They really don't (at least for performance). The article gets several things wrong. For example, the purpose of calibration is not to stack rank, but to ensure consistency (IE 3.5 means the same thing across different orgs), etc. Managers don't sit in a room and try to force fit a curve.