Post-hoc feature attribution methods are progressively being employed to explain decisions of complex machine learning models. Yet, it is possible for practitioners to obtain a diversity of models that provide very different explanations to the same prediction, making it hard to derive insight from them. In this work, instead of aiming at reducing the under-specification of model explanations, we fully embrace it and extract logical statements about feature attributions that are consistent across multiple models with good performance. We show that a partial order of feature importance arises from this methodology enabling more nuanced explanations by allowing pairs of features to be incomparable when there is no consensus on their relative importance. We prove that every relation among features present in these partial order also holds in the rankings provided by existing approaches. Finally, we present use cases on three datasets where partial orders allow one to extract knowledge from models despite their under-specification.