Allomorphy is currently the focus of intense investigation within syntactic approaches to morphology, since this phenomenon is constrained by locality conditions that are most probably syntactic in nature. Within Distributed Morphology (DM), two main locality conditions are entertained for an allomorphic interaction between a trigger and a target to be possible: membership of the same Spell-Out domain (a hierarchical condition) and adjacency (a linear condition). The standard assumption is that the first condition is sufficient enough to constrain the size of the domain in which allomorphic interactions may take place. In this paper I show that things are actually more complicated once we take into account two factors. First, the difference between so-called Grammatically Conditioned Allomorphy (GCA) and Phonologically Conditioned Allomorphy (PCA). Second, the difference between three possible contexts of interaction: 1) within the phonological word, 2) between phonological words, and 3) between the clitic and its host or cliticizing site. While context 2 is currently being investigated, little attention has been paid to context 3, in particular when set in contrast with context 1. Specifically, I observe that while PCA seems to always involve a more embedded (inner) trigger across all three contexts (to the extent that it is actually attested in context 2), the pattern for GCA varies from context to context: it allows both an outer and an inner trigger in context 1, it is not attested in context 2, and it seems to allow only an outer trigger in context 3. Focusing on contexts 1 and 3, I show how the standard theory of allomorphy in DM does not account for all the patterns, and I develop a version in which M-Words, that is, maximal heads, are cycles for Vocabulary Insertion.