Initially, the team identified over two million addresses as potential sybils but later refined their criteria to minimize false identifications, resulting in a more precise classification.