Consider a diffusion spreading through a network. Given a snapshot of the history, can the starting point be determined? I will discuss the ideas and problems surrounding this question in two contexts: for simple random walk / Brownian motion, and rumour spread in social networks. For random walks, I will sketch how the theory of self-intersections lends a hand; for rumour spread, I will present the state-of-the-art, an algorithm called adaptive diffusion, discuss its shortfalls, and suggest a path forward.