Despite recent progress giving lower bounds on the second eigenvalue for directed configuration models, there is a fundamental gap between lower bounds on the spectral gap and an exact characterization of mixing behavior. For undirected sparse random graphs the cutoff phenomenon is now fairly well understood through the work connecting spectral gaps to hitting times of tree-like neighborhoods, but the directed setting introduces non-reversibility, which breaks the standard spectral tools. Non-reversible chains do not have a simple variational characterization of the spectral gap, the singular value decomposition of the transition matrix replaces eigenvalue analysis in a complicated way, and typical directed random graphs have highly asymmetric local structure. These obstacles mean that even knowing the second eigenvalue to high precision does not immediately translate into a mixing time with cutoff because the relationship between spectrum and total variation distance requires new techniques in the non-reversible case.