To support inference algorithms, Bean Machine represents the model as a probabilistic graphical model. A probabilistic graphical model is a directed acyclic graph where each node is a random variable and edges between nodes represent dependencies between random variables. During a single iteration of inference, MCMC assigns a specific, concrete value to each of the unobserved random variable functions in your model. We refer to this set of assignments as a World in Bean Machine.
In an MCMC method, worlds are computed sequentially. A new world is "proposed" based on the random variable assignments from the current world. In each inference step, an MCMC method iterates over all unobserved random variables and proposes a new value. The world is updated to reflect this change; that is, likelihoods are updated and new variables may be added or removed. This updated world will either replace the existing world or be discarded as determined by the specific inference method. The value associated with each variable at the $i$th inference step is returned as the $i$th sample for the variable.