Getting fractionalized rebel groups to agree

The New York Times has an op-ed by Robert Frank on how to deal with violent crime, suggesting police view the problem as a dynamic game, where if they announce and credibly target one gang for crackdown, they can quickly achieve success, and then move on to the next gang, all the while credibly threatening to revert to the first gang if they act up, and thus quickly bring about a general reduction in crime as long as the "order" is public knowledge... turtles all the way down kind of thing.  And then over in The New Yorker Jon Anderson deals with a similar problem- ethnographic understanding of the favela gang lords in Rio de Janeiro.

And this reading about warlords and gangsters got me thinking about the opposite problem, which is Special Envoy Gration's problem, of how to sequentially get a bunch of small fractured rebel groups to agree to something.  Of course, the something that they have to agree to is one thing, and then how to get them to agree is the other thing.  Suppose we knew what the something was; say it was that they would publicly announce a chain of command and a structure of allegiances... i.e., who is to be king, and who sub-king, and so on.  Moreover, simultaneously, the king would announce what the conditions were for serious negotiations with Khartoum.  Finally, the whole group would commit to some costly action to show their credibility (like abandoning their positions and massing in some area protected temporarily by UNAMID).  So how to get the fractured groups to agree?  Do you start with the largest group and then work down to smaller and smaller groups?  Or start with the smaller groups, form a coalition, and then approach the larger groups.

In these kinds of problems, it is usually "garbage in, garbage out" style theorizing, in that certain assumptions will get you one direction and other assumptions will get you the other direction, largely because in strategic situations like this modeling the outcome depends on what you assume about the behavior of the actors, and for sure the assumption that they are "rational" present-discounted risk-averting calculators with infinite horizons blah blah is not a good assumption, but that means that there is little basis to choose from alternative assumptions.  But I still wonder whether they might not be some robust (i.e., unable to argue with the logic) result out there.  Or some codification of common sense- maybe something like, "there is no robust algorithm, so just keep experimenting patiently with various formulas until you hit paydirt."