I review several alternative linking hypotheses for relating eye tracking data from the VWP to cognitive theories and models. While some models are able to simulate VWP data surprisingly well (such as the TRACE model), there is still ample ambiguity to resolve in the meaning of fixation proportions over time, despite decades of work with the VWP. I also present a simple fixation model based on probabilistic sampling from an underlying lexical activation that allows simulation of individual trials. Unsurprisingly, a properly-parameterized sampling procedure approximates the underlying activation patterns when sufficient trials are averaged together. However, the utility of simulating trial-level behavior is not in reconstructing central tendencies (which can be derived directly without simulating fixations), but in addressing, for example, individual differences. I also discuss critiques and misunderstandings of linking models to the VWP, and analogies to a simpler paradigm - lexical decision - to illuminate the logic of linking hypotheses in the VWP.