I started reading this paper expecting to find solutions to the problem of survivability and predictability of automated negotiator algorithms in an unreliable, underlying message-passing infrastructure, and found something interesting, but much less relevant.

The work that is described consists of the definition of a framework that allowed the authors to implement and test different negotiators. The two main components of the environment have been specified in pseudocode following an event-driven paradigm: a supervisor in charge of managing the environment, and negotiators that have been defined in an abstract way, allowing for the implementation of diverse algorithms and the introduction of varied behaviors (survival behaviors are just a possibility; they also tested differences between two negotiation strategies). Both pure reactive and mixed reactive-proactive behaviors are allowed. Survivability has been implemented as a proactive behavior, but it has been dealt with in a way that is too simplistic.

Using this environment, you can run simulations and study patterns of behavior. The authors conducted some experiments combining different types of algorithms, and they found a quite obvious result (that the algorithms with survival behaviors survived better), and an apparently not so obvious one (that communities of negotiators with survival behaviors performed more efficiently and with higher stability).

The authors argue, and I agree, that it is important to be able to reason about and predict the behaviors of complex negotiation algorithms. Unfortunately, though, the proposed framework, although useful, does not solve this problem. It only allows one to experiment in order to try to discover these behaviors.