The real threat to trust in science isn’t outright fraud, but the pervasive tweaking of research designs and models