The adaptive capabilities of the agents in the community act as a counterbalance to some of the problems of breaking up traces of cooperative behavior for storage in the distributed collective memory. When solving a problem, the agents check collective memory for plans which have indices which are ``similar" to the current setting. From these, the agents select the plan fragment which accomplishes the largest number of current goals. It is possible for the agents to select different plan fragments since they each individually determine what is ``similar." When this happens agreement is reached in the exact same manner as when the plans were generated from first principles; i.e., cooperation is achieved for competing versions of cooperating behavior via communication.
In another problem-solving session that both agents consider similar, HL and HTO retrieve the improved fragments produced in Section 3. In the subsequent action, HL can anticipate the requests of HTO. In this case, the plan fragment is not yet optimal since the agents expect to unload the boxes in the same order that they loaded them instead of in the reverse order. During execution, they unload them in reverse (as they must). During the subsequent analysis, collective memory is updated to reflect this so in the future the chunk will be optimal. By reusing her chunk, the heavy lifter can simplify her cooperative interaction with the hand-truck operator and forego the reasoning involved in forming other plans.
Figure 6 shows the end of the second execution trace and the corresponding improvement.