Somebody's probably already thought of this but I hadn't, so here goes:
Today, we regulate for anti-trust by looking at market share in important industries. Today we regulate banks largely by looking at their total leverage. But, despite the media nattering on about companies that are "too big to fail," that's not really the problem. The problem is that some companies are too connected to fail. If you rip them out of the network of financial and supplier relationships, the damage done cascades through the whole network.
So why not regulate from a graph-theoretic basis? We not only want to know about leverage and market share, we want to know about how many other entities that hold debt or credit, or how many suppliers and customers, will be hurt if a given entity goes belly-up. If we had this information in a standardized reporting structure, regulators could look for excessively connected nodes or nodes whose average network distance was substantially shorter than the network diameter.
These are the companies that are most likely to cause trouble if they fail. Regulators could concentrate on these companies and, if necessary, take preemptive action to reduce the damage they could do to the rest of the network.
No idea whether this is feasible or what the unintended consequences are.