First, a new method is introduced for extracting gradient subcategorical (e.g., within-heavy) syllable weight hierarchies from parsed metrical corpora. The method is illustrated for Sanskrit and Finnish. For example, in both languages, VC and VV rimes are categorically heavy, but in both, distributional asymmetries reveal that the poets treat VC as significantly lighter than VV. Second, it is demonstrated that gradient weight can be used to probe syllable/interval structure, e.g., whether the onset is parsed with the preceding (as in Steriade 2009) or following (as traditionally) nucleus. Finally, two independent case studies in Epic Sanskrit metrical statistics are presented for which the assumption of discrete syllabification of segments is problematic. Both cases support a model of syllabification in which singleton segments are gradiently parsed between syllables, akin to ambisyllabicity, only with gradient bifurcation rather than categorical sharing.