Yes - you can see how they did this: The signal is very noisy (which is to be expected from those small directional couplers) so they just average multiple measurements. Not ideal, but at least it makes measurements possible. Specs are undoubtedly reduced below 10MHz, but so far, I think good enough for typical HAM use.
hendorog and I looked at this when we got the SVA1015X beta FW and were able to get sub 0.1dB accuracy from his and Siglent VNA Cal kits.
Undoubtedly the coupler is not great down below 3MHz or so. It is well out of its spec'd range. However that is compensated (i.e. hidden from view) by the inbuilt SOL calibration in the device, so you see that as noise on the trace. Other VNA's don't have a built in calibration and on them you can easily see the basic performance of the coupler just by attaching a 50 ohm load.
And of course you can do your own cal to improve it further. And when you do a cal, the magic of the error correction maths makes it a perfect coupler again!
However, when the coupler is operating in this low frequency region where its directivity is poor, what is observed is the noise and also measurement drift increase. The error correction maths amplifies the directivity which is great, but it also amplifies all of bad stuff, like noise and drift, which is bad.
So we see more noise, and the device becomes more sensitive to things like temperature changes. So turning on averaging, having a good warm up and a stable environment help. Doing the calibration and the measurements quickly also matters more in that region.
For HAM use, having the capability to do the measurement is generally more important than the convenience.
With that in mind, there is the option of using an external coupler between port 1 and 2, save the sweeps, and do the calibration on a PC. I am keen for Siglent to support OSL calibrations on Port 2 at some stage to make this a bit easier.
Another point which was raised by member graybeard is that calibration standard sweeps cannot currently be averaged on the device. The SVA just does one sweep and thats it. However if you do the error correction on a PC then you can use averaged sweeps of the calibration standards and then apply the error correction to averaged sweeps of the devices. Obviously that will take longer though, and so there will be a trade-off in drift to get the benefit of reduced noise