I've been looking at frequency difference, phase difference and Allan deviation for the most part, but I took a look at the latest prototype's time deviation and had a bit of an "aha" moment.

Imagine the time deviation as being as if you had fed the frequency standard into a clock. After watching the clock for tau seconds, how far off might you expect the clock to be?

What's telling here is that after around 3000 seconds, the clock would have a basic uncertainty of 10 ns. It just so happens that the GPS module's specification for nominal PPS accuracy is... wait for it... 10 ns. I haven't done it yet, but if I were to graph this unit for a week, I would expect that line to proceed horizontally out to the horizon.

The phase discriminator unit compared to the old unit has the same performance so far out to around tau 100s. It then does just a little bit worse up to around 1000s, but then becomes better much faster.

The dark green graph is an old unit, the pale blue the new one.