Stacks are the 'optimal' combination of multiple warps on the same skycell. For the 3pi this may only be 10 or so warps, but for the Medium Deeps it can be several hundred. Stacks are also astrometrically and photometrically calibrated. These images can be accessed and downloaded through the PS1 archive.
At the time of stacking, the pixel values in stacks are rescaled to a zero-point of 25 + 2.5* log10(exposure time), based on the input warp calibration. However, as the final ubercalibration of the data has not taken place at this time, this zero-point may be slightly incorrect (usually at the hundredths of a magnitude level). However, the stack fluxes/magnitudes in PSPS have undergone the final calibration process.
The exposure time for a stack is given in the FITS header. It consists of the unweighted sum of the individual warp exposures which were passed to the stacking process. Note that during the stacking process some of these warps can be excluded. This is not reflected in the exposure time, so the actual on-sky time for each stack is not necessarily the same as the exposure time given in the header.
Due to the gaps between the OTAs , and general cosmetic masking, the number of input warps per pixel with valid data on a stack varies. This variation can be quite large on the 3π (it is not so much of an issue on the Medium Deeps). As a result the true PSF at a given location on a stack can vary discontinuously between pixels (if the input warps have different FWHM). This acts as an increased source of error in the stack PSF photometry, as the PSF model fitted to the stack is continuous. Better results may therefore be obtained by using the Forced Mean Warp data (as the true PSF is expected to be continuous over a warp).