CRPS when sigmas not 1

[LucasMMK]

When calculating the continuous ranked probability score, the true label "y" is also scaled by the difficulty:[/+]
[+]lower_errors[cpd_index]*(y-cpd[cpd_index])sigma [/+]
[+]+ higher_errors[cpd_index](cpd[cpd_index+1]-y)*sigma[/+]
[+]Shouldn't the ground truth stay as it is and only the calibrants are multiplied by the sigma?