How sampling is done for continuous variables

[icarosadero]

I have read the sampling section on the methodology of the paper many times, but I am still unsure about how positive and negative samples are drawn. If I pass a continuous label $c$ along with some neural data $x$, and take a sample $x_t$ and $c_t$ how does CEBRA determine whether it belongs to a positive or negative class? Does 'positive' mean "has almost the same value as a reference $c_\tau$" and 'negative' otherwise?