The research agenda is to apply the thesis “no correlation without causation” within the framework of causal modeling to resolve various probability-related puzzles, and I will use the two-envelope paradox as a case study to illustrate how this agenda may be carried out. In the literature on the two-envelope paradox, a distinction is commonly drawn between probabilistic and non-probabilistic versions of the paradox. Attempts to resolve these have typically focused on one version or the other, with some even claiming that a solution to one version cannot address the other. I will argue that, within the framework of causal modeling, it is possible, on the one hand, to clearly lay out the causal assumptions underlying certain proposed solutions in the literature, and, on the other hand, to provide a unified solution to what may initially appear to be two distinct forms of the two-envelope paradox.