An LQ optimal tracker will achieve zero tracking error when the input penalty is removed. However, the problem becomes singular, and a solution hides behind several coordination transformations. We will highlight a discrete time version of this singular LQ optimization. We will also present data-based formulas to design this optimal tracker at the absence of a parametric plant model. We will show that numerical formulas which use input and output data from open-loop system tests are constructed in the way as to eliminate the need for explicit knowledge of plant model parameters.