In the case of the two 'dashboard' graphs that was with me turning the potentiometer up to compensate for lower input level and achieve the same output level (to show that the distortion changes even though the output stage is not providing more/less power).
In the case of the graph below, that was obtained with the potentiometer staying at the same position
Similar graphs (showing IMD not THD but doesn't matter too much to demonstrate the effect) for other amps (for which I also after verified that the behaviour occurred even when actual output level was the same but input voltage only changed, thereby eliminating the output stage as the cause of the change in distortion):
Woo WA23:
Riviera AIC10:
If you are keeping the potentiometer at the same position and sweeping input voltage, you will obviously have a variable output voltage. And at low input levels (and output levels) the SINAD is going to be dominated by the noise of the amplifier. As the input signal climbs and the S+N ratio improves, SINAD will rise until it reaches a maximum. Then as the input signal continues to climb, distortion from the tube stage will continue to rise and SINAD will fall again. Your hump-shaped measurements are expected in a tube amp.
Now the dashboard results with the potentiometer adjustment, I cannot explain, there is something I'm missing here. I can't find any information on the topology of the Kallyste Amethyste, it isn't easy to discern in the photos unfortunately. But if we are talking a common cathode gain stage (the most common tube gain stage), and you turn the volume pot to maximum and put 70mVrms into the tubes grid vs. using a 1Vrms input and adjusting the pot to get the same output level, in both situations the tube is seeing 70mVrms into its grid. With a 70mVrms sine wave with negligible amounts of distortion in both cases, the output distortion will be roughly the same.
The tube doesn't care what voltage goes into the pot. In a triode gain stage, distortion will rise linearly with input level, but in the above example, the input level at the grid is identical (or at least, it should be).
The only other significant difference between the 70mVrms situation and the 1Vrms situation is the source impedance will be different in the two scenarios, as the potentiometer itself contributes depending on its position, but in any normal tube gain stage, with a competent source, the higher contribution of source impedance from the pot should not impact distortion figures, and certainly not by a factor of 10.