Sorry, I didn't see the source. But that is not the most accepted definition of covariate. There is no requirement that the "values of the covariate be known" á priori. The covariate must be identified and measured.
"All models are wrong, some are useful" G.E.P. Box
