Hi Sichen,
I think that AERC 1.1 is the technical document where it is published, I don’t think my original report ever got published in a quotable way.
The roller shade XML files in CGDB are generated from measured spectral data that is converted to XML files using modified Kotey to get the angle dependence and then a very crude model for splitting direct and diffuse components are used (documented in the WINDOW technical manual sect 12.4, https://windows.lbl.gov/sites/default/files/software/WINDOW/WINDOW%20Technical%20Documentation.pdf). Greg derived a better way to spread the diffuse component out so that reciprocity is conserved but that has not been implemented yet.