I liked your reframing of the prisoners dilemma as a kidnap dilemma. It is true, most game theory studies are stateless and model reciprocity as a series of transactions, which is subtly different. There also tends to be no observer effect, or at least I haven't seen a study. The observer effect would tend to propagate trustability status ahead of actual transactions. It may have the effect of adding some enforcement to the golden rule by decreasing the effectiveness of bad actors. It isn't entirely clear-cut however, a bad actor strategy may be able to target good actors more effectively as well. Nash's work partially extends the agent model to groups. I should catch up on research in the field and/or run my own simulations. I think there is untapped utility as applied to trust networks. Its not rocket science after all ; )

Nice post, thanks.