The supply chain sits as a directed graph across several tiers, and it goes below the company level: down to the individual material and the production line that depends on it. If a supplier fails or a vulnerability lands, what follows from it is who that reaches, along which paths and in what order.
A temporal graph neural network reads the graph as a whole: it learns from past events which failures actually travel, along which edges and with what delay. Propagation is then not a set of rules imposed on the supply chain but a result drawn from it.