In this paper I argue for a violable optimality-theoretic constraint penalizing branching constituency in moras. I examine this constraint in relation to a problem in Buriat syllable structure in which a consonant appears in a position normally reserved for the second vocalic member of a nucleus but does not contribute to syllable weight. In general, the maximal constituency of a Buriat syllable is defined by a CVVC structure, and these syllables are treated as heavy in the stress pattern of the language. This form obeys the properties of typical Buriat syllable structure that vowels may be long or short (long includes diphthongs) and there are no complex onsets or codas.Yet one case appears to be an exception to these generalizations. CVNG (N=a velar nasal, G=a voiced velar stop) is the only syllable form in Buriat with a final consonant cluster. Furthermore, it has two apparently conflicting properties, namely that it is maximal in its constituency, so CVVNG is not possible, yet it patterns as light with respect to stress. Indeed, consonants never contribute to syllable weight in this language. In a derivational model, the CVNG syllable appears to require the elaborated nuclear moraic model of syllable structure proposed by Shaw (1992, 1993); however, I show that with a violable branching mora constraint, the facts of Buriat syllable structure can follow under a simpler model of syllable structure. I argue that this constraint in combination with standard syllable structure constraints and the independently motivated notion of head prominence can predict precisely the range of Buriat syllable forms and their weight without making use of the nuclear moraic model. Concluding discussion shows that the analysis proposed here has interesting implications for both maximal constituency effects and moraic domination of coda consonants.