Two observations about the relationship between sound patterns in the phonological grammar of a language were foundational in motivating the proposal of a prosodic hierarchy with trees defined over phonological constituents in the 1980s: (i) sets of multiple, disparate patterns—suprasegmental and segmental, the application of phonotactic restrictions and processes— cluster together over domains; (ii) the patterns are in containment relationships, whereby the distribution of (non)application of one process invariably implies the (non)application of others. However, work on prosodic and intonational phonology in the past decades has raised doubts about the universality of clustering and containment. At the same time, we argue that Autosegmental-Metrical analyses of intonational phonology have slipped into narrowing the scope of sound patterns used to motivate phonological constituents to the distribution of prosodic boundary tones. We explore the consequences of the strong hypothesis that a span of segmental material is a phono- logical constituent if and only if it is delimited by at least one boundary tone. We show that clustering and containment can be understood to be at least partially respected under this hypothesis. But adopting it might also lead us to miss diversity in the organization of phonological patterns of natural language.