All the above results are based on the observations of SatNOGS Network and on using ikhnos to analyze them. In short, ikhnos tries to project the expected curve of signals (red line) on a SatNOGS observation waterfall.
It does that based on the time of observation, station location and on the TLE that were used, thus you expect to see a straight red line for the used TLE. So, let’s see what we get for observation 1831423 (I’ll show only 4 of the 9 TLE sets as the rest are too off):
As you can see, OBJECT RB is totally off, OBJECT RC is off enough to say for sure that it is not QARMAN. Someone would say that QZ is also off enough, however as the calculations are lacking of some degree of accuracy and precision comparing to other tools, I give some slight possibility to QZ, but as you see I have chosen the RG OBJECT to be followed in Network as it seems to fit better than all the other TLE sets. Feel free to use ikhnos, it is actually a python script, and check other good observations of QARMAN or some from Phoenix as QARMAN is also visible there.
On why you get best results for RC object is probably because the objects are still close. You can see on the ikhnos results above that the expected curve for the signals using RC TLE set is very close to the observed one. To understand better how close they are check the screenshot from Gpredict bellow which shows all the deployed OJBECTS (+ ISS) over a station (light blue dot) in Australia. Notice how all OBJECTS are visible from the station at the same time at the polar plot:
You can see that RC, QZ and RG are in a high elevation and not so much separated to be received well enough from the same station.