The pointing refinement step looks for sources in common with 2MASS and uses these common sources to refine the coordinate information in the headers of the BCDs. Note that 2MASS wavelengths are much different than MIPS wavelengths!
It is not recommended that you use the Spitzer offline pointing refinement tool for MIPS, since it requires much tedious parameter tweaking and a good understanding of the quality of astrometric source matches. It is important to note that the accuracy of the refined pointing keywords appended to each scan mode BCD FITS header by the SSC strongly depends on the number and quality of the 2MASS-to-24 micron source matches available. Please use these values with care if you by any chance replace the standard BCD WCS with these refined keywords prior to mosaicking. If the mosaic turns out smeared (or sources appear double) then in all likelihood, the refined pointing is suspect and you will need to revert to the raw pointing.
In the end, if you do not plan to use these refined pointings, the mosaicker uses the raw image pointing reconstructed from the telescope boresight telemetry and a calibrated model for the scan-mirror. The blind pointing uncertainties of MIPS are currently ~1.2´´ for 24, ~1.7´´ for 70, and ~3.9´´ for 160 microns, all 1 sigma radial.
The pointing refinement pipeline will only run on 24 micron scan mode data using exclusively 24 micron point sources that are common with those from the 2MASS public catalog. It operates on the individual scan-legs of scan-mode observation in order to maximize contiguous sky coverage and hence the number of potential 2MASS matches. Earlier testing revealed that 2MASS-to-24 micron point-source matches are very rare in a BCD frame. The online pipeline performs optimum registration of scan-leg mosaics using 2MASS astrometry if a sufficient number of matches exist, then, the refined (intermediate) scan-leg pointings are mapped into the WCS of each BCD frame. Even for large scans, the online pointing refinement pipeline may not yield reliable results (or nothing at all) since 2MASS-to-24 micron matches are far and few between at all galactic latitudes.