The incremental derivations of phase theory (Chomsky, 2001, 2005, etc.) raise the question of the directionality of derivations (den Dikken, 2018). In this squib, I outline some arguments for top-down generation, arguing that grammatical theory must treat as axiomatic the notion that derivations mirror actual structure-building processes in the brain; given the limited nature of temporary memory capacity (Cowan, 2015), bottom-up generation does not survive long under such scrutiny. Next, I briefly review some theory-internal evidence (Phillips, 1996, 2003; Chesi, 2007, 2015) with respect to conflicting results from constituency tests. Finally, I seek to characterize the phases with respect to what Chomsky (2005, 2007) calls the ‘duality of semantics’, arguing that roots are inserted in the upper phase (CP) for discourse purposes, but are interpreted with respect to the core proposition in the lower phase (vP).