Bridging the gap between single- and multi-model predictive runtime verification