I've published performance charts showing the characteristic performance of the units.

These charts were made by connecting one GPSDO with the PLL firmware running into one channel of a Keysight 53220A TIA. An FE-5680A Rubidium standard was connected to the other channel. The 53220A was set to generate time intervals between the rising edges of each, and TimeLab was set up to acquire a day's worth of samples. For this run, the oscillator was connected to an active antenna mounted above the roof being shared through a 5 way active GPS splitter. This configuration yields typical PDOPs of 1.2-1.7 on average, with mostly 7-11 satellites being used.

There are a couple of occasional, momentary discontinuities in the frequency difference. The origin of them is unknown, but the logging indicates that they were contemporaneous with reductions in the 1000 sample accuracy, meaning that you'd know from the LEDs that the accuracy was impacted.

The frequency difference graph shows an offset. That's likely caused by the fact that the FE-5680A is uncalibrated. It's frequency is likely quite stable (more stable than the device under test by at least 2 orders of magnitude), but its absolute accuracy is unknown.