I study the implications of a key fact: that many different global value chain (GVC) networks aggregate up to the same multi-country input-output data. I argue in favor of networks where the use of inputs varies depending on the use of output - in contrast to the current literature based on networks where all output uses the same inputs. I provide evidence for this approach using Mexican customs data: cars exported to the U.S. use a higher share of imported American inputs than those exported elsewhere. This heterogeneity matters since both quantitative counterfactual estimates and measures of globalization such as value-added trade vary across GVC networks. I argue that GVCs are better measured when leveraging additional information: incorporating Mexican customs data implies that 30% of U.S. imported Mexican manufactures is U.S. value returning home - higher than the conventional estimate of 17%.