Reinhard Selten's paper reexamines the concept of perfect equilibrium points in extensive games. He introduces a new definition of perfectness, distinguishing between subgame perfect and perfect equilibrium points. A perfect equilibrium point is always subgame perfect, but a subgame perfect equilibrium may not be perfect. Selten shows that every finite extensive game with perfect recall has at least one perfect equilibrium point. He argues that the normal form is inadequate for representing extensive games and introduces an "agent normal form" as a more suitable representation. The paper defines extensive games with perfect recall, strategies, expected payoffs, and normal form. Kuhn's theorem is restated, showing that in games with perfect recall, behavior strategies can be used instead of mixed strategies. Subgame perfect equilibrium points are defined as equilibrium points that induce equilibrium points on every subgame. A numerical example is presented to illustrate the concept of subgame perfect equilibrium points. It shows that not all intuitively unreasonable equilibrium points are excluded by the definition of subgame perfectness. Selten then introduces a model of slight mistakes, where players have a small probability of making errors, leading to strategic deviations from the equilibrium. The paper defines perfect equilibrium points as limits of equilibrium points in perturbed games. It shows that perfect equilibrium points are subgame perfect and that every perfect equilibrium point is a subgame perfect equilibrium point. The paper also discusses the difference between perfect and subgame perfect equilibrium points, showing that not all subgame perfect equilibrium points are perfect. Finally, the paper discusses the decentralization property of perfect equilibrium points, showing that the question of whether a behavior strategy combination is a perfect equilibrium point can be decided locally at the information sets of the game. Local equilibrium points are defined, and it is shown that in perturbed games, these local conditions are equivalent to the usual global equilibrium conditions. This leads to a decentralized description of a perfect equilibrium point.Reinhard Selten's paper reexamines the concept of perfect equilibrium points in extensive games. He introduces a new definition of perfectness, distinguishing between subgame perfect and perfect equilibrium points. A perfect equilibrium point is always subgame perfect, but a subgame perfect equilibrium may not be perfect. Selten shows that every finite extensive game with perfect recall has at least one perfect equilibrium point. He argues that the normal form is inadequate for representing extensive games and introduces an "agent normal form" as a more suitable representation. The paper defines extensive games with perfect recall, strategies, expected payoffs, and normal form. Kuhn's theorem is restated, showing that in games with perfect recall, behavior strategies can be used instead of mixed strategies. Subgame perfect equilibrium points are defined as equilibrium points that induce equilibrium points on every subgame. A numerical example is presented to illustrate the concept of subgame perfect equilibrium points. It shows that not all intuitively unreasonable equilibrium points are excluded by the definition of subgame perfectness. Selten then introduces a model of slight mistakes, where players have a small probability of making errors, leading to strategic deviations from the equilibrium. The paper defines perfect equilibrium points as limits of equilibrium points in perturbed games. It shows that perfect equilibrium points are subgame perfect and that every perfect equilibrium point is a subgame perfect equilibrium point. The paper also discusses the difference between perfect and subgame perfect equilibrium points, showing that not all subgame perfect equilibrium points are perfect. Finally, the paper discusses the decentralization property of perfect equilibrium points, showing that the question of whether a behavior strategy combination is a perfect equilibrium point can be decided locally at the information sets of the game. Local equilibrium points are defined, and it is shown that in perturbed games, these local conditions are equivalent to the usual global equilibrium conditions. This leads to a decentralized description of a perfect equilibrium point.