The literature about side-channel attacks is very rich. Many side-channel distinguishers have been devised and studied; in the meantime, many different side-channels have been identified. Also, it has been underlined that the various samples garnered during the same acquisition can carry complementary information. In this context, there is an opportunity to study how to best combine many attacks with many leakages from different sources or using different samples from a single source. This problematic has been evoked as an open issue in recent articles. In this paper, we bring two concrete answers to the attacks combination problem. First of all, we experimentally show that two partitionings can be constructively combined. Then, we explore the richness of electromagnetic curves to combine several timing samples in such a way a sample-adaptative model attack yields better key recovery success rates than a mono-model attack using only a combination of samples (via a principal component analysis).