This study focuses on the algorithm RBSC-SubGen, which is originally offered for generation of vocabulary decks but can potentially be applied for a variety of problems. We first study the resilience of RBSC-SubGen against various hyper-parameters by testing it under various constraints. Our results indicate that RBSC-SubGen is sensitive to subset size S, followed by desired RBSC coefficient ρ * , permissible disparity ε and, finally, parent set size L. We then offer a simple modification for reducing the computational load of RBSC-SubGen and avoiding saturation. In particular, we offer to carry out an intermediate-level verification of the desired conditions. Additional tests show that the rate of saturation and the number of iterations decrease, whereas the disparity in the RBSC coefficient increases within acceptable limits.