Since the deviation is negative, it is not noise. Here is the linearity using very narrowband FFT method:
Each peak needs to hit the corresponding horizontal line. We see deviations starting around -100 dB and it gets worse, always showing lesser amount. Not that I measure it past -120 but you can see it is all pretty wrong in that region.
Noise will not always show positive in a single scan. But the single tone stuff nails it. Many thanks.