemma with some probability δ of continuing after each round, a Subgame Perfect Nash Equilibrium may be found which induces cooperation instead of defection in each round. However, in a repeated prisoner's dilemma with a fixed, finite number of rounds, defection occurs in the last round and unravels all the way back to the first round via backward induction.
How minimal can the uncertainty about the future be while still permitting such a cooperative strategy? For example, if prisoner's dilemma will be played for at least 19 rounds, and then a random event with probability δ will occur to determine whether a 20th round will be played, is that enough to break the backward induction conclusion that defection is inevitable? How fickle is the backward induction conclusion to the inclusion of randomness at single points of the game?