CLTs for mechanistic interpretability research. Compatible with circuit-tracer for generating attribution graphs and analyzing model circuits.