I noticed, that it seems your Guard is set "To LO" instead of "OPEN" on 3458A, while Guard is not connected to LO at the reference. Because of that we had quite some shifts during MM2020mini that were reproducable afterwards.
Has Frank measured the reference the same way you did, I mean the same wiring scheme?
I followed 2ppm hunt in MM thread
The measurement for LTZ#2 was taken as shown on picture, Guard closed and connected to case of LTZ#2.
I have done comparisons between all possible configurations for guard open/closed and guard connected to case or to low on LTZ#2.
Afair there was no significant difference in the readings, unfortunatly I lost the notes, so nothing left as foundation
Edit: I'm also curious about your uncertainty values, they seem a bit low. For comparison, Keysight Böblingen gives +/-2.5ppm on our 3458A calibration. Why you add only 0.5ppm/2 to your uncertainty? Maybe you can elaborate on that numbers?
For the reference values I used the mean values provided by Dr. Frank before and after return and added half of that drift to the uncertainty:
There is no added uncertainty for my measurements, I just used uncertainty Dr. Frank gave for his references.
After return, Dr. Frank told me that both references drifted by ~+0.5ppm.
As I said, I have no experience with this kind of calculations, if this is wrong, please push me on the right path.