In short each node would have its own risk management profile. Entering a transaction using any settlement path or asset type (even multi-hop transactions involving multiple paths or assets) is the explicit decision of the actor in question, against that risk profile, available information, and its own priorities (eg. must complete by X date, prefer to avoid Y actor/system/asset, etc.)

Because the properties of each settlement network, asset type or actor are able to be formally defined, cross-system automation is possible and new systems with interesting properties are able to present themselves on a fair and equal footing.

Of course, with such lofty goals, it was never finished (I was keen but asked to direct attentions to urgent areas elsewhere), but recently I have begun to revisit it owing to my study of physical logistics networks and scheduling algorithms for my current startup http://8-food.com/