For the GGX distribution used for specular sampling, resources generally state that you should square the roughness to get the correct alpha value. This, however, assumes (to my knowledge) that the roughness value is perceptual (i.e non-linear). Since the LabPBR standard stores perceptual smoothness, and the conversion to roughness already converts it to linear sqr(1.0 - specular_map.r), it is incorrect to then square it again.
For the GGX distribution used for specular sampling, resources generally state that you should square the roughness to get the correct alpha value. This, however, assumes (to my knowledge) that the roughness value is perceptual (i.e non-linear). Since the LabPBR standard stores perceptual smoothness, and the conversion to roughness already converts it to linear
sqr(1.0 - specular_map.r), it is incorrect to then square it again.