Commonly weights are viewed as monolithic quantities which can be estimated from data, elicited from experts or are known a priori. In this paper we question this assumption, positing that frequently weights reflect a series of underlying processes, each effectively contributing a component of the overall quantity making up the weight in the given aggregation or reasoning context. As such, we consider weights as a set of entangled components, defining them using the framework of compositional data, a field of statistics designed specifically to model components of a whole. We proceed by focusing on two well established processes–and thus weight components–in the field of data aggregation: the weighting of sources–underpinning linear averages, and the weighting of evidence–as underpinning ordered weighted averages. Using these two aggregation processes, we demonstrate how compositional geometry can be used to disentangle weights estimated from data according to the underlying processes and discuss the impact of doing so – including an improved capability for explaining both the weights and the weighting approaches.