Extended search and relaxation are presented as negotiated-search methods in the introduction, but how negotiation is used is unclear. It says that these methods are used when an agent recognizes a conflict. The methods seem internal to an agent and do not require any interaction between agents.

      When speaking of solution acceptability, an example of the "best" proposal are those with lowest cost. The authors go on to say that "When this proposal is used as the basis for a complete solution however, all other agents must produce proposals that are higher cost than they might have been with some other base proposal." I see no justification for saying that other agents must produce higher cost proposals. It often happens with greedy algorithms, but it is not a steadfast rule.

      I found the explanation of how the system actually works to be quite vague. In particular, what interactions do the agents have with the Framework Controller and how does this facilitate negotiation between agents? It is unclear what initiates a negotiated-search method in an agent. There are portions of the article that seem to indicate that they are self-initiated inside an agent, called by the Framework Controller based on its decisions on how to integrate agent proposals, and initiated by other agent messages that the Framework Controller simply passes to the target agent.

      The control flow of the system is also confusing. Two phases are mentioned: 1) an agent cycle and 2) a framework cycle. In the agent cycle, what invokes each agent? Are there any commands sent to the agents as they are invoked or is everything the agent needs contained in the shared memory?