Implicit conversion from float -10.25 to int loses precision