Bouncing completion of eternal inflation

Using a purely kinematical argument, the Borde-Guth-Vilenkin (BGV) theorem states that any maximal space-time with average positive expansion is geodesically incomplete, hence past eternal inflation would be necessarily singular. Recently, discussions about the broadness of this theorem have been resurfaced by applying it to new models and/or challenging the space-time maximality hypothesis. In the present work, we use reference frames of non co-moving observers and their kinematical properties in order to inquire into the nature of such possible singular beginnings. Using the spatially flat de Sitter (dS) space-time as a laboratory, this approach allows us to exhaust all possibilities bounded by the BGV theorem in the case of general spatially flat Friedmann-Lema\^{\i}tre-Robertson-Walker (FLRW) geometries. We show that either there exists a scalar or parallelly propagated curvature singularity, or the space-time must be past asymptotically dS (with a definite non-zero limit of the Hubble parameter when the scale factor becomes null, hence excluding certain cyclic models) in order to be extensible. We are able to present this local extension without violating the null energy condition, and we show that this extension must contain a bounce. This is a mathematical result based on purely kinematical arguments and intuition. The possible physical realization of such extensions are also discussed. As a side product, we present a new chart that covers all de Sitter space-time.