Implicit conversion from float -2.6785714285714306 to int loses precision