Greetings Ian,
It looks like the data only go out 18 hours. For example, the time variable associated with Geopotential_height_surface goes out to 1080 Minute since 2017-11-27T18:00:00Z :
This looks to be somewhat variable between runs, and likely reflects the experimental nature of the HRRR grids coming from GSD. From my understanding, as new features or GRID outputs are added, it is possible that the run takes more than the allotted time (something less than an hour), and if the processes are not finished, everything is canceled so the next run can begin. I believe this (working out the timing of the run) is part of the normal process and has to be worked out before new stuff goes to NCEP.
Cheers,
Sean