Im doing some phasing and genotype imputation of a target data set using a reference panel and Beagle.

The way I understand it the phase of the target sites that overlap with the panel are going to be inferred by using the phase of the reference panel, beagle then imputes sites to the target based on this inferred phase.

What I don't get is how does the phase of sites that are in the target but not the reference panel get determined? Beagle obviously outputs phase for sites not in the reference, but how is it doing this? Or is it just that the incorporating the phased reference panel helps the model get a better starting point for phasing everything, including sites that don't overlap with reference?