Ironically enough, the OPPOSITE generally happens more often (cells get flagged as unusable despite there being room because there's a slight overlap). I remember the manual bringing this up and mentioning that the right grid size is paramount for good results, but I'm not sure if the exact same text was ported to GMS2.Second, mp_grid_path gives you a path even if your object is not able to pass through because of large collision mask. So it only cares if a path exists with continuous free cells. I also verified it. It was silly to expect that since the function doesn't take any object as a parameter.