Hmmm, I see what you are saying, but I am not sure I agree that my inline load and cable are not flat enough at 25MHz bandwidth. If what you say is true, then adding another length of cable should show a difference again if the cable is at fault. I will try that tomorrow and see if I get a result that would indicate that my cable and/or load are at fault. I will also do it with the same input settings on the scope.
Please excuse my naivety with this frequency realm as I never really dealt with things like this above 10MHz and sometimes I played with things around 25MHz but not with any precision.
You can not really check your load impedance with changing cable lenght. You may see some change but can not classify what is really going on. Your test method is somehow ok, this I do not push to garbage. But your test accuracy is not enough for claim if SDG1000 is in specs or not. You can only tell that "my setting this and that show this and that image. But starting claim part of decibels... (I can see there is even 1/100 of dB resolution in your numbers but your system really can not be even close with this needed accuracy)
This next example is not true in this test setup but just for give some weak light for thinking dB's and 0.1dB's and 0.01dB's
Think
pure resistance 50ohm (not RLC hybrid)
Then connect 20pF pure capasitance parallel with it.
Next can not exactly do due to many reasons but if we forget phaase etc things and simply think that C is "resistance" with some other than just DC.
Lets look reactance map. With 25MHz this
example 20pF reactance is 318 ohms.
(btw with 100kHz it is 79.6 kohm and with 100MHz 79.6ohm)
Lets do deep simplification. 50ohm parallel with 318 ohm is 43.2ohm.
Is your load impedance now 50ohm. Not. What it means with dB's. (you can calculate)
(in real system there may be things what make situation more or less bad due to other R and LC reactances in total setup. Example: Oscilloscope input is RLC hybrid... there is RLC Network what may or may not affect more os less with different frequencies together with cable)
In real life things are much more complex. Due to other reactances of course.
This kind of "simply thinking calculkation" give only imagine what may happend and if it affect lot of or perhaps less. But it is well enough for tell that 0.01dB's are sure total garbage.
Oscilloscope like Rigol (and most of other simplemachines) do have only 50ohm +/- something resistor connected parallel with input by relay. It is not true 50ohm impedance over freq span. HF power meters heads are totally different "animals" and also example good spectrum analyzer inputs. And even more bad it go if we use coaxial T and external 50ohm terminator. Feed thru terminator is some amount better but still there is scope input reactance with it.
My point is only for accuracy. If we talk 0.1dB or even 0.01dB things in flatness your measurement is not valid. Using 50ohm termination and 50ohm cable and oscilloscope is ok for many kind of measurements. But it need understand what kind of error sources it may have. And just forget 0.1dB and 0.01dB thins.