Daniel Cebra had pointed out that the Vz dependence of the refMult edge (h) looks symmetric about Vz=0 after accounting for a linear dependence. I went ahead and fit the data to a line and applied a linear correction to compare the positive and negative Vz values. You can see in the figure below that after the correction, the Vz dependence is pretty symmetric. This tells me that the distribution is shaped mostly by the physical detector configuration, not by fluctuations in the refMult-edge fit quality. [discussion continued after figure]