Implicit conversion from float -21.31034482758622 to int loses precision