Right, my point is that while they converge to exactly one second intervals as they should, the minimum adjustment step for timing a single second is fairly coarse, and it will bounce around between adjustment bits so that averaged over a long period, it's stable, but in the short term, it is not very finely smoothed. If the adjustment step was smaller, the variation from the intended center point could be much smaller. I'm measuring around +-25ns from an exact one second signal total variation, but the minimum adjustment step is around 20ns, so that individual short term measurements based off just the PPS signal could vary considerably, this has been the case across the ublox modules I've measured so far.
If that signal was disciplining an OCXO with a long integration time, then you used the OCXO's output to create another PPS signal, that one could be considerably more stable, probably an order of magnitude or two more than the 2x10^-8 worth of minimum adjustment variation I can see on the module's PPS output. I think this is probably around as good as you can get from a commercial GPS signal without additional smoothing, but the hardware needed to smooth it further is probably already in the box.