Bermúdez-Otero (2013) argues that the Spanish lexicon stores whole stems complete with their theme vowels, rather than storing roots whose inflectional class features condition the insertion of particular theme vowels (as argued for in much Distributed Morphology work on Romance, such as Oltra-Massuet 1999; Oltra-Massuet & Arregi 2005). The argument turns on an apparent cyclicity paradox which Bermúdez-Otero (2013:65, 71) dubs the problem of the missing cycle. Bermúdez-Otero argues that this problem cannot be avoided if roots and categorizing heads are held to be atoms stored in the Spanish lexicon. The paradox can be circumvented only by embracing the notion of 'stem' and taking the stem to be the unit stored as a lexical primitive. This conclusion, if correct, would have far-reaching implications for the theory of the architecture of the grammar, since the notion of stem does not and cannot have any status as a primitive in a theory that is committed to the ideas of syntactic hierarchical structure all the way down and the strictly local determination of conditioned allomorphy (such as Distributed Morphology)-see Embick & Halle (2005). The notion that a morphologically complex unit like a stem could be stored, and behave as a unit with respect to conditioned allomorphy, is not compatible with either of these tenets. Therefore, if Bermúdez-Otero's argument is right, it follows that some of the fundamental architectural claims of Distributed Morphology are wrong.