I would expect to see reflections in the unterminated stub, since, as seen from the stub there is a piece of 50 Ohm coax terminated in 25 Ohms. Reflections from the unterminated end would not be absorbed by the source (as seen from the stub). So you have both an impedance mismatch at the source and at the load (unterminated stub). Attached is a quick and dirty LTspice simulation (blue is source, red is terminated load, green is unterminated load). Obviously a shorter stub would shift the standing waves to higher frequencies, but not even in the ballpark for ppm-level performance.
A 50 ohms system is not suitable for ppm accuracy. Termination accuracies, splitter mismatch, contact resistances... makes this impossible. Pretty basic, btw.
Either you have a very short cable system in which you do not have (too high) line losses or you use sensing. Everything also obviously strongly depends on max frequencies you need to achieve. If its the 100kHz you mentioned then you can do sensing to reach best results.
You should have a spec from your boss against which you work. Ppms, frequency, lenght... That would be a starting point for the group. Otherwise this will continue to be the fruitless discussion it has been so far.
If you "sweep" the high-impedance stub say from 1 megaohm to 10 gigaohm, what does this do to your frequency response plot? Also, does the unterminated stub cause an impedance mismatch between the terminated stub and the source?
If you "sweep" the high-impedance stub say from 1 megaohm to 10 gigaohm, what does this do to your frequency response plot?
No change at all. The reflection coefficient is pretty close to -1 anyway. I only put that resistor in to give Spice some finite resistance there.
Also, does the unterminated stub cause an impedance mismatch between the terminated stub and the source?
Yes, see the red lines in the plot. The closest things to this I can think off is those signal pickoff adapters that either used a power divider or use the shortest possible stub (a few mm long). And again, not in the same ballpark as a 10 ppm accurate source.