It is remarkably difficult to reconcile unitarity and Vilenkin’s wave function. For example, the natural conserved inner product found in quantum unimodular gravity applies to the Hartle-Hawking wave function, but fails for its Vilenkin counterpart. We diagnose this failure from different angles (Laplace transform instead of Fourier transform, non-Hermiticity of the Hamiltonian, etc.) to conclude that ultimately it stems from allowing the connection to become imaginary in a section of its contour. In turn this is the unavoidable consequence of representing the Euclidean theory as an imaginary image within a fundamentally Lorentzian theory. It is nonetheless possible to change the underlying theory and replace the connection’s foray into the imaginary axis by an actual signature change (with the connection, action and Hamiltonian remaining real). The structural obstacles to unitarity are then removed, but special care must still be taken, because the Euclidean theory a priori has boundaries, so that appropriate boundary conditions are required for unitarity. Reflecting boundary conditions would reinstate a Hartle-Hawking-like solution in the Lorentzian regime. To exclude an incoming wave in the Lorentzian domain one must allow a semi-infinite tower of spheres in the Euclidean region, wave packets traveling through successive spheres for half an eternity in unimodular time. Such a “Sisyphus” boundary condition no longer even vaguely resembles Vilenkin’s original proposal.