It’s a good point. The BLG and CSP results appear to be identical if used downstream of the gamut mapper, as long as that is set so all (noticeable) negative values are compressed to >0. But when dialling back the effect of the gamut mapper to search for optimal settings, as soon as it allows some negative values through, the BLG and CSP results will begin to diverge.
I will experiment with a modified OCIO config implementation of the BLGs, using a transform to the native input colour space for each DRT, followed by a simple 3D LUT.
EDIT: Updated config linked here.