Initially, the team identified over 2 million addresses as potential Sybils but later refined their criteria to minimize false identifications, resulting in a more precise classification.

Go to Source