I compared the results from the old version of Annual Daylight (1.3.0) and latest version (1.3.2) with the workflow attached GH files. I made the same assumptions for both b/c they have exactly same algorithm and parameters.
heatmap visualization from 1.3.2 ver. :
graphs shows results from UDI, UDI_LOW, UDI_HIGH, and DA in order.
We have made a few improvements to the accuracy of direct sun modeling between the last stable release of LBT 1.3 and the latest development version of the LBT plugin. So that probably accounts for most of the differences, especially the change with UDI_HIGH. Also note that Radiance uses stochastic methods to generate rays so no 2 Radiance simulations will be ever be perfectly equivalent to one another. Using better radiance parameters will give you more consistent results.